For about eighteen months, any team that could wrap a model in a chat box and call it an "agent" could raise money. That window has closed. Two of the better-known coding-agent startups shut down this quarter, and at least one large platform quietly folded a competitor's team into its own.
How we got here
The wrapper was never the hard part. Once the frontier labs shipped their own agents and the open tooling caught up, a thin layer over someone else's model had nothing to defend. Distribution, real workflow integration and trust turned out to be the moat, and most of the field had none of the three.
What this means for you
- Assume any single-product agent vendor may not be here in a year. Price that risk in.
- Favour tools that are easy to swap out. If your workflow is welded to one vendor's proprietary format, you have taken on their survival risk as your own.
- The frontier labs and the big platforms are the safer bet for anything you plan to depend on, even if a startup is briefly ahead on features.
A feature is not a company. This quarter the market started charging admission to learn the difference.
The quiet winners
Consolidation is not all bad news. The teams that survive are the ones solving a real workflow, not demoing a trick. Expect fewer launches, better products, and a lot less noise in your feed by the end of the year.
Worth your time
- The month's agent funding and shutdown tracker, if you want the raw numbers.
- Our forthcoming white paper, The Shifts in Software, which looks at where this settles.