Introducing Agent Mode: Your autonomous engineering teammate
Agent Mode changes how you build. Instead of writing one file at a time, you describe the feature — and Dubleu plans, writes, migrates, and deploys it end to end.
Arjun Mehta
March 28, 2026 · 6 min read
Building software has always been an exercise in context switching. You write a component, then jump to the API, then the database schema, then the deployment config. Each switch costs time and momentum.
Agent Mode is our answer to that. Flip the lightning bolt toggle in the chat header and Dubleu enters a fully autonomous loop. It writes a structured task plan to .tasks/plan.md, works through each step — writing files, running SQL migrations, fixing TypeScript errors — and signals done when it's finished. You stay in the loop without doing the work.
How it works
When you send a message in Agent Mode, Dubleu first writes a plan. Every task is tracked in .tasks/plan.md as a checkbox list. You can watch it tick through in real time.
The agent writes code across as many files as needed. It understands your project's structure — existing components, types, database schema — and generates code that fits. When it hits a TypeScript error, it fixes it before moving on rather than leaving broken code for you to find.
Database-aware by default
If your project has a Supabase database, Agent Mode knows about it. It reads your existing schema before generating API routes or database queries. When it needs a new table, it emits a migration and runs it — credentials handled, row-level security applied automatically.
What's next
We're working on cross-project coordination — letting a single agent session span multiple linked projects in the same org. If you're building a monorepo or a separate frontend/backend, that's coming soon.
Try it today: open any project, flip the Agent Mode toggle, and describe your next feature.
Share this post
Found it useful? Pass it on.