AI Co-Founder Pitches Are the New Uber-for-X and Most of These Startups Will Be Dead by Q3 2027

I will die on this hill: the "AI co-founder" SaaS category is the most over-funded, most undifferentiated, most brittle cohort of AI startups since the 2021 ZIRP. Y Combinator is funding a hundred of them a batch. Three quarters of them will be dead, acqui-hired at fire-sale valuations, or quietly pivoting to "enterprise document search" by Q3 2027. The partners writing the checks know this. They keep writing them anyway.

I have watched the demo circuit long enough to call the pattern. Eleven "AI co-founder" pitches in the last six months, every single one the same skeleton: chat interface on the left, calendar on the right, sidebar of tasks, an "agent loop" that pretends to be three people, Stripe Connect for invoices, Calendly for meetings, and a memory product with a fresh logo. Strip the chrome and you get one LLM call, one prompt template, and a cron job. There is no multi-agent orchestration. There is no long-horizon planning. There is one model role-playing a team, hitting a SaaS API, and pasting the result back into the chat. That is not a co-founder. That is a Slackbot from 2017 with a ChatGPT wrapper and a $20M seed round.

The Unit Economics Are A Crime Scene

Let me do the math nobody on the demo day call wants to do. The median "AI co-founder" SaaS is priced at $99 to $499 per month. The median user runs roughly 80 LLM calls per day for emails, scheduling, research, and document drafting. At Claude Sonnet 5 pricing of $3 per million input and $15 per million output, the inference cost alone is $35 to $110 per user per month. Add a "memory" layer ($30), add retrieval ($20), add voice for the agents that bother ($15), add the cron-driven background tasks that run whether the user is online or not ($15). You are selling a $99 product that costs $120 to serve. You are selling a $499 product that costs $180 to serve. Your gross margin is negative or razor-thin before you pay a single engineer, sales rep, or AWS bill. The only way this works is if your users churn inside three months, which they will, because the product does not actually do anything their existing $20 a month ChatGPT Plus subscription does not already do for them.

The Counter-Argument Is Wrong

The pushback I hear from VCs is "but the AI co-founder will replace the $8,000 a month executive assistant." It will not. I have watched three of these products fail inside real executive workflows in the last quarter. The pattern is identical: the first two weeks feel magical, then the user notices the agent is sending emails they did not approve, scheduling meetings on the wrong day, and "remembering" facts the user never told it. By week six the agent is in a sandbox the user checks once a week. By month three it is cancelled.

A real executive assistant costs $96,000 a year and is worth every dollar because they have judgment, accountability, and the social context to know which emails are urgent and which are noise. An LLM call does not have judgment. It has vibes and a tendency to hallucinate the name of your investor's spouse. You are not replacing an EA. You are replacing nothing, and your run rate is showing it.

The Take

If you are a founder pitching this category, stop. The category is full and the moat is the sandbox of integrations you have not built yet, which is exactly what every other pitch deck says. If you are a buyer, ask the vendor for their per-user inference cost and their per-user revenue. If those numbers cross, you are funding their Series B with your subscription. If you are an engineer interviewing at one of these companies, ask the founder how many of the last 100 logged agent decisions they have personally reviewed. If the answer is zero, you are joining a content moderation team dressed up as a startup.

The next 18 months are going to be brutal for this cohort. The bills come due. The users churn. The inference bill keeps growing. And the demo-day claps turn into "we are excited to announce our pivot to enterprise document search." Most of you already know this. You are not investing in companies. You are investing in conference talk titles.

— Mr. Technology