@Hi_Mrinal For 1, are the verifiable checkpoints overly deterministic? Seems like having some sort of a cheaper agent that lightly wraps these steps might be helpful so it can route back to a previous step if necessary
@stallmanseesyou@kentcdodds yeah that's what most mcp servers do. but we should have meta mcp tools, so an agent can inspect the tools available lazily instead of front-loading them all
@davis7 Composer 2.5 has been excellent. You don't need a frontier model for everything, and I feel like 2.5 is the perfect model for most tasks. If something I'm doing is type heavy or really requires some thought, I might switch to a better model.
Welcome @corsairdotdev as a Gold Sponsor of BuildX'26.
Building integrations shouldn't be the hard part. Corsair is creating open-source infrastructure that helps developers ship faster.
Register Now:
https://t.co/LArOu5GBxN
#BuildX26#Corsair#OpenSource#AI#AIAgents#SaaS
Tool calling is broken right now, and Corsair fixes it.
Corsair lets your agent connect to anything online via APIs. It stores all data in your db so you can access it anytime.
Built in webhooks, secure auth handling, strong types, and impossible to use wrong.
Add it to your project with `npm i corsair`
@kentcdodds another benefit is you can quickly deactivate the feature if something isn't going to plan.
instead of having to wait to rollback changes and redeploy, you can just toggle it off.
@zaimiri makes sense! if you want to make it even better, id suggest you try corsair. itโs easier for an agent to work with and builds more complex workflows reliably
@aakashgupta agent wars will always be on the API layer. if any product has sufficient demand for an API layer, it has to eventually offer one or risk losing their customer base.
cua and browser use are great on the UI layer, but i think they are a bandage solution until that product has api