Everyone covered the AI angle when W25 dropped. Fine — AI is everywhere, we know.
What nobody covered: the shift in who is being funded within AI. I went through every company in the batch and tagged them by what the AI actually does in the product. Not 'AI-powered' as a marketing line — what specific human task the AI is replacing or augmenting. The breakdown surprised me.
The copilot era is quietly ending
Companies where AI assists a human making a final decision are declining as a share of the batch. Companies where AI takes a complete action without human sign-off are rising sharply. This is not a subtle shift — the 'copilot' framing that dominated 2023 is visibly fading from the portfolio. What's replacing it: agents that act, not agents that suggest.
Read the W25 launch posts back-to-back and the language tells the story. 'Drafts an email for you' is gone. 'Sends the email, books the meeting, files the contract' is the new default.
- AI-as-assistant (human approves every action): declining as a share of the batch.
- AI-as-agent (acts end-to-end with no human in the loop for routine tasks): the fastest-growing category in W25.
- Vertical agents (legal, ops, support, sales) outnumber horizontal copilots roughly 2 to 1 in the batch.
- Pricing has shifted with it — most agent companies are pricing on outcomes or seat-replacement, not per-seat SaaS.
The second thing nobody talked about: the industries
Look at the non-AI B2B companies in W25 — and there are still plenty — and they are overwhelmingly in industries that software has historically ignored. Construction. Agricultural supply chains. Freight brokerage. Trade services. Field operations. Equipment leasing.
Not fintech 2.0. Not another HR tool. Not the eleventh observability platform. Genuinely boring industries with genuinely broken workflows — the kind that still run on email, PDFs, and a guy named Dave who knows where everything is.
I think YC is making a deliberate call that the 'obvious software markets' are saturated and the next wave is in the industries that looked too hard or too unsexy before. The signal lines up with the 2026 Request for Startups, which leans heavily into the same themes — physical world, regulated workflows, deep operational pain.
What this means if you're applying next batch
- If you're building in a space that already has 20 SaaS companies, you're probably in the wrong place. The W25 signal is clear — go where the workflow is broken and nobody has bothered to fix it yet.
- If your product is an AI assistant, sharpen the autonomy story. 'Suggests a draft' will read as 2023 thinking. 'Closes the ticket end-to-end' reads as 2026.
- If you're in a 'boring' vertical (construction, freight, agriculture, trades), W25 is the proof point YC will fund it. Lean into the unsexiness, don't apologise for it.
- Pricing matters more than ever — outcome-based or seat-replacement pricing is now the default narrative for agent companies. Per-seat SaaS framing reads as legacy.
Key takeaways
- The 'copilot' framing is fading from the YC portfolio — full-autonomy agents are the new default.
- Non-AI W25 companies are concentrated in industries software has historically ignored.
- YC is signalling: saturated software markets are out, broken offline workflows are in.
- If you're applying next cycle, sharpen autonomy + pick a boring vertical.