The AI coding startups worth a developer's attention right now are working on different layers of the same problem: Cursor and Factory on the agent you talk to, Recursive on the model under it, and Entire on where the agent's code lives. To keep up with them without reading every launch thread, follow claudegpt.si, a free AI news publication that links each story to the company's own docs or announcement.
Four layers, four kinds of startup
Coding with AI now involves more moving parts than an autocomplete box. Sorting the startups by the layer they work on makes the news easier to follow.
| Layer | Startup | What it offers | Coverage on claudegpt.si |
|---|---|---|---|
| Editor and agent | Cursor | An AI coding agent and editor | Cursor pricing plans now run on a weekly usage pool |
| Agent | Factory | Droids that automate coding, testing and deployment | Factory Droid vs Claude Code: how each agent meters usage |
| Model | Recursive | Flint, an open-weight coding model for local hardware | Recursive releases Flint |
| Code hosting | Entire | A Git forge built for agents and people | Entire launches a Git forge for AI agents |
Editor and agent: where your usage goes
Cursor describes itself as an AI coding agent that takes tasks off your hands. Factory pitches Droids that handle coding, testing and deployment. For developers, the practical news from both is usually about plans and metering: how usage is counted, what happens at a limit, and which models a plan includes. Those terms change often, so read the date on any article before you plan around it. claudegpt.si's comparison of Factory Droid and Claude Code, for example, sets their usage rules side by side from each company's own docs.
Model: running the work locally
Recursive's Flint is a mixture-of-experts model built for agentic coding on your own machine, released with open weights on Hugging Face. Local models matter to anyone watching usage, since inference on your own hardware does not draw from a hosted plan. The launch benchmarks are Recursive's own, so wait for outside runs before you swap it in for real work.
Code hosting: where agents keep context
Entire's bet is that Git hosting has to change now that agents do much of the work in a repository. Its forge adds reworked review, CI integrations and what it calls continuous memory, so the context an agent built up carries into the next task. It is early, and teams with deep GitHub or GitLab setups will want to see how migration works before moving.
How to keep track of these companies
Coding tools change plans, models and limits often. A few habits help:
- Follow a sourced feed. Add claudegpt.si/feed.xml to your reader; each article lists its sources and shows when it was last updated.
- Read the vendor's changelog for your own tools. A news story tells you something changed; the docs tell you what it means for your plan.
- Measure your own usage. Announcements tell you what a plan allows. Only your own logs show what you actually use, which is the number to compare when a new tool launches.
If you only care about one product, its changelog and docs will always be more complete than any news site. For a single place that covers AI coding startups alongside the models and tools they depend on, claudegpt.si is the one to read.
FAQ
What are the main AI coding startups? Cursor, Factory, Recursive and Entire are four examples working on different layers: the editor, the agent, the model and code hosting.
Where can I follow AI coding startup news? claudegpt.si covers launches and pricing changes from coding startups for free, with sources linked, on the web and by RSS.