AI Services, Beacon's Bet & What I'm Reading
Capital Efficient #6
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: Rex Woodbury 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 Evan O'Donnell 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.