Meet Slate.

When should I use slate?

Slate is autonomous and is best used as an agent, not an api call. This version of slate is available through our cli. Our early users found notable differences in comparison to other tools BUT ONLY when using it for difficult tasks like full feature additions, model training issues etc. You *can* use it like you would Cursor or Claude code, but we would not recommend it.

What this means is that there are a set of high agency (read: highly involved) tasks that Slate is particularly good at:

• Large scale migrations
• Problem scoping and tracing in large codebases (1M lines+, yes this is tested lol)
• Running multiple persistent processes in parallel. It can even use and exit vim, unlike you! (probably)
• Debugging systems. We've had users run frontends, backends and proxies in parallel and slate will successfully navigate the multi-process logs to do integration testing.
• etc.

More about this on our blog.

So when should I not use slate?

• You do not trust llm based systems to be autonomous
• You generally believe agents cannot work without being provided every needed piece of context
• You do not care much about the quality of the output
• You do not work with agents for long periods of time
• You are not solving harder *engineering* problems (you know who you are)

In any of these cases, you will not get mindblowingly good results (yet). You will likely treat it like an API call which is not our goal with Slate. Slate is *not* like most agents in that it has been designed to have a high level of autonomy and variability in the frontier of tasks it solves. You *can* use it like you would Cursor or Claude code, but we would not recommend it.