@theo It's still being built for devs' way of thinking, although Claude’s desktop app is a disaster that should be the new look for an AI agent. Not that you can open it in Cursor where each thing is a project, not a chat, IMO.
@theo There’s no way a Chineese model is 2 dollars to run this 850b model and Anthropocene cost 50 for fable
The more expensive the. Keel the dumber less efficient it is. Check out my substack
Cognition Labs understood something early that everyone is starting to realize now:
AI agents should not feel like another app. They should feel like another person on the team.
That is why Slack made so much sense for Devin.
A bug gets reported in a thread. Product and engineering discuss it. Then someone writes:
“@Devin, please fix this.”
The entire conversation becomes the prompt.
Devin goes into its own environment, works asynchronously, tests the code, opens a PR, and reports back in the same thread.
Slack is not where Devin does the work.
Slack is the management layer.
That is the important distinction.
The agent works across GitHub, Linear, Jira, the codebase, CI, and its own environment. Slack is simply where humans assign, clarify, supervise, and receive the result.
Cognition realized that autonomous AI is not mainly an interface problem.
It is an organizational problem.
Once AI can actually complete work, the best interface is often not a separate chatbot or dashboard. It is the place where work already begins.
Slack.
Email.
GitHub.
Jira.
Linear.
The future of AI agents is not everyone constantly opening another AI app.
It is AI becoming an addressable member of the organization.
Tag it like a coworker.
Give it the context.
Let it do the work.
Review the result.
Cognition saw that with Devin before most of the market did.