How Has AI Impacted the Start Up Model?

Ask a founder how AI changed their company and the answer is usually about tooling. Code ships faster. Support replies write themselves. All true, and all the least interesting part of the story, because faster tooling has arrived in software every few years since the first compiler.

The change worth arguing about is a change in measurement. For two decades a startup’s health was read through headcount growth, burn rate and the size of the last round. Those numbers describe less every quarter. The number that describes the most now is revenue per employee: money the business makes, divided by the people it takes to make it. AI did not invent that measure. It made it the one that separates companies.

What gets missed is that businesses already sat at the far end of that measure, and they were not in Silicon Valley. Online gambling platforms have run on those economics for years. Shuffle, a crypto casino and online gaming platform where deposits and withdrawals settle on-chain in coins and stablecoins, publishes a catalogue of casino slots Canada titles that costs roughly the same to operate whether a hundred people play them or a hundred thousand. No inventory. No shipping. No warehouse headcount rising with demand. Each game is RNG-driven with a house edge fixed in its maths, so revenue is arithmetic across volume rather than a negotiation per customer, and the marginal cost of one more player rounds down to compute and payout rails.

That is a structural observation, not an endorsement of the sector. It is a useful one, because it means the economics AI-native startups are reaching for are not hypothetical. Somebody has run at that ratio long enough for the failure modes to show. Keep the measure in mind. It comes back below.

The batch where it stopped being a trend

At the March 2025 Demo Day, Y Combinator CEO Garry Tan said about 80% of the presenting companies were AI-focused. He also said that for roughly 25% of current YC startups, 95% of the code was written by AI, and that the Winter 2025 batch was growing 10% per week in aggregate, faster than any cohort before it.

That batch held roughly 160 to 163 companies, and 58 of the 163 were building AI agents specifically. Read those numbers together and the accelerator stops looking like a cross-section of software and starts looking like one bet placed 160 times.

Where the seed money went

Global seed funding across every sector came to about US$36.17 billion in 2025. Around US$15.25 billion of it, roughly 42%, went to AI-focused companies. That share was about 30% in 2024. In 2020 the AI slice was US$3.43 billion in total.

The growth is not the interesting part. The overall seed pool did not expand to make room for the new category. AI took share, and every point of that shift came out of something else that used to get funded.

Two companies that show the shape

Lovable, an AI app builder, reported about US$17M ARR with roughly 30,000 paying customers in February 2025. By around July it reported about US$100M ARR with roughly 45 employees. November brought US$200M ARR, and by mid-2026 the company reported about US$500M annualised, with 146 full-time staff on the books as of February 2026.

Replit tells a similar story from the usage side: more than 40 million users as of Q4 2025, more than 50 million by March 2026. Divide revenue by staff and you get a figure that would have read as a typo in 2019.

What revenue per employee breaks

Once revenue per employee becomes the number the board watches, several habits stop working at once. Hiring plans go first. A functional org chart assumes each new function needs a team, and the measure punishes that assumption. Fundraising signals go next, because a round sized to pay for headcount says something unflattering when the comparable company earns four times as much with a third of the people. Multiples built on growth rate alone wobble too, because they ignore whether the growth needed 40 hires or four.

This is where the casino comparison earns its keep, and where it stops. A gambling platform reaches an extreme revenue per employee figure because the product is a fixed mathematical edge applied to volume, and volume costs almost nothing to serve. The edge does not need staffing. An AI startup reaches a similar ratio for a different reason: the marginal unit is inference, and inference carries a real price that moves. Two identical numbers on a slide can rest on completely different ground. Founders who copy the number without the reason behind it find out during their first bad month of usage-based cost.

The costs AI did not compress

Nobody has automated money movement. Fraud review, chargebacks, failed authorisations and cross-border settlement stay stubbornly human and expensive, which is why payment gateway optimisation keeps showing up as a line item in companies that otherwise run on eight people and a model provider.

Compliance is the same. So is trust and safety, and so is the slow work of answering a regulator. Gambling operators learned that years ago and staff accordingly. Worth saying plainly about that sector: the house edge holds over time, no strategy or system removes it, and those products are for adults 18 and over playing with money they can afford to lose.

What nobody can tell you yet

Some confidently quoted numbers do not exist. There is no published figure for average YC founding team size in 2025, so treat claims about shrinking founder counts as anecdote wearing a statistic’s clothes. Full-year 2026 AI seed totals are not out either, which makes every projection about this year a guess with a chart attached. The measure is real. The precision people apply to it often is not.

Where the model actually landed

The start up model did not get easier. It got narrower. A small team can now serve a large user base, so the excuse for a bloated org has gone, and so has the excuse for slow revenue. The harder question is whether the ratio lasts. A company holding a high revenue per employee figure because its marginal cost is genuinely near zero can keep it. A company holding it because inference is currently cheap is renting the number.

Founders should know which one they are. Most do not ask until the invoice tells them.

One note on the example above. Online gambling is adult entertainment for players aged 18 or over, and the house edge behind those economics is built into every game by design.