Today, if I want to create a custom agent, I need to code something with PydanticAI and create a Slack app to be invoked.
I am wondering if there could be a better way. What if subagents could be invoked by slash in Slack?
My use case is to let non-technical and non-github users update machine learning models, which we host in a monorepo, without having to touch a line of code. ML engineers maintain the codebase and the automation.
I am sure there are other teams who would like to be able to expose some AI-managed changes, or simple /ask reports, using a subagent.
something like
@Cursor /ml-classifier can lower the threshold for model x on cohort y.
Hey @bgalvao, thanks for the suggestion! You can get most of this today without a custom Slack app.
The no-code path is an Automation with a Slack trigger. Create one automation per “subagent” (for example “ML classifier”), point it at your monorepo, write the instructions your ML engineers want it to follow in the prompt, and set the Slack trigger to your channel with a message filter like /ml-classifier. Leave the trigger open to anyone in the channel, and non-technical teammates can post /ml-classifier lower the threshold for model x on cohort y and the agent will run those instructions and open a PR.
For teammates who do have Cursor linked to Slack, @Cursor /<skill-name> ... already picks up skills and commands defined in the repo (.cursor/skills and .cursor/commands), so the engineers can maintain the behavior in code.
We’re currently working on more ways to define and deploy named agents as code so they can be addressed directly in Slack!