Acyclic (YC F26) splits a coding task across AI agents in their own sandboxes, runs them in parallel and merges one patch. Judgment calls come to you in the terminal. The founders' stated goal: make human decisions the bottleneck.
Voice AI assumes you're alone. The Subvocal Company (YC F26) is building silent speech smart glasses for offices and public places. The founders use dictation heavily, but saying prompts out loud next to each other just isn't realistic.
https://t.co/2IJ6s0GITe doesn't just give each coding agent a copy of your code. It snapshots the whole running machine (processes, ports, database, memory), so every task starts in its own fresh cloud devbox. If that holds up, the bottleneck becomes review.
Lark's AI agents read a distributor's existing ERP and draft the daily reorder list. At Interline Hardware, purchasing says a three-hour job is now about 30 minutes of review. In the demo, POs under $5K to approved vendors go through on their own.
Voice AI assumes you're okay talking out loud. The Subvocal Company (YC F26) is building sensor glasses that transcribe mouthed words into text, and the founder says a minimal, natural murmur is meant to be enough. The target: offices and cafes.
Canos turns a vet's blood draw into a cancer signal for dogs. In its example report, an illustrative lymphoma case drops below 0.5% tumor fraction during treatment, then comes back at 7.6% around week 20. That isn't a diagnosis on its own. It tells the vet to look again.
Most startups skip paid ads because they're complicated and easy to waste money on, the founders say. https://t.co/GkXgaThaLj (YC F26) wants AI agents to do the agency grunt work, with a person still steering, and to become the Stripe of paid ads.
For travelers with weak passports, the flight is the easy part. Veeza AI (YC F26) pulls your passport details, fills the visa forms and works the government portals with computer-use agents. A human expert checks it all before submission.
Audora (YC F26) wants to turn fan fiction into full-cast audiobooks. You paste an AO3 or Wattpad link and it aims to give each character their own voice. The real test isn't chapter 1. It's whether everyone still sounds the same by chapter 40.