Welcome to the latest edition of Capital Efficient. This one is a little different — a single long essay instead of the usual deal roundup and reading list. We’ll be back to the regular format next issue.

The Demand Side of Vertical AI

Something is shifting in vertical AI that the mega-rounds in AI-native platforms and AI services businesses miss. The most interesting thing about AI is not the competition it enables — it’s how AI expands which markets exist.

Over the last few years, I underestimated how fast frontier models would match trained professionals in their quality of output. I also underestimated how broad the categories of expertise would turn out to be. As an investor, I had one perspective, but after getting more hands-on building with these tools my sense of possibility has changed. This spring I spent a few days rebuilding a community health navigator for a New Haven clinic that had lost its federal funding for navigator roles. That small project was enough to change my read on where things are heading.

To borrow from Gibson: what I’d call AGI-ish is here; it’s just not evenly distributed. Most of the analysis I read, including my own, is still sorting which $50B legacy markets get re-divided by which AI-native entrant. That analysis isn’t wrong, but it’s only a part of the discussion.

The more pressing question is which markets come online at all, and what kinds of companies form to serve them. That’s the demand side.

What the Supply-Side Story Misses

Let’s start with the supply-side consensus, because everyone reading this is likely already fluent in it.

Vertical SaaS is under threat as a category. The moats that justified venture-scale outcomes — enterprise-readiness, workflow IP, integration surface area — are collapsing into the model layer. The thing the software used to do, the model can now do. The thing the model can’t yet do, a small AI-native team wraps around a foundation model and ships in a quarter.

As a result, many investors are pivoting towards backing AI-native services businesses that eat the workflow end-to-end and provide a managed service offering to an end-user instead of an efficiency platform. Crosby for legal work, Hanover Park for fund administration, Corgi for business insurance, and many more like them launching every week. The playbook is straightforward: don’t sell the tool to the professional — be the professional, at 10x the leverage and a fraction of the cost. I wrote about this in Five Approaches to Vertical AI. It’s accurate — but it’s also the part of the story everyone else is writing. Amidst this technology shift, workflow automation is most at risk. The best companies in that bucket are evolving: into platforms of record, or directly into AI-native operators competing with the customers they used to sell to. Those that stay tools get subsumed by the model layer.

Subscribe now

The Bigger Question: Demand-Side Expansion

It’s clear we now have near-infinite affordable intelligence on tap. So how does that impact customer demand?

Professional services have historically had fixed minimum price points set by human labor costs. A senior lawyer. A commodity trader. A Medicare enrollment counselor. Each had a compensation floor, and the floor excluded entire categories of buyers. SMBs couldn’t afford the lawyer. Regional distributors couldn’t afford the trader. Underfunded clinics couldn’t staff the counselor. Demand was always there but the unit economics were not. AI collapses this compensation floor and can meet latent demand that had previously been priced out.

So what does this look like? Three illustrative examples:

• Commodity hedging. An airline or an agricultural giant — think Delta or Cargill — has commodity traders on staff. A regional food distributor doing $200M in revenue does not. It can’t afford to hire away a Wall Street senior trader, it doesn’t have the FCM relationships, and it can’t hit the bank desk minimums. So they don’t hedge. The market, at their scale, didn’t exist. Pillar raised a $20M seed this month led by Andreessen Horowitz to solve exactly this. Their platform automates hedging for commodity-driven businesses — metals recyclers, import-export firms, food traders — at a price point the old desks couldn’t touch. The founder, a former macro trader, put it plainly: “Sophisticated institutions had access to tools, infrastructure, and talent, while the actual producers, importers, and manufacturers driving global trade had little to no access. Risk management was treated as a luxury, despite being essential…[o]ur goal is to make hedging as accessible and ubiquitous as payments or accounting software.” The market for hedging had priced these companies out of the discussion — but now this is a service that’s been made accessible to the long tail of hypothetical customers.

• Legal access for small businesses. Everyone understands “I can’t afford a lawyer.” A five-person company with a contract dispute, an IP question, or a vendor who won’t pay faces a $5,000-$15,000 minimum engagement at any real law firm. Most never engage. They eat the loss, sign the bad contract, and Google their way through it. An AI-native legal service that drafts the demand letter, reviews the MSA, or files the motion doesn’t need to replace the lawyer. It needs to exist at a price point where the small business picks up the phone, with this new service becoming the lawyer they never could afford. That small business wasn’t a client of the legal industry before. Now it is.

• Bankruptcy access for low-income Americans. A typical Chapter 7 filing can cost over $2,000 in attorney fees — money the people who most need bankruptcy protection don’t have. Legal aid is overwhelmed, and most eligible filers never even file. They live with the debt instead. Upsolve, a 501(c)(3) non-profit, built a free AI-native self-filing tool that has helped more than 16,000 families eliminate over $1B in debt. A consortium of foundations has been funding Upsolve Assist, an AI financial counselor that analyzes individual situations and provides personalized debt-management guidance at no cost. It isn’t a cheaper bankruptcy lawyer — it’s a new shape of legal service the old unit economics didn’t allow. The families that Upsolve serves couldn’t access these kinds of legal or financial advisory services before. Now they can.

Demand-side expansion comes in two flavors:

On one end: existing services reaching buyers who were always excluded. On the other end: categories of organization that were never economically feasible before, now becoming viable — a three-engineer shop shipping AI tools for a hundred under-resourced clinics, AI hedge funds for the middle class, social services navigation for cities that lost their federal funding. AI’s ability to bring metered on-demand intelligence to any problem is the commonality across these ideas.

The further out you go, the more interesting (and weirder) the companies can get — and the less the existing industry even recognizes them as competition, because they’re serving a previously unaddressable market segment.

The pattern itself isn’t new. Clayton Christensen’s non-consumption framework gets at this: when the cost floor drops, markets that were always there become reachable. What’s new is the speed, the fact that distribution becomes the moat rather than technology, and the kinds of companies it enables.

A View From One New Haven Clinic

Federal budget changes last year cut the community health navigators at a clinic in New Haven — the people who help under-resourced patients find health, housing, food, and immigration services. The funding stream that paid for the navigator role dried up, but the need did not.

I spent a few days and built the clinic a live bilingual navigator with 192 local resources, routing by need, and a simple tracker — enough to prove the bottleneck wasn’t engineering talent but whether anyone would build the thing at all. I’m not launching it in this piece, but it did clarify something for me. The hardest unsolved problem in AI isn’t capability. It’s whether the model ever reaches the clinic.

Writing the software was the easy part. The hard part was the question I kept running into at every step: who does this actually reach, and how? A front-desk volunteer? A bilingual community organizer? A link shared in a WhatsApp group? I haven’t solved this. That’s the point. The engineering was trivial. The reach problem is the actual work, and the essay you’re reading is partly me admitting I don’t know the answer yet.

The version of the story that works is one where this kind of tool becomes the default for under-resourced institutions — not because someone builds it for them centrally, but because the unit economics of building it have collapsed to the point where the tool gets built locally, by one person, in days. That’s the far end of the spectrum I described above. Not an existing service reaching a new buyer. A category of organization that had no business model before, with one now.

A non-profit development shop of three to five people, serving 100 under-resourced institutions at a time, is now a viable organization. The same logic applies to clinics, community colleges, small municipalities, legal aid offices — the parts of the public and social sector running on donated time and grant cycles. Whatever shape the for-profit operators take, someone also needs to build the version that serves populations the commercial version was never going to reach.

The tool only exists because the primitive underneath it is cheap, reliable, and accessible enough for one person to build on in a weekend. That’s not guaranteed. That’s a lab decision.

Timing, Distribution, and the CAC Problem

The timing is unusual. Most technology transitions go: incumbents die, then new things emerge in the wreckage. This one is different — new markets are forming before old ones finish dying. Demand is moving faster than displacement, because the bottleneck was never supply. It was the buyers who were excluded from the market in the first place. Create intelligence cheap enough, and the excluded buyers become customers immediately. Incumbents take longer to unwind than new markets take to form.

The moat is distribution, not technology. The real defensibility for the companies at the far end of this expansion isn’t the AI — it’s being the first to build go-to-market for populations no one has ever sold to. Pillar doesn’t just automate hedging; it invents the sales motion for metals recyclers who’ve never bought a financial product. A non-profit dev shop doesn’t just build tools for small clinics; it earns credibility with administrators who’ve never deployed AI. Technology is becoming a commodity. The relationship with the underserved buyer is the asset.

The risk. The sharpest pushback is predictable: SMB acquisition cost is the real ceiling. You can’t acquire a $30/month customer for $1,000 and build a venture-scale business.

Two responses. First, when you’re being the service rather than selling a tool for it, revenue per customer is structurally higher. A SaaS tool charges $50-200/month. An AI-native service doing the actual work — the legal filing, the hedge, the staffing, the benefits enrollment — charges on the value of the output. Revenue per customer is 5-20x higher. CAC math changes when the revenue line changes.

Second, and more honestly: CAC itself is a distribution problem that’s been unsolved for thirty years across CRM, cloud, and fintech — not because no one tried, but because the cost lives in trust-building, not lead generation. Fragmented buyers, no procurement infrastructure, long sales cycles relative to contract value. AI changes the revenue side of the equation. It doesn’t automatically change the trust side. Which is why the deployment-reliability question I’ll get to at the end of this essay is so important: it’s the mechanism by which acquisition costs actually come down over time (or don’t).

The supply-side story is about which $50B markets get re-divided. The demand-side story is about which human interactions (and businesses) become possible for the first time.

What Would Prove This Wrong

Three things that could prove me wrong:

• Frontier model plateau — if frontier models stall out (which is not a bet I would take given the last 24 months), AI doesn’t get reliable enough to bring these new organizations into being.

• Agentic reliability — if AI can’t move past the last mile of “plausible but wrong,” copilots survive but full agents don’t due to a lack of trust. Again, not a bet I would take.

• Regulatory capture — law, healthcare, and finance are the biggest excluded-buyer categories and also the easiest for incumbents to lock down with licensure rules. Watch state bars and insurance commissioners in 2026-2027.

The popular falsifier — enterprise-readiness — is a speed bump, not a moat. Models keep eating it.

What a Startup Even Is Now

Which is why the word “startup” is doing less and less work.

The venture-backed company of 2021 was a software vendor. It shipped a product, sold seats, recognized ARR, and defended a moat built out of integrations and enterprise controls. Five to ten years to $100M ARR. Exit path either public market or strategic acquirer. The venture-backed company of 2026 is something else. It doesn’t sell software to the professional — it is the professional.

Its P&L looks like a services business, while the company tries to make its margin structure look like software. Its moat is distribution into populations no one has ever sold to. Its competitive set is not other software vendors — it’s the labor pool of humans doing the work today, at a cost structure that excludes a whole swath of potential buyers the company could otherwise serve.

The key difference: it’s not a leaner software company — it’s a completely different kind of business.

And the shape keeps shifting. A single immigration attorney running 500 active cases. A boutique M&A advisor doing sub-$10M deals the big banks won’t touch. A two-person firm managing 10,000 rental units. A patent-filing shop that replaces a $2M/year associate team with one partner and a model. None of these are “startups” in the 2021 sense. They’re a new class of organization, and the commentary hasn’t caught up.

The playbook for founders: stop benchmarking yourself against SaaS. Your comp set is incumbents writ large — the law firm, the brokerage, the consultant — not Salesforce. Your hires are engineers plus domain operators, not engineers plus product managers. Your board wants people who ran a business in the vertical, not more software executives. Your fundraising story is about creating a market and capturing services margin, not ARR growth at 120% NRR. The faster you stop pretending to be a SaaS company, the faster the P&L starts making sense.

The question I’d ask any founder right now: are you building the 2021 startup or the 2026 one? The difference is not incremental — and the market will eventually tell you which you are.

The Constraint Is Reach

If the demand-side expansion is the real story of the next decade, value accrues at two layers at once.

At the application layer, operators win. They own the buyer relationship no one else has ever built. The tax-planning firm for doctors and dentists priced out by family-office retainers, the fractional CFO for $10M-revenue companies who couldn’t justify a full-time hire, the VA benefits advocate for veterans waiting years on their claims — each becomes indispensable infrastructure for a buyer the existing market wasn’t built to serve.

At the primitive layer, the lab that ships the model every operator builds on captures the largest share of a pie that just got much bigger. Not because the lab won the existing market — because the market it’s serving is the one the existing offerings couldn’t reach.

Those aren’t in conflict. They’re complements. The demand-side expansion creates a newly-formed pie. Operators capture the application layer of it. Labs capture the primitive layer of it. Both layers are structurally larger than they would be in a world where AI was only redistributing markets that already existed.

The obvious tension: labs have repeatedly moved up the stack. Consumer chat products, agents, vertical assistants. If labs capture application-layer value directly in the categories I’m describing, the operator thesis narrows. My bet is that the deployment contexts that matter most here — asylum hearings, title objections, Medicaid enrollment — are the ones labs are least equipped to own, because they require vertical trust-building labs aren’t structured for. But that’s a bet, not a foregone conclusion.

Where the primitive layer gets interesting is that the deployment context for demand-side expansion is unlike the context frontier labs optimize for by default. The asylum applicant can’t tell whether a misworded pleading will cost her the hearing. The solo inventor can’t evaluate whether the claim language the model drafted leaves his IP exposed. The first-time home buyer can’t know whether an AI-drafted title objection actually protects her deposit. Trust at the point of deployment is the entire product.

The difference between labs stops being what shows up on an MMLU leaderboard. It starts being whether the lab is investing in silent-failure detection in populations who won’t push back when a model gets it wrong. If the failure is invisible to the lab, it never gets fixed — and if it never gets fixed, the whole category of deployment stalls at the populations who most need it.

This is also the CAC answer from earlier. Distribution into excluded populations is expensive because trust is expensive. Silent-failure detection isn’t a lab-ethics problem — it’s the mechanism by which acquisition economics eventually work, because a service that quietly fails on the populations it claims to serve never builds the reputation that lowers CAC over time.

Not every lab is trying to win this game. Some are optimizing for reach and consumer attention. Some are optimizing for benchmark dominance. A smaller number are treating fragile-deployment reliability as the entire product.

If AI is going to expand which markets exist, that is the constraint that matters. Not whether the model is impressive in isolation, but whether it can be deployed where trust is thin, feedback is delayed, and the buyer has no margin for error. And when that happens, what kinds of businesses become possible?

Welcome to the latest edition of Capital Efficient. Let’s get into it.

Five Approaches to Vertical AI

Something is breaking in vertical software - and how to respond to it is top of mind for founders and investors alike. SaaS multiples have compressed to roughly 3x. AI-native startups face margin pressure from inference costs that didn't exist in the prior era. And AI can now work autonomously across multi-hour tasks - capabilities that open up entirely new ways to build businesses, not just new ways to build software.

These shifts are unlocking business models for vertical AI that look very different from the last generation of venture-backed startups. A year ago, vertical AI meant one thing: sell high-ACV software to an industry that hadn’t been digitized yet. That playbook is still alive, but it’s no longer the whole story. The best founders I’m talking to aren’t just asking “what’s the best vertical AI platform to build?” - they’re asking a harder question: “given everything AI can do right now, what’s the highest-leverage way to build a business in my industry?” This is my attempt to map the five distinct approaches I see working in 2026 - and where I think each one goes from here.

1. AI-Native Services: Compete, Don’t Enable

The traditional vertical software playbook was straightforward: enable an industry to digitize by selling it category-specific SaaS. But as AI gains the ability to take real-world actions, startups can now compete directly within an industry, not just sell software to it. An AI-native law firm goes after legal spend, not IT spend within law firms. An AI-native accounting firm is the accountant, not the software vendor for bookkeepers. The addressable market for competing is dramatically larger than for enabling, and AI allows new entrants to operate with margins approaching software-level economics while capturing revenue streams previously off-limits to technology companies.

My logic: vertical AI platforms are betting enterprises will shift labor spend to software spend, justifying outsized ACVs. But that premise cuts both ways. If AI is genuinely powerful enough to automate at that scale (and command those ACVs), you should be able to use the same tooling to compete directly as a services business; but at margins that legacy incumbents can’t touch. And if AI stalls out and that automation promise doesn’t fully materialize, the high-ACV platforms flame out anyway, because their valuations were built on ACV assumptions that never arrived.

In a lot of verticals, competing directly might be the better bet. Same tooling, better positioning, without the dependency on your customers agreeing to spend record-setting amounts on software and a much larger end market. Some examples we are already seeing in this bucket:

• Legal: Crosby is an AI-native law firm for startups backed by Index & Sequoia that has raised $25MM in the last year. On the other end of the spectrum, Norm Law is an AI-native Big Law competitor that has raised $140MM from Blackstone, Bain & Coatue.

• Accounting: Quanta raised a $15MM Series A led by Accel and is going direct to startups as the accountant, rather than selling software to CPAs.

• Insurance: Coverage is a Sequoia-backed AI-native insurance broker capturing margins with AI-driven efficiency, having raised ~$50MM. They offer businesses insurance policies directly vs. building SaaS for incumbents.

• Investment Banking: OffDeal is an AI-native investment bank doing the deals, not selling deal-sourcing software, having raised $17MM led by Radical Ventures.

• PR/Comms: Blazel is an AI-native content and PR platform drafting pitches, generating media lists, and monitoring coverage at startup-friendly pricing, having raised $7MM led by Sierra Ventures.

The playbook: take an expensive or inaccessible professional service, rebuild it with AI at the core, capture the full value of the efficiency gain rather than selling it to incumbents for a fraction.

Another way to think of this model is Tech-Enabled Services 2.0. We saw lots of startups raise in 2010-2020 for tech-enabled services bets but they largely failed due to brutal unit economics. AI changes that cost structure and makes this model worth revisiting. The key difference: back then, a legal tech startup still needed armies of paralegals and associates to deliver the work; today, a 20-person team can do what once required hundreds. AI-native services also has a more forgiving market structure - vertical software tends to cluster around one or two category winners (we will get to these), and the 10th-best platform is dead. But the 10th-biggest insurance wholesaler or law firm is still doing hundreds of millions in revenue.

2. Vertical AI Platforms of Record

Another way founders can win in Vertical AI is to build the Platform of Record for their industry. This is a more traditional venture bet - use AI to either replace a legacy vSaaS player or build an AI-native platform of record (POR) for an industry that hadn’t seen a venture-scale outcome. So what does this look like? Picture a High-ACV software platform designed for orchestrating agents and copilots. Every industry knows it needs to implement AI but most don’t know where to begin. By building a vertical AI platform of record, you can sell an easy-to-implement, compliant command center for driving AI automation within a category.

Examples of startups building vertical AI platforms of record include Harvey (legal, $5B valuation, $190MM+ ARR), Hebbia (finance/consulting), Abridge (healthcare, $550MM+ raised). Within Bowery’s portfolio, Harmonyze is building the platform of record for franchisors.

Voice as a wedge. One of the most effective entry points into building a POR is voice - especially in industries where workers aren't at desks. Field sellers, construction crews, home health aides, and technicians were largely bypassed by the last wave of cloud software. Voice meets them where they are.

• Hospital Systems: Abridge used voice-first clinical documentation as its wedge into large health systems, then expanded into RCM and broader platform capabilities. Voice capture was the beachhead; the data flywheel is what made it a POR. Over $550MM raised, $2.75B valuation.

• Home Healthcare: Roger Healthcare records in-home patient visits via voice and auto-fills the OASIS assessment, government-mandated Medicare documentation that takes clinicians 2-3 hours per patient. Home health aides are in patients’ living rooms, not at computers. Voice is the only interface that makes sense.

• Field Sales: Bowery portco Enata is building the second brain for field sellers in categories like F&B, CPG, pharma, and med device. Reps brain-dump via voice throughout the day; Enata handles CRM inputs, follow-up emails, and ERP coordination, giving sellers back 2-3 hours daily.

In these categories, you’re often competing with no technology at all, so you don’t face the incumbent displacement problem that plagues most enterprise software. People share richer context by talking than by typing into a CRM. Whoever captures that voice data first in a given vertical creates a compounding advantage, and from there can earn the right to become the POR.

The risk: vertical software markets cluster around one or two winners, and many obvious verticals already have semi-entrenched leaders with massive war chests. For later entrants, the question is whether you can find a wedge, like voice, that provides a differentiated path before the window closes.

3. Workflow Automation

Vertical AI investors have poured money into two categories - healthcare and insurance - around the workflow automation opportunity. These are both industries that have lots of well-defined processes, a heavy paperwork burden, and clear rule-based guidelines for decision making. Digitizing high-volume, repetitive processes in industries that still default to fax, PDFs, phone calls, and spreadsheets is something AI can do right now with current capabilities, so it’s no surprise we’ve seen investor and founder interest in solving these challenges.

• Healthcare: The most common targets for AI-native automation are prior auth and claims submissions. Prior auth is a massive, multi-step process involving payers, providers, and mountains of documentation still largely done by hand. The inefficiency has real consequences - delayed care, provider burnout, and billions in administrative waste annually. Companies like Adonis and Cohere have raised hundreds of millions solving prior auth bottlenecks.

• Insurance: The $7 trillion insurance industry still runs on PDFs, Excel, and disconnected systems. The cost is real - slow underwriting cycles, missed submissions, and underwriters spending more time on data entry than risk judgment. The most common workflows being attacked by AI-native players are submissions intake and underwriting. Emerging players in this space include Sixfold and Further AI.

The best companies in this bucket won’t stay here. They’ll use workflow automation as a wedge to evolve into a Vertical AI Platform of Record or into an AI-native competitor using their workflow expertise and data advantage to compete directly with incumbents they initially served. The window for pure workflow automation as a standalone business may be closing as foundation models improve and horizontal tools handle simpler automation increasingly well. The companies that use their foothold to move up the value chain will build something durable.

4. Biz in a Box + AI

The last generation of vertical SaaS built for small businesses largely stalled out. Even the seeming winners of the 2015–2022 wave of biz-in-a-box software hit a ceiling at $10–20MM in revenue and never broke through - not because the products were bad, but because SMBs simply won’t pay meaningful ACVs for standalone software. The unit economics never worked. The vertical AI era makes this model worth revisiting. The new playbook flips the model: give small business operators a digital hub to run their operations, then monetize by owning a painful back-office task or a piece of their labor spend, taking a cut of revenue instead of charging a traditional software subscription.

One thing that’s exciting about this approach is these platforms don’t just serve existing SMBs; they enable new ones to form. Many aspiring operators never launch because barriers are too high: credentialing with payers, regulatory compliance, sourcing supplies at competitive prices. Platforms that solve these on Day One make first-time operators competitive with established players from the jump. Therapy is a canonical example. Alma and Headway recognized that independent therapists face brutal administrative overhead (credentialing, claims, billing, compliance), and that the back-office burden, not clinical skill, is the primary barrier to independent practice. Some examples of modern biz-in-a-box startups that are making it work:

• ABA Therapy: Alpaca Health automates insurance onboarding, claims processing, and reimbursement optimization for ABA therapy providers serving children with autism, taking ~5% of topline. It removes a critical barrier to entry for new providers and wins when the provider wins.

• Restaurants: Owner.com gives independent restaurants the digital stack that chains like Domino’s spent billions building: AI-powered websites, online ordering, SEO, and marketing automation - plus “AI Executives” that handle marketing, finance, and ops tasks. A first-time pizzeria owner gets access to the same conversion and reordering infrastructure as a national chain.

• MedSpa: Moxie provides biz-in-a-box for MedSpas with better pricing on injectables, supplies, and equipment via collective purchasing power. A first-time operator gets Day One access to pricing that would normally take years of scale to negotiate.

The formula: provide the business-in-a-box (cost of building this now nearing zero), layer on a high-value service like credentialing, claims, sourcing, or better input pricing, and take a percentage of revenue. The best platforms won’t just be tools. They’ll be the reason new businesses get off the ground in the first place. Closely related to AI-Native Services (#1): the common thread is that AI can now do work, so you monetize the output, not the tool.

5. Custom Automation Platforms

Every Fortune 500 company is strategizing around how to implement AI - to cut costs, drive growth, and maintain competitive edge. But for the very largest enterprises, most vertical AI startups are not broad enough or malleable enough to slot into their idiosyncratic workflows. These organizations have decades of calcified processes, complex technical architecture, and requirements no off-the-shelf product satisfies. Historically, when these businesses couldn't find software to automate something, their alternative was hiring consultants to build internal tools. The products those engagements produce are rarely best-in-class, and they don't compound. Now a third path is emerging: Custom Automation Platforms (CAPs).

• Distyl AI - Founded 2022, raised $175MM at a $1.8B valuation in a September 2025 Series B led by Khosla Ventures and Lightspeed. Its Distillery platform deploys AI agents directly into Fortune 500 workflows across healthcare, telecom, and manufacturing.

• Percepta - Incubated by General Catalyst, launched publicly in October 2025 with no disclosed investment amount. Deploys engineers directly inside enterprises via its Mosaic platform, with clients including Fortune 500 companies and state governments.

• Brain Co. - Founded 2024, raised a $30MM Series A led by Elad Gil and Jared Kushner’s Affinity Partners. Bridges Silicon Valley AI talent with large enterprises and governments via a strategic partnership with OpenAI.

What makes CAPs distinct is their heavy forward-deployed model: they start by doing unscalable things for individual enterprises, but each engagement builds reusable automation components for a proprietary technical platform. Distyl calls theirs Distillery. Percepta calls theirs Mosaic. All three are trying to use bespoke deployment to build a core asset that compounds, then scale it across lookalike customers. The key open question for the category is who builds and maintains these workflows in the long run: the CAP's forward-deployed engineers, or the customer's own teams? The answer determines whether this model scales like a software business or plateaus like a services one.

The playbook: custom, n-of-1 automations. Charge implementation plus ongoing maintenance. Each engagement builds reusable components for a core technical asset, following the Palantir playbook. Automate claims for a top-five payer, then go to the other four with a speed advantage. Right now this looks like consulting. The bet is it evolves into something with real technical leverage and software-like scalability. One dynamic worth watching: Anthropic and OpenAI are hiring their own forward-deployed engineering teams to deploy into massive regulated organizations and drive token consumption, putting CAP startups in direct co-opetition with the model providers themselves.

These five buckets aren’t rigid categories. Companies move between them. The best workflow automation companies evolve into platforms of record or direct competitors. The best biz-in-a-box platforms are really AI-native services with outcome-based pricing. A Custom Automation Platform that builds enough reusable components starts to look like a vertical AI POR for the F250. The framework is less a taxonomy than a map of entry points, and the most interesting companies are the ones that use one bucket as a wedge into another. The question I’d ask any vertical AI founder right now: which of these five are you building for your industry, and why?

Share Capital Efficient

Deals

• ElevenLabs raised a $500MM Series D at an $11B valuation in early February to cement its position as the default voice AI infrastructure layer. This came on the heels of news they had passed $300MM in ARR. Sequoia led, with a16z, ICONIQ, and Lightspeed all participating. If you’ve interacted with a synthetic voice in the last year, there’s a decent chance it was running on their models. The question at $11B is whether they stay an API business or push into end-user products - my bet is both, and the volume of voice interactions flowing through their platform gives them a data advantage that compounds over time.

• GenLogs raised a $60MM Series B to build an AI-powered freight intelligence platform. The round was led by Battery Ventures with IVP, Cathay Innovation, Venrock, and Autotech Ventures participating. GenLogs operates a nationwide network of roadside sensors and satellites to give insurance firms, freight brokers, and government agencies real-time visibility into U.S. trucking - a dataset that simply didn’t exist before them. They’re already serving ~100 customers. The real moat here isn’t the software, it’s the compounding data asset: every sensor deployed and every mile tracked makes the intelligence layer harder to replicate and more valuable.

What I’m Reading

It was a banner week for X articles on vertical software - a few interesting ones:

10 Years Building Vertical Software: My Perspective on the Selloff (@nicbstme) After nearly $1T wiped from software stocks, founder Nicolas Bustamante argues LLMs are eroding vertical software’s premium moats — learned interfaces, custom business logic, bundling — while proprietary data, regulatory lock-in, and network effects survive. A moat redrawing, not a moat destruction.

Rebuttal to “10 Years Building Vertical Software” (@atelicinvest) The Unemployed Capital Allocator pushes back: the “AI kills software” narrative misunderstands what software actually is — accumulated domain knowledge delivered through a trusted relationship — and distribution moats, organizational inertia, and enterprise change complexity will protect incumbents far longer than the bears expect.

In Defense of Vertical Software (@gsivulka) Hebbia’s George Sivulka argues the value of enterprise software was never the code or interface — it’s “process engineering,” the deep understanding of how a specific team does their specific job — and that last-mile organizational knowledge, calcified over years of collaboration, is precisely what LLMs cannot commoditize.

Welcome to the latest edition of Capital Efficient. Let’s get into it.

Weekly Radar

Compete vs. Enable

I’ve been thinking more about what vertical AI’s form factor is as the cost of code declines. I see two paths for founders building venture-backable businesses in vertical AI.

Path 1: Build the platform. Get to a category early, get some heat on you, and try to build one of the breakout vertical AI platforms. This generally means selling high ACV software that is “AI-native,” offering agents and copilots that make workers more efficient. A big reason for why these companies command such rich valuations is that the efficiency (i.e., labor replacement) they offer will tap into human capital budgets and drive outsize ACVs. Examples: Harvey, Hebbia, Legora.

Path 2: Be the competitor. If a vertical AI platform can truly drive outsize efficiency, why not just compete directly? A few reasons this is increasingly attractive:

• The end-market is much larger. Compare all legal spend to all legal IT spend. This also gets around ACV sticker shock. The current vertical AI bet is that it taps into human capital budgets and can charge massive ACVs over time, but there’s a psychic limit that will take time to overcome for what people want to pay for software in an old-line business. Going direct sidesteps this entirely.

• You hide your technical asset. If every law firm is using Harvey, they’re all equally productive. If you build your own secret sauce and compete directly, you could outcompete these AI-enabled firms by being AI-first. Your innovation is your moat, not a product you’re selling to everyone.

• The risk cuts both ways. Yes, it’s risky to compete directly. But you can build a bigger top-line business, tap into markets that already have a kingmade vertical AI platform, and here’s the thing: if the AI isn’t good enough to compete directly, it means the labor replacement isn’t real, the ACVs never materialize, and none of these vertical AI platforms get over their pref stack anyway. You can’t have it both ways.

“But what about margins?” If AI can truly automate the work, an AI-native services business should have margin profiles approaching software, not traditional services. And if it can’t, the entire vertical AI platform thesis falls apart anyway, because the labor replacement that justifies those massive ACVs was never real. Same labor, same automation, different business model.

And the market structure is more forgiving. Vertical software usually sees clustering around one or two category winners. If you’re the 10th-best vertical AI platform in a category, you’re dead. But if you build an AI-native insurance wholesaler and come in 10th, you’re still doing hundreds of millions in revenue.

Examples of what I’m talking about:

• Build a staffing agency vs. sell recruiters AI sourcing tools

• Start a law firm vs. sell lawyers AI copilots

• Run a creative agency vs. sell marketers AI content generators

• Offer accounting services vs. sell accountants AI bookkeeping software

• Start a market research firm vs. sell analysts AI research assistants

In each case, the AI-native competitor captures the full value of the efficiency gain rather than selling it to incumbents for a fraction.

With the cost of code falling, we’ve seen VCs gravitate away from SaaS and towards defense tech, robotics, and neo-labs. I think the next phase of the cycle is founders building competitive businesses, not software providers, across a whole range of categories. Think of it as a return to the early days of venture backing companies like FedEx: real businesses with real operations, not just a software layer. I’m personally excited for this next wave of AI businesses, and if you are building one, get in touch.

Share

Deals

• AI meets RIA: Zocks, an AI assistant for financial advisors, announced a $45MM Series B led by Lightspeed and QED. This comes on the heels of direct competitor Jump raising a $20MM Series A led by Battery last year. A lot of big vertical AI businesses like Zocks are going to start as AI notetakers and spiral outwards from there. We’ve seen this playbook work in healthcare with Abridge, which started as a clinical documentation tool and expanded into a broader AI platform for health systems. As the gen 1.0 vertical AI companies mature, expect to see a clustering around one or two industry defaults in the same way we saw in vertical SaaS (i.e., if you are a GC, you are on Procore or Autodesk Construction Cloud).

• Voice AI for Field Sales: Bowery’s first Voice AI investment is now officially out of stealth. Enata is building a second brain for field sellers. Enata lets field reps capture visit notes and customer context by voice, building a living record of relationship history across their territory so nothing slips through the cracks. Unlike coaching tools designed for compliance monitoring, Enata is built for high performers who want a second brain to cover more ground. With a founding team (Justin, Marty, Smit) that combines technical expertise and field sales know-how, we couldn’t be more excited to partner with them. We led their first round of funding with our friends at Breakers VC and SV Angel. Check out their launch video below.

What I’m Reading

• The Adolescence of Tech: The Oracle has come down from the mountain and it’s worth reading. Dario’s implication is that the fast takeoff just got faster and we should brace for impact (or liftoff, if we take the right steps now). If you are in any way interested in AI and its implications, this is a must read. This one’s long, so I’m not going to recommend anything else for this week, lest readers get distracted.

Weekly Radar

AI-Native Big Law

Last month I wrote about Crosby, a new AI-native law firm targeting startups. Now there’s a new entrant going after the other end of the market, Norm AI.

Norm AI started in 2022 as a regulatory agent company selling compliance software to major financial institutions. But last month they announced something exciting. Norm AI is launching Norm Law LLP, a full-fledged law firm that intends to compete directly with Big Law.

Source: https://www.normlaw.com.

Norm Law LLP would be the first AI Services startup I have seen that aims to compete directly with blue chip providers for their top clients, instead of simply serving the long tail (i.e., startups and SMBs). In conjunction with the launch of Norm Law LLP, Norm announced they had raised another $50MM from Blackstone - the business has raised a total of $140MM to date. Other investors include Bain Capital, Citi, TIAA, Coatue, Henry Kravis, and Marc Benioff.

Norm Law is branding its approach as “legal engineering.” The startup has hired a few dozen attorneys with Big Law pedigrees and has them working hand in glove with in-house AI engineers to convert legal workflows into agents that they can offload their lawyering onto. Lawyers design the logic while engineers build the agents, with lawyers then overseeing their deployment.

Norm has also brought on some heavy hitters to beef up their legal bona fides including Ben Lawsky (former NY DFS Superintendent, familiar to any crypto investor), Troy Paredes (former SEC Commissioner), and Dan Berkovitz (former SEC General Counsel). Now, they’re hiring partners, associates, and paralegals. So how does this shake out?

• Crosby wins the startup market. Startups are AI-native themselves, prioritize speed and price over pedigree, and lack existing vendor relationships. Crosby could also keep growing with these startups and eventually become equivalent to a tech-focused law firm a la Cooley or Gunderson.

• Norm Law (and its competitors) knock out Tier Two Big Law. The bottom half of the AmLaw 250 lacks the brand to survive on reputation and is too slow to modernize from within. AI-native firms like Norm Law can win here and establish themselves as true alternatives to expensive second and third-tier Big Law firms which charge rates commensurate with the AmLaw 20.

• Tier One Big Law survives. Firms like Skadden, Paul Weiss, et al. will end up adopting Harvey/Legora or building engineering talent in-house (most likely the former). They have the relationships, brand, and reputation that will probably help them ride out the transition and will adapt just like they have through past shifts.

Expect other AI startups in accounting, comms, and professional services to make a similar pivot. Competing directly as the service provider may be a better value prop than being another IT vendor, if you can actually drive up margins with AI. And if you can’t, your AI platform probably didn’t work that well anyway and won’t command a good ACV.

Subscribe now

AI Agents Cross the Chasm

Kenn So’s () annual AI Trends report dropped last week. Kenn is a very bright thinker on all things AI and the whole thing is worth reading, but two findings stood out to me.

• AI is matching human professionals in quality. On the GDPVal benchmark, which tests whether AI can produce professional-grade deliverables like 3D engineering models and financial analyses, GPT-5.2 matches or beats human experts (avg. 14 years experience) more often than not (~70%).

There is certainly some benchmaxxing in these numbers but assume this trend holds.

• The capability of these models is informing adoption. AI adoption is insanely fast vs. other disruptive tech (see chart below) and better performance begets more usage. A year ago I used AI a lot less - the same interfaces (Claude/ChatGPT) are materially more useful now. Not to mention new entrants like Google/Grok. Adoption of AI is far ahead of PCs, smartphones, or the Internet when they were at this point in their lifecycle and is only poised to accelerate.

Source: @ Generational

Claude Code Weekend

Claude Code had a big few weeks over the holidays. Every VC and vacationing startup employee was cranking away on side projects. I was one of them.

In two days I built and deployed a birding app (basically a crude, localized version of Merlin). I built agents to automate a bunch of personal workflows. I built a vertical AI ARR/employee site (watch out for its debut). And all of this with…well, let’s just call it, limited technical ability.

It’s cliché and hype-y, but it’s also real. And it’s extremely fun. The gap between what you can think up and what you can ship is shrinking. Everyone should be playing with this stuff. It’s entertaining, it’s educational, and it shows you how far things have come (and offers a preview of where we are heading).

My view: we’ve crossed the Rubicon in the last few months and I’m coming around to the fast takeoff (TM) theory.

And for vanity’s sake - a screenshot of my janky Merlin competitor is below. As you can see, Claude isn’t great at centering birds in preview images just yet.

Deals

Slow week for announcements given the holidays, but two mega-acquisitions dropped since our last edition.

Manus → Meta (~$3B, ~$125MM run rate)

Meta bought Manus for a rumored $3B at the end of December. I tried Manus when it launched and didn’t fully get it, but Wide Research and other agent-driven search features won over converts. One rumor is that Meta may use the tech to help businesses run ad campaigns and offer them a SMB toolkit. Another is that this is just a talent buy given their AI struggles to date (see LeCun v. Wang). I don’t think this ends up an Instagram-tier GOAT acquisition, but let’s check back in a year.

Groq → Nvidia ($20B)

Nvidia also had a buzzer beater license & hire acquisition, paying $20B for Groq’s assets and key talent in the waning days of December. The deal was structured as a “non-exclusive licensing agreement” - presumably to dodge antitrust review. Groq was valued at $6.9B three months ago and Nvidia paid 3x that. This helps Nvidia head off a potential competitor and further solidify its place in the AI hardware ecosystem.

What I’m Reading

Capital in the 22nd Century: Some really big-brain thinking on the logical endpoints of the more extreme “AI will eat the world” views. Worth a read with an open mind.

Dan Wang’s 2025 Letter: Every other VC has already recommended this on X, but I’d be remiss not to include it here. Great annual letter from a close observer of tech and China.

The Billionaire Baby Boom: Head-spinning WSJ piece on billionaires using unscrupulous surrogacy agencies to have dozens of children. And I thought it was just Elon.

The Epstein Investigation: Good nitty-gritty reporting by the NYT investigative team on Epstein’s early years and his lifelong history of fraud.

Welcome to the latest edition of Capital Efficient. Let’s get into it.

Vertical AI Defensibility

"Why won't OpenAI just do this?" is the question every vertical AI founder fields from skeptical investors. Karpathy’s 2025 LLM Year in Review offers a useful rebuttal. Using Cursor as his case study, he extrapolates to explain the broader app-layer opportunity. In short, "LLM apps" provide real value through:

• Context Engineering: A best-in-class vertical AI platform pulls in all relevant context, structures it optimally for the task, includes pre-built expert-level prompts, and lets users skip the “setup” phase entirely.

• Orchestration: Quality models are proliferating fast. How many startups were making calls to Gemini six months ago vs. just OAI/Anthropic? Knowing which model excels at which task is increasingly complex. Vertical AI apps need to orchestrate calls across multiple LLMs, optimized for each step of a given workflow.

• UX: ChatGPT’s empty text box and Claude Code’s black terminal don’t give typical business users much to work with. The best vertical apps harness LLM power for discrete industries and make it simple for workers to incorporate AI into their day-to-day.

• Autonomy: AI outputs aren't perfect. Depending on the task, a human-in-the-loop is necessary. Vertical apps should let users make the tradeoff between automation, quality, and efficiency, choosing when to let AI run autonomously vs. when to double-check its work. This flexibility is key for real-world adoption.

My view: the labs will make immense money providing inference to this next wave of apps. Think of foundation models as raw commodities and vertical AI as the finished product. Vertical AI companies are already showing they can generate revenue faster than cloud-era software businesses. Every company needs to leverage AI, but horizontal models and their UX won’t cut it for most workflows.

What Karpathy calls “LLM Apps” (and what I call Vertical AI) is how AI actually gets deployed in the economy beyond prosumer and generalist use cases.

Model Cost Deflation + Anticipatory Building

The lesson of the last three years: build ~50% ahead of model capabilities, use the best available models to wow customers, and assume costs will fall. Cursor did exactly this and is now the fastest-growing software company in history. When they started building AI code gen, foundation models couldn’t reliably produce junior-level code. But they were positioned to capture the market as models improved.

This is counterintuitive for founders trained on cloud-era SaaS economics, where gross margin was sacred from day one. In AI, that logic inverts. The companies winning right now are the ones spending aggressively on inference to deliver a product that feels like magic, even if unit economics look ugly in the early innings. The bet is that cost curves will bail you out, and so far that bet has paid off every time. Founders who optimize for margin too early end up with a mediocre product that loses to a competitor willing to burn more on compute.

Customers always want the best model. Model pricing collapses over time. So max out quality over margin. Worst case, you lower model quality later to cut costs and drive up margin at scale. And by then the “discount” model will be better than today’s SOTA.

For a data-driven breakdown of AI input cost deflation, check out Tomasz Tunguz’s Gemini 2.5 Flash analysis here. I included an excerpt below from his great piece which drives the point home.

Source: https://substack.com/home/post/p-181936170.

Deals

Late December isn’t known as the hottest time for deal announcements, but a few caught my eye.

• Chai Discovery: Chai raised a $130MM Series B led by Oak HC/FT and General Catalyst, as well as a slew of other A-list investors. AI drug discovery isn’t a vertical I’ve spent much time on, but it has been gobbling up VC dollars. If it works, the medical breakthroughs could be world-changing. One thought I’ve found myself coming back to: how far into development do these companies take the drugs vs. when do they sell them off to the majors? My bet is some go all the way - starting off by partnering before developing their own drugs to compete against the Eli Lillys and Pfizers of the world.

• Lovable: Running away with the title of Sweden’s most valuable startup, Lovable raised a $330MM Series B at a reported $6.6B valuation. The round was led by Capital G and Menlo Ventures. Lovable surpassed $200MM in ARR last year and is now aiming for the enterprise with the goal of becoming “the last piece of software” a company ever buys. I am not sure even the founders believe that, but it’s a hell of a narrative.

What I Am Reading

• Raising A Series A in an AI First World: Solid founder service from Susan Liu at Uncork with spot-on advice for startup leaders thinking about raising an A in today’s market. The short of it: if you aren’t pre-empted, and aren’t growing very fast, be prepared to do a Seed-II. Well worth a read.

• Paying $3 for $1: The Google Doc Guru himself, Chris Paik, uses game theory to break down the dynamics at the frothier end of today’s private markets. His view is that the music stops once the public markets won’t accept AI asset prices, but given the proliferation of late-stage capital and tendency of tech co’s to stay private indefinitely, I can’t really say when that will be.

• How The Phone Ban Saved High School: NY Mag on the positive impact of NYS banning phones in schools. It was the phones after all. If Yondr were public, I’d be piling into it - there’s no future where phones aren’t eventually banned in schools nationwide as this gets replicated.

And if you’ve made it this far, I truly appreciate the readership and hope you have a great and restful holiday break. Until next year, thanks for reading.

Welcome to the latest edition of Capital Efficient - let’s get into it.

Weekly Radar

The State of Enterprise (Open)AI

OpenAI released their State of Enterprise AI report this week. While in part a marketing document, it still has some good nuggets. The cliche about the future being unevenly distributed continues to hold true: AI adoption varied wildly across industries. Tech industry adoption of AI-coding was off the charts and should almost be viewed separately, but healthcare and manufacturing weren’t far behind, each seeing a 7-8x YoY spike in usage among OpenAI customers.

One caveat - this data set only applies to a subset of OpenAI’s enterprise customers - no API data is included, so lawyers accessing OpenAI via Harvey or bankers accessing it via Hebbia don’t show up here. Still an interesting graphic.

Educational services was the biggest laggard, and slow growth in professional services stood out, though I suspect that’s a methodology issue.

Two other fun facts:

• The UK and Germany are the largest ChatGPT Enterprise markets outside the US by customer count.

• Japan has the largest number of corporate API customers outside the US.

As OpenAI goes global, I wonder if being early in the Japan/Germany/etc. OAI office ends up having similar cache that having been part of an Uber geographic launch team used to carry. Read the whole thing here.

Homebrew Software Raises The Bar For The Pros

Over the last few weeks, a number of stories about white collar professionals vibe coding custom software have gone viral, accompanied by the (incorrect) take that vertical AI is dead because users will build their own tools instead of buying off the shelf. One such example: a lawyer who built their own case management platform.

No doubt ambitious, semi-technical professionals are experimenting with Claude Code, Lovable, and others to see what they can jury-rig together. But that doesn’t mean homebrew software will replace the role vertical software plays within industry.

I tend to agree with the standard pushback from vertical investors who aren’t ready to write off their portfolios to a future where every worker becomes a vibe coder. The counterargument is usually threefold:

1. This homebrew software will break and there’s no one to maintain it.

2. It won’t be compliant with any security frameworks, meaning IT will block its deployment.

3. Building this software is a bad use of the firm’s time. They should focus on their core competency.

All fair points. But the rise of in-house AI tools should still matter to vertical AI founders (and investors). With some workers building their own “verticalized” systems - and all workers increasingly turning to foundation models for task assistance - the bar for what a vertical AI platform needs to do to be interesting and command a healthy ACV is getting higher.

If you can solve the same workflow with ChatGPT/Claude, or if someone ambitious on the team can use AI to build an internal tool that solves the same problem a vertical AI startup is using as their wedge, it gets challenging to convince buyers to allocate budget to your product.

So how do vertical AI companies stay ahead of the ambitious associate with a Claude Code subscription? The moat comes from orchestrating complex, multi-step workflows that would take months to replicate in a homebrewed tool, data flywheels that get smarter with every customer interaction, deep integrations with unfriendly systems of record, and real domain expertise baked into product by default. The bar is higher, but there are still many ways for vertical AI to win and to build defensibility over time.

Deals

Unconventional AI raised a $475MM Seed round this week. They knew exactly what they were doing calling a half-billion dollar financing a “Seed” - and everyone still took the bait. The round was led by a16z and Lightspeed. Unconventional is building energy-efficient compute for the AI era, claiming it will draw on principles of biology to do so. The company is led by Naveen Rao (former head of AI at Databricks). This round values the company north of $4B.

Quanta raised a $15MM Series A to offer full-service accounting to startups. The company uses AI to manage bookkeeping, reconciliations, tax filing, and FP&A for startup darlings like Browserbase, Decagon, and Braintrust. The round was led by Accel and aligns squarely with my view that there may be more opportunity competing directly with legacy service providers vs. selling them software.

What I’m Reading

Claude’s Soul - An intrepid researcher extracted Claude’s “soul” document from the Opus 4.5 system prompt. Worth reading to understand how the labs are trying to shape model behavior.

How AI Alters Work - Anthropic surveyed its own staff about how AI is changing the way they work. My biggest takeaway: AI is reshaping the mentor-mentee dynamic in many junior roles, with juniors “self-educating” via AI instead of walking over to a coworker or Slacking them a question. The AI is almost certainly more accurate, but it also has an interpersonal cost.

They Killed My Source - Shane Goldmacher in The Atlantic reports on his digital relationship with an alleged Iranian spy. A good window into the role the Internet plays in modern espionage and a real page turner.

Weekly Radar

Thoughts on AI-Native Services

The fastest-growing AI-native startups tend to either i) sell to other startups or ii) sell to very large incumbents (think Delve for startups getting compliant vs. Harvey for global law firms trying to retain an edge).

And in professional services, we are seeing growth-stage firms buying middle-market players, rolling up competitors, and infusing them with AI. Two recent examples:

• BVP teaming up with Centerbridge to buy out an accounting firm (CRI) as a roll-up platform.

• Thrive backing Crete as an acquisition vehicle for accounting firms; Crete reportedly has a mandate to deploy $500MM on roll-ups over the next two years.

An interesting route, but likely a growth or PE-profile return. These are existing businesses; and you’re making them more efficient, not building from $0 to $X00MM. To be fair, it’s also less risky.

An approach with more upside: build AI-native services firms from scratch. The catch is that professional services run on trust, and clients rarely switch vendors. But startups are comfortable relying on AI, often lack existing vendor relationships, and prioritize speed and price over pedigree, making them the perfect entry point.

And the timing is right. Models have gotten good enough at knowledge work to handle the bulk of what eats up junior employee time. Inference costs keep falling. While traditional services firms remain inherently cautious and slow to adapt, creating an opening for insurgents.

Crosby is executing well here. They’ve built an “agentic law firm” targeting high-volume legal work for startup clients, and they boast customers like Cursor and just raised a Series A led by Index and Elad Gil. They’re not competing with white-shoe firms; their competition is boutique startup firms or overworked junior associates at megafirm startup practices who often end up treating this work as their last priority, but at sky-high rates.

Selling AI-native services to startups unlocks venture-scale growth. Over time, you can either stay in that lane if the market proves large enough, or move upmarket as your offering matures and your customers scale. What other startup pain points could you build a Crosby for?

Accounting

Every startup eventually transitions from cash-based to accrual-based accounting, and most struggle to find affordable help preparing for audits and investor due diligence. An AI-native accounting firm could automate categorization, reconciliation, and reporting at a fraction of traditional CPA pricing. Picture pasting your cash-based financials into a platform, answering some prompts, and getting accrual-based output with modules for monthly close, revenue recognition, and audit prep baked in. A self-serve mode handles routine tasks so your team reserves billable hours for higher-value advisory work.

Communications

Most startups can’t afford a traditional PR agency. The alternatives are doing nothing, hiring a junior freelancer who doesn’t understand the space, or burning founder time on outreach that goes nowhere. An AI-native PR firm changes the calculus. It drafts press releases and pitch emails in minutes, generates targeted media lists based on beat and past coverage, and surfaces warm leads by tracking journalist engagement. Founders get AI-generated briefing docs before interviews and real-time monitoring when coverage drops. All at startup-friendly pricing and startup speed.

Design and Creative

Every startup needs pitch decks, marketing collateral, and brand assets, but traditional agencies charge $15-30K for a brand identity and freelancers are hit or miss. An AI-native design firm could generate pitch deck templates, produce ad variations at scale, and spin up landing pages in hours, all while enforcing brand guidelines and managing asset versioning. Human designers handle the high-judgment work like core brand and key visuals; AI produces the long tail of deliverables.

These are three AI services bets we’re actively studying. If you’re building an AI-native services business, I’d love to hear from you.

Deals

Beacon (or, Constellation…Meet AI): General Catalyst made waves in vertical software last week with the public launch (and concurrent Series B) of Beacon, their AI-native Constellation Software competitor. Beacon emerged from stealth with a reported $250MM raised to date and a $1B valuation.

Constellation Software ($70B market cap as of today) has historically purchased legacy, sub-scale vertical software companies (often several in the same sector), rode out the cashflows, let the software degrade, and eventually ported users to their best offering in a given market. This has worked well as they focus on mission critical systems that end-users can’t or won’t switch off.

Beacon’s bet is straightforward: buy vertical software companies, then use AI to modernize and refactor previously untameable codebases across hundreds of legacy, industry-focused players. The goal is to acquire the same large but fragmented customer bases in the Constellation mold, but modernize them instead of managing their decline.

The question is how much operating and technical leverage Beacon’s AI teams can extract from the latest coding tools. Only time will tell if they can actually improve what are often fairly stale offerings.

It’s also not a coincidence that Beacon is hiring in Toronto (Constellation’s HQ) with job listings that mirror the work of Constellation’s acquisitions team. Constellation will inevitably bring on AI engineers and attempt some version of this themselves. But Beacon’s DNA, leadership, and GC’s financial backing make them a formidable threat. My guess: within 18 months, Constellation announces an AI modernization initiative of their own. By then, Beacon will have a head start and a playbook they can’t easily replicate.

For sub-scale vertical software founders, this is a good thing because it means another credible acquirer ready to deploy capital. Between Beacon and Percepta, GC is making a clear bet that the next wave of value creation comes from applying AI to existing workflows, not just building new ones.

What I Am Reading

• New York Is An Industry Town: makes the case for NYC as the HQ of applied AI. The basic argument: models get built in SF, but use cases get built in NYC. A good counterpoint to the rampant SF hype.

• Can The AI Boom Pay For Itself: Data-driven piece by exploring the “are we in a bubble” question by comparing inference costs and cap ex spend. He is also running a monthly tracker to refresh this data set over time - the current trajectory is “so far, so good”.

• The Runaway Monkeys Upending the Animal-Rights Movement: Ever wondered where research monkeys come from? The New Yorker goes inside the industry and highlights some strange bedfellows among its opponents.

  Share Capital Efficient with a friend.

  Share

  Subscribe now

I’m Pat McGovern and welcome to the fifth edition of Capital Efficient. Let’s get into it.

Weekly Radar

Browser Agents

If LLMs are AI’s brain, browser agents are its hands. ChatGPT and Perplexity both recently launched agent-focused browsers, and a new crop of startups is building out the connective tissue between LLMs and the digital world. From SaaS apps to web portals to dashboards - the browser is where modern desk work lives. For those getting up to speed on the space, I’ve included a helpful market map below courtesy of Theta.

For vertical AI startups, browser agents are often the missing execution layer - the bridge between intelligence and action. They enable startups in logistics, healthcare, and insurance to actually do the work they aim to augment or automate. These domain-specific agents, capable of handling complex, industry-tailored workflows, are especially interesting. It’s still an open question whether that layer becomes a standalone market or gets commoditized as general-purpose agent infrastructure improves.

There’s also a future where leading vertical AI platforms build their own browser agents, and where their execution quality determines overall product quality. And there’s another timeline where foundation models develop such powerful agents that the category itself becomes irrelevant.

Re-architecting the internet to be agent-first will be a 5-10 year process. Until then, browser agents that can navigate the web like a human will be essential if AI is going to fulfill its potential in the workplace.

VC Expectations → ARR Inflation

It’s time for a reset on early-stage revenue expectations.

Here’s the current state of play:

• VC’s publish select, unrepresentative examples of super high-growth companies, largely ones that are prosumer or selling to other startups. And then message that any startup not growing $0→$5MM in Year 1 isn’t interesting.

• Founders overcorrect in response. They inflate ARR in their decks, claim one-time pilots are recurring, fudge selling months, and position estimated expansions and unsigned revenue as ARR.

• VC’s get mad that founders are lying to them (and, in the process, committing borderline securities fraud).

My view: building a durable, high-growth business isn’t always going to allow for a $0 → $5MM Year 1 ramp. VC’s - let’s all take a breath. And founders - just tell the truth - there is nothing worse for trust than having to walk back and reclassify revenue the bulk of your ‘revenue’ on call #2 once the questions start flying.

Disclaimer: this is not aimed at any particular founder; this is an observation of an incentive structure in today’s early stage market which isn’t particularly healthy for trust building.

Deals

OpenAI Goes Public Benefit: At long last, OpenAI announced its plan to make the switch from non-profit to a quasi-for-profit model. As part of this transition, Microsoft will get a 27% stake in the successor entity, and OpenAI will be reconstituted as a public benefit corporation overseen by the OpenAI Foundation. MSFT also gets a 20% revenue share in OpenAI, but only until an expert panel determines AGI has been achieved. With OAI revenues continuing to ramp into the tens of billions, this is setting both parties up for a fun “are we there yet?” game on the question of AGI.

Seeing AGI popping up this week in MSFT corporate press releases is another sign we have crossed the rubicon re: AGI going mainstream. Founders should position their products around the assumption that baseline AI improves dramatically in the next few years. Build where the puck is going.

Legal AI Faceoff: This week was a big one for legal AI announcements, with Harvey and Legora both announcing $150MM rounds (although Harvey’s was done at 5x the valuation). Many of the fastest-growing vertical AI platforms sell high-ACV, enterprise products to professional services firms who are facing pressure to adopt AI to remain competitive. The rare back-to-back announcement screenshot below is courtesy of yesterday’s Axios Pro-Rata:

Legal AI is already seeing winners emerge and companies like Hebbia and Rogo are playing similar roles in finance/consulting. AI in these categories is seeing adoption by blue chip customers who would have never trusted AI even two years ago.

Accounting still doesn’t have a clear early winner as the AI platform of record. And most of the innovation capital there has been going towards AI-flavored rollups. As foundation models get better at wrangling spreadsheets, expect an analogous funding arms race in that space.

What I Am Reading

• The AI Wildfire: Dion Lim’s fantastic piece in uses the metaphor of AI as a wildfire (vs. a bubble) to take stock of the tech landscape. Fun fact from this one: AMZN traded at ~$5/share in 2000 before drawing down 95% to ~$0.30/share, only to then return an aggregate ~720x over the next 20 years (the biggest swing by an order of magnitude of any of the Web 1.0 and 2.0 public co’s he analyzes). Dion’s breakdown of Fire-Test KPIs by different players in the AI value chain is a sharp lens for thinking about durability. Worth reading the whole thing.

• Blowing Bubbles: Max Schoening from Notion lays out where he sees AI as having achieved true PMF. I think we are starting to see it in certain verticals at the app layer as well (e.g., prior authorization management, legal doc drafting) but, Max makes a strong case for the early breakouts.

Share Capital Efficient

Weekly Radar

Vertical AI vs. The Foundation Models

Every enterprise-focused app is now in quiet co-opetition with the foundation models, even if they haven’t realized it yet. In horizontal markets, this is already a given: OpenAI launched a browser today. They will undoubtedly launch a CRM in due time, they will go after Google’s ad business, and they will continue to build out large, mass-market applications for core business tasks. Imagine what Google did with G-Suite but on steroids.

Recent remarks I heard at a talk by Anthropic’s Chief Customer Officer were enlightening on what this means for the vertical software players. In his talk, he explained that for very large, regulated orgs (e.g., hospital systems, insurance carriers), they are currently sending teams of what are effectively FDEs to work with these huge enterprises and get them consuming Anthropic tokens. Anthropic is positioning itself as a safer, more regulation-friendly alternative to OpenAI with the aim of winning over more risk-averse institutions.

In my last post, I wrote about Custom Automation Platforms and how they represent an alternative way to build vertical AI for the enterprise. The foundation model labs are clearly planning to spin up their own version of this, as a way to directly capture compute spend and cut out the software-provider middleman.

The rub: many vertical AI startups pitch investors by explaining how they will start in the SMB or mid-market of insurance or healthcare, and eventually they will earn the right to move up market and win huge enterprise contracts. To do so, they will need to compete not only with the CAPs, but also with the foundation models that drive the core of their products.

IDE for X

AI agents are here, and they work, kind of. But there is increasing awareness that they will need more contextual infrastructure and guidance than initially anticipated. While AGI is reportedly on the horizon, many practical gaps remain.

Coding is well suited for an agentic workforce. There’s an established IDE and there’s fairly clear rules about what “good” looks like. Deploying agents in other verticals is more complicated. For most roles, tasks are multimodal, only loosely codified, and touch on many different working environments.

For example, an insurance broker may find themselves needing to log into and update their own CRM, a commission-tracking platform, email, and an insurance carrier portal, all before lunch time. 

The agent-meets-world challenge can be seen in the rush of capital into RL-environment startups and the immense sums the foundation labs are spending on RL training. This can also be seen in the glut of browser agent startups looking to give AI agents the ability to reliably navigate the browser-based workflows that make up most of today’s white-collar work.

Founders face a fundamental choice: try to build AI agents on top of existing legacy systems vs. take the time to rebuild the scaffolding their agents need to thrive.

The tradeoff depends on the industry. In slower, legacy categories, you often have to play the existing game; in faster, tech-forward sectors, there’s an opportunity to rip-and-replace the stack and rebuild for agents. Startups taking the rip-and-replace approach tend to focus on AI-native companies and provide them with new infrastructure. For example, Campfire and Rillet’s ERP customers are Series A–D startups, not traditional businesses. Serving legacy firms is harder and requires more integration, but there are far more of them to go after.

News Round Up

• Serval: The demand for “AI for IT” continues apace, with Serval announcing a $47MM combined Seed/A round led by General Catalyst. Instead of laying automation onto legacy systems, Serval’s bet is to build an AI-native IT automation from the ground up. Most of their public customers are startups (e.g., Clay, Mercor) where there’s less legacy ops debt to tackle.

  Starting fresh helps avoid some of the challenges I outlined above - it’s much easier to build something like an IDE for IT Ops, and then build an agent on it vs. letting it run wild on your existing, human-centric systems.

  OpenEvidence: Medicine is proving to be a field where AI - with just its current capabilities - can have a big, real-world impact. Yesterday, OpenEvidence announced a $200MM Series C at a $6B valuation spearheaded by GV, Coatue, and Thrive. The business aims to supercharge doctors and nurses by allowing them to turn patient symptoms into clear diagnoses via natural language search, and helping them with workflow automation around things like prior-auth requests.

  I’ve had some of the highest-paid medical specialists in NYC click off Epic and turn to WebMD - at a minimum, OpenEvidence is an improvement on that.

What I Am Reading

• Karpathy x Dwarkesh - This fits more under “What I Am Watching” but Andrej Karpathy sat down with Dwarkesh Patel for a discussion on the state of AI that went briefly viral. I’m still digesting it, but depending on who you ask about the interview, it either means that the AI bubble is popping or that the opportunity has never been greater. Decide for yourself.

• Why Vertical SaaS is Riding the Waves of AI to New Heights - Great piece by giving his take on how recent advances in AI capabilities will accrue to vertical software co’s.

• BVP Demand Gen Guide - Bessemer put out a solid overview of B2B demand generation tactics for the modern startup. Worth a read for anyone trying to crack mid-market / enterprise buyers.

Weekly Radar

The Rise of Custom Automation Platforms

The last few weeks have seen the public rise of a new category of software companies that don’t really have an agreed upon name yet - for the sake of simplicity, let’s call them Custom Automation Platforms. Every Fortune 500 company is strategizing around how they can implement AI in their business to save money, drive topline growth, and maintain their competitive edge.

But for the largest enterprises, most vertical AI startups are not broad enough or malleable enough to slot into their idiosyncratic workflows. And vertical AI platforms can struggle when pitted against decades of calcified internal processes and procedures.

So how do large companies solve this? Historically, when these businesses can’t find an off-the-shelf piece of software to automate something or surface some business insight, their alternative has been to hire consultants to build internal tools. Evidence of this can be seen in the oft-cited reports around Accenture generating billions in “AI” bookings. However, the products that these consulting engagements produce are rarely best in class.

Now, a third path is emerging: the rise of custom automation platforms (CAPs). These CAP startups look to automate some of the most complicated workflows in the largest and most demanding enterprises (think top healthcare payors, CPG brands, hardware manufacturers, etc.)

These businesses have a heavy forward-deployed aspect and start off looking like consulting , with the hope that over time they can build a core technical asset with enough reusable automation components to allow them to scale like a high-growth software startup. With the cost of code generation falling exponentially, it makes this CAP model, which would previously have been too costly and time intensive, look instead like an exciting new way to deploy AI in the enterprise. CAPs are fundamentally new types of software businesses that have only now been made possible by advances in AI.

And custom automation platforms are having a moment. There were three big developments in the CAP space over the last few weeks:

• Distyl: Distyl, the CAP that has been in market the longest, announced a $175MM fundraise which valued the business at almost $2B - a 9x jump from their Series A valuation. This round was led by top funds like Lightspeed, Khosla, DST Global, Coatue, and more. Some examples of Distyl’s recent custom automations include an AI system for PA decisioning built for a top healthcare payor and an AI insights engine for mapping demand and supply chain changes facing a top OEM.

• Brain Co.: Brain Co. came out of stealth with $30MM in funding led by Elad Gil and Jared Kushner, as well as a bevy of tier-one angels. Brain Co. describes themselves as building “AI for the world’s most important institutions.” They are intentionally trying to stay broad and avoid verticalization, instead looking to serve a wide variety of industries that share common challenges. Like Distyl, they are taking an early view towards healthcare and supply chain as places where they can make early inroads.

• Percepta: Not to be left out, shortly after the news of Brain Co.’s raise broke, General Catalyst announced their own newly incubated “AI transformation company” called Percepta which will focus on “transforming critical institutions using applied AI.” Percepta is swimming squarely in the same lane as Brain Co. and Distyl. They are hiring Applied AI Engineers in the US and Europe and are targeting large organizations in the “industries that power the world” including energy, healthcare, and manufacturing.

Using Consulting to Build Tech

All of these CAPs are doing unscalable things for the moment in hopes of building out their core technical platform, which should improve as they successfully automate more and more enterprise work. For now, much of what they do doesn’t look different than consulting, but each of these projects creates more reusable components for their proprietary platforms.

In terms of their flagship technical asset, Distyl’s technical platform is called Distillery, while Percepta offers a core asset called Mosaic, which enables enterprises to “deploy production-grade, industry-specific AI workflows in minutes instead of months.” All of these seem to draw inspiration from Palantir and their Gotham/Metropolis platforms.

One big question in the CAP category is how many of these workflows get built by forward-deployed engineers vs. customer employees, and who is responsible for maintaining them in the long run.

The CAP space will likely produce mulitple large winners, and while the very top of the market (e.g., F200) is getting a little crowded in terms of VC-backed CAP startups, I think there are still many opportunities in this space. If you are building a Custom Automation Platform, I’d love to hear from you.

News Round Up

Prepared, a public safety vertical software platform that had raised $150MM from funders like a16z and General Catalyst, announced it would sell to Axon, a leading public safety vendor best known for its tasers and body cameras. Axon has already been selling municipalities a GPT-wrapper called Draft One for writing police reports and this partnership should be a big unlock for Prepared given Axon’s existing relationships with nearly every hypothetical Prepared customer.

This is seemingly an early exit for Prepared with the business having raised an $80MM Series C in Q2 2025. My speculation is that selling AI to the government is still fairly painful, and they had to cut a deal with Axon to get access to customers at scale. This makes me wonder about the fate of many other public safety AI companies that have raised in the last few years using Prepared as a rapidly growing comp. Prepared was reported by some outlets as having sold for ~$800MM but there’s been no official confirmation of that number. For context, Axon’s market cap today is $56B so that scale acquisition would be a mere drop in the bucket for them. Axon generated $670MM in Q2 2025 revenue, $290MM of which came from their rapidly growing software business.

EvenUp @ $2B. This summer, Harvey - the leading legal AI platform for large law firms - announced a Series E at a $5B valuation with reports surfacing that the business had surpassed $100MM ARR. Now the plaintiff’s bar is getting in on the act with EvenUp announcing a $150MM Series E at a $2B valuation. EvenUp targets plaintiff-side personal attorney firms and has been expanding its offering after establishing a foothold drafting automated demand letters. An impressive valuation given the size of the law firms it targets.

What I Am Reading

• Ethan Ding’s Substack (a/k/a Mandates) - I really enjoyed two pieces in Ethan’s Substack - the first on OpenAI’s potential ads strategy and the second on the agent builder / cloud provider tension. Both are must-reads and very thoughtful.

• ICONIQ State of Software 2025 - ICONIQ has released their latest State of Software report - always worth a read to get a pulse on the market.

  If you enjoyed this edition of Capital Efficient, sign up below to get on our list.

Weekly Radar

Vertical AI → Business Model AI

Driving IPO-scale outcomes in the land of vertical software isn’t easy - this is evidenced by the fact there are only about ~20 public vertical SaaS co’s founded in the last 25 years. Post-GPT, interest in vertical software has exploded, and now founders are slicing their verticals thinner and thinner to stand out from dozens of venture-backed competitors (e.g., we aren’t software for truckers; we’re software for refrigerated box-truck operators on the West Coast.) While this can make you distinct, it can also cap your addressable market.

I’m increasingly interested in vertical software that tackles an entire business model from Day One rather than locking into a narrow sub-industry. While each industry has its nuances, certain types of vertical software are more generalizable to adjacent categories than founders often admit — and a larger initial customer base can be a big advantage. Our portfolio company Harmonyze is a great example of this - they are building for franchisors (which is more of a business model than a single vertical as this encompasses everything from QSRs to gyms to hardware stores and swim schools). Another example would be building an OS/command center for PE-backed businesses rather than just for car washes or for laundromats or for fast food restaurants, etc.

n.b. - this idea of business model software goes against the typical advice of start very narrow and expand outward; but for Vertical AI given the potential speed of product development (and expectations around speed of revenue by Series A/B investors), it may be worth considering a broader initial GTM target base. If you do start narrow, have a plan to widen your target fast - not in 2-3 years. Also, for really complex workflow automation (e.g., prior auth for providers/payors, this is far less applicable).

This phenomenon of “business model AI” can also be seen in the roadmaps of Hebbia, Harvey, and others who are evolving from “AI for financial firms” and “AI for law firms” to the more generalizable “AI for Professional Services” approach. Harvey’s public ICP is AmLaw 250 Law Firms but PwC and other consulting firms have become some of their most important customers. Hebbia has also branched out from finance into Big 4 consulting. This land grab by Vertical AI startups into industries with similar business models is one I expect to continue.

News Roundup

Figma’s IPO Delivers

As predicted last week, outsized demand for new tech IPOs coming to market led to a huge pop in Figma’s first day, with trading closing at $122/per share and as of today at noon, the stock was still at ~$80/per share (more than 2x above where banks had initially priced it). While it’s great to see the IPO window nominally open, companies still need to be doing several hundred million in revenue with a 40% YoY growth narrative to excite the public markets and these are relatively few and far between among today’s late-stage, venture-backed private co’s.

Harvey Crosses $100MM ARR

Harvey announced this week that it crossed $100MM ARR. This aligns with my observations that there are two kinds of AI-native software companies that have demonstrated an ability to scale very quickly

• Startups that sell low-cost, PLG to thousands of prosumer users (e.g., Cursor) and;

• Startups that sell very high-end platforms to the blue-chip players in their category (e.g., Harvey for Legal, Abridge for Health).

To date, I haven’t seen as many AI-native companies serving the mid-market that have exhibited this kind of explosive growth from $0 → $100MM ARR but that may change as AI creeps into more of the economy and becomes table stakes for all businesses.

The Next Brex vs. Ramp

A new rivalry is taking shape in the land of enterprise resource planning. Rillet - an AI-native ERP - announced this week it raised $100M from a16z, ICONIQ, Sequoia, and others. This comes on the heels of DOSS raising $18MM in a Series A led by Theory Ventures in April 2025. I predict DOSS and Rillet will follow an analogous trajectory with VCs pouring cash into each of them as they publicly duke it out to be the one, true AI-native ERP. In my view, the market for ERPs is a big one and multiple winners will emerge.

What I Am Reading

• “SEO is Dead. Say Hello to GEO.” (NY Mag) - John Herrman with an overview of the challenges the SEO-centric web will face as AI tools replace search engines for information discovery. The future of ad-tech (and the open web) will be written by the decisions the foundation model labs make over the coming years.

• “Bessemer Venture Partners (2025)” (HBR) - Earlier this summer, Harvard Business Review released a case study on the multi-stage firm’s expansion into Seed. Interestingly, the case shows Kent Bennett citing Andreessen’s push into Seed as a forcing function for BVP doing the same. Good read for any pure-play Seed fund trying to understand multi-stage motivations.

Weekly Radar

Vertical AI -> Custom Enterprise Software

The opportunity AI presents to automate large swaths of back-office and middle-office tasks in corporate settings is undoubtedly real. But exactly how AI’s promise and abilities will be delivered to buyers is still being worked through. Many of today’s best vertical AI founders are building module-based platforms with an eye toward ultimately becoming the default platform for their given industry to automate repetitive, formulaic workflows.

However, in the very largest organizations, there is a certain scale and style of business where vertical AI platforms - which are inherently “one-size-fits-all” - may not be enough. Large companies often have complex and dated internal storage systems, mountains of technical debt, and highly specific internal processes they don’t want to deviate from. It is very difficult to build an industry platform that is applicable across the top end of any given market; and those that succeed often do so by offloading customization to entire ecosystems of SI’s and consultants.

To date, the needs for automation and internal tooling within the F500 have been filled by “digital transformation” consultancies and we have seen the Accentures and Deloittes of the world make hundreds of millions in top line by serving as custom dev shops building internal AI tools for the world’s largest orgs. My belief is much of this consulting revenue will be eaten by productized versions of the services these firms provide, and these products will be built by best-in-class technologists who can drive faster implementations and better outcomes.While I love the traditional vertical AI business model, I am also increasingly excited about platforms that can enable rapid deployment of automation solutions and internal apps within the F500. With many predicting that the cost of code is heading towards zero, companies will demand infinite customization and will need a trusted technical party to enable this shift.

The work these custom enterprise software platforms do may start off looking like consulting, but they will evolve to have some kind of technical asset at their core (think Palantir’s Gotham or Metropolis) that can be re-used across engagements and each engagement should serve to make the core asset faster, more effective, and give it a broader range of capabilities. These businesses will rely on FDEs to meet customer needs and will need to command high ACVs to justify their human capital and deployment costs. A handful of companies (Distyl, Braintrust, etc.) are already working in this space but it’s a massive market, with tons of white space, and is one I believe will have multiple winners targeting different markets and different size enterprises.

If you are building this kind of software, I would love to chat. Get in touch by replying to this email.

Early AI Winners & Large But Under The Radar Verticals

Some of the recent growth-stage fundraising announcements I’ve been seeing have me re-evaluating the vertical AI opportunity set. Many of the more obvious verticals have seen multiple, extremely well-funded AI-native challengers break out in the past 18-24 months and new entrants looking to compete are being crowded out and having trouble raising. Elad Gil had a great piece last week which surveyed how verticals like law and medicine are shaking out and seeing early winners emerging. With many “obvious” verticals seeing the early AI winners become semi-entrenched, what are the second order use cases that can scale to an IPO-level exit?

Some areas that may fit this description where I am spending time: equipment dealerships, franchisors, building supply distributors, home healthcare agencies, insurance agencies/carriers, pharmacies, and medical device distributors.

News Roundup

Money Can’t Buy You Class

Kylie Robison from Wired broke the story this week that Meta has been offering Thinking Machines Labs employees comp packages in the $100MM-$1B range to jump ship. And even more shocking - most of these TML employees have turned Meta down. This is either a big vote of confidence in the value of TML’s yet-to-be released product or a sign of how toxic the Meta brand is in the eyes of some engineers. Escalating AI engineer pay raises an interesting question - how can vertical AI companies attract top AI talent given the riches on offer at Google/Meta and the foundation model companies for this skill set?

Toma Brings Voice AI To The Dealership

Toma announced a $17MM Series A from a16z to build a voice-first OS for car dealerships. This is a good early example of startups using Voice AI as a wedge to break into what has been a notoriously tough space to crack. As the ceiling on what software is capable of doing rises with each new foundation model release, expect to see more big vertical funding rounds. It will be interesting to see if Toma can eventually displace Reynolds, CDK, etc. From my perspective, Toma fits the new “default” building model for vertical AI founders who are selling to mid-market and enterprise companies, where the founders need to fully embed with their early customers. Toma’s story also highlights the growing importance of design partners to serve as an AI sandbox as you tinker with early product.

Anthropic’s War Chest Grows

Anthropic is reportedly raising $5B at a $170B valuation - this is up 3x from their last raise six months ago. It’s a two horse race at the moment between OpenAI + Anthropic in the lead; this dynamic may also explain Zuck’s eye-watering willingness to pay up for their alumni in an effort to draw even.

Figma’s IPO Is The Hottest Ticket In Town

Figma’s $1.2B IPO is reportedly 40x oversubscribed and the expected IPO listing market cap has grown from ~$13B to $20B over the last week. Hold on for a serious first-day pop. And get ready for even more startups to call themselves the Figma for X (or is it the Cursor for Y?). Does this mean the IPO window is open again?

What I Am Reading

• “Drones & The Future of Warfare” (New Yorker) - the legend Dexter Filkins reports from Ukraine on how drone warfare has changed the battlefield and what it means for national security. Must read for any defense-tech investor.

• “The Boring AI Questions That Actually Matter” (Margins by Ranjan Roy & Can Duruk) - good check-in on where we are in the AI platform shift and what chapters remain to be written.

• “The Website at the End of the Internet” (NY Mag) - John Herman chronicles the rise of Reddit, its breakthrough into the mainstream, and the platform's relationship with AI.

• “Mary Had Schizophrenia, Until She Didn’t” (New Yorker) - fascinating article about the link between mental illness and autoimmune diseases, and how autoimmune treatments might have a huge “off-label” use case hiding in plain sight.