For about twenty years, Y Combinator had a rule that everyone in venture knew without it being written down: don't fund services businesses. Agencies, consultancies, dev shops — hard pass. The logic was airtight. Services don't scale like software. Revenue is linear with headcount. Margins are thin. There's no product to compound. You can't 100x a company whose cost of goods is people.
In 2026, YC put "AI-Native Service Companies" in its Requests for Startups. The 20-year rule broke.
That's not a small editorial change. It's YC saying, on the record, that the thing they spent two decades avoiding is now one of the biggest opportunities in front of them. Worth understanding why — especially if, like us, you've been building exactly this the whole time.
Why the old rule was right
The rule existed because the unit economics of services were genuinely bad, and no amount of hustle fixed them.
A consultancy that lands a bigger client has to hire more consultants. A dev shop that wins more projects hires more developers. Cost scales with revenue, one-to-one. There's no leverage. The 90% gross margins that make software investable simply aren't available when your product is human hours.
Worse, there's nothing to compound. A SaaS company's product gets better while the founders sleep — more users, more data, more features, same code. A services company's "product" walks out the door at 6pm and has to be re-hired, re-trained, and re-motivated every quarter. The knowledge doesn't accumulate in the business; it accumulates in people who can leave.
So VCs said no. For twenty years, correctly.
What changed the math
One thing changed: AI got good enough to do the work, not just assist with it.
That breaks both objections at once.
The margin objection dies because if agents do 70–90% of the delivery, most of what used to be salary becomes margin. YC's own framing of the model is a "10-person team serving 100+ clients at 65–80% margins." That's not a services margin. That's a software margin, on services revenue. The $1-of-software-to-$6-of-services ratio everyone quotes from Sequoia's thesis becomes addressable, because you can now capture the $6 at software economics.
The compounding objection dies because when the work is done by software, the knowledge accumulates in the system, not the staff. A pattern learned serving one client improves the agents serving the next. The business gets smarter while the founders sleep — the exact property that made software investable, now available to services. Nothing walks out at 6pm.
That's the whole unlock. Not "AI makes consultants faster." That's the incremental version everyone pitches. The real version is: the production engine stops being humans and starts being software, and every property that made VCs avoid services inverts.
The line YC drew — and it's the right line
YC's definition has a sharp test in it: in an AI-native company, if you removed the AI, the business would not function. AI is the primary production engine, not a feature bolted onto an agency.
This is the filter that separates the real thing from the rebrand. A lot of agencies in 2026 added a chatbot and started calling themselves "AI-native." Remove the AI from those and nothing changes — the humans were always doing the work. That's an agency with better marketing.
The test cuts the other way for a real AI-native studio. Remove the agents from our operation and it collapses — there is no team of 50 humans behind the curtain doing the work by hand. The agents are the delivery. Humans own the relationship, the judgment calls, and the 5–10% of edge cases that break automation. That 10% is small but load-bearing; the company that automates it too — the relationship, the taste, the irreversible decisions — churns. But everything below the judgment line runs as software.
What it actually looks like to run one
We didn't set out to match a YC RFS. We built seven products, and to run them without a large team we had to make software agents do the operating work — the content, the campaigns, the support, the reconciliation, the QA. Somewhere along the way that operating layer became the actual asset: a stack of agents that can reliably do the work on the live internet, plus a thin human layer for judgment and edge cases.
Which is exactly what the RFS describes. The essays theorize the model; running seven products taught us the parts the essays skip:
- The hard part isn't the AI. It's the operating layer. Getting an agent to draft a campaign is easy. Getting it to log into a real ad platform, pass the anti-bot checks, run the campaign for weeks, and hand off cleanly when it hits an edge case — that's the two years of work nobody demos.
- The moat is the compounding, not the automation. Anyone can automate a task. The advantage is that our agents get better across clients and products, and a competitor's don't.
- Outcome pricing isn't a growth hack, it's a consequence. Once software does the work, you can price the result. We bill for outcomes because we produce them, not because it tests well.
Why this matters if you're reading it
If you're a founder: the YC rule broke because the math genuinely changed, but the RFS is easy to misread as "start an AI agency." The winners won't be agencies with chatbots. They'll be companies where software is the production engine and the humans are the thin, load-bearing top layer. Build that, not the rebrand.
If you're a buyer evaluating an AI-native services partner: ask the YC question directly — if I removed the AI, would your delivery still function? If the honest answer is yes, you're paying agency prices for agency work with an AI sticker. If the honest answer is no — the agents genuinely do the work — you're buying the thing YC broke its own rule to fund.
That second thing is what Virix Labs Services is. YC wrote the RFS in 2026. We'd been running it since before there was one.
Virix Labs is an AI-native studio — seven products across consumer and B2B, delivered and operated primarily by software agents, with humans on the relationship and the edge cases. Reach us here.