Lately, I’ve been reflecting on how we can further optimize software development with AI. And then it hit me—what if we had a workspace, like Trello, specifically designed not just for humans, but to orchestrate collaboration with AI models?
https://aifa.dev will have this in its function
What’s Wrong With the Usual Approach?
These days, we often let AI decide how to implement a feature. It feels a little like having a Tesla on autopilot—smooth and automatic, but sometimes you just want more control.
What if we gave that autopilot a “manual gearbox”?
The Trello-for-AI-Coding Vision
Picture this:
- Project Capture: The Trello-like system “swallows” your whole project—files, structure, dependencies. No more “but it works locally.”
- Micro Changes: Every AI request introduces the smallest possible change. Forget “build the whole component.” Instead: “Add email validation,” “Create a User interface,” “Add error handling.”
- AI-Powered Decomposition: Given a new feature, AI generates a step-by-step breakdown—each micro-task is a Trello card, each card is a direct AI prompt.
- Automated Testing: Once a micro-task is completed, the card moves to a “Testing” column. Code changes are committed, your dev environment or Vercel auto-detects the update, and the server restarts to show the result.
- Automatic Feedback and Iterations: If something fails, the logs are attached to the card and it’s sent back for rework. Next time, the AI has the full error context and previous attempts, so it gets smarter.
Why This Model Works
- Control: You orchestrate the process instead of letting the AI go fully solo. You set the rules, the model works in your framework.
- Transparency: Every change is visible and logical—no black boxes, just a chain of understandable steps.
- Resilience: A bug in one micro-task won’t break everything. The system “learns,” improving on every iteration.
- Predictability: Smaller, sequential changes mean more reliable, less surprising outcomes.
A Practical Stack
It’s surprisingly doable with modern tools:
- Frontend/Backend: Next.js (leveraging parallel and interceptor routes)
- Source Control: GitHub API
- Deployment: Vercel (handles builds, previews, and production deployments)
- CI/CD: Automated via GitHub–Vercel integration—failed builds are caught early and flagged with notifications.
- Branching: Parallel feature development and safer merges become default.
The Philosophy
Isn’t it a bit weird—giving manual control to an automated process? Maybe. But even Teslas have steering wheels and pedals. Sometimes, control is more important than full autonomy.
What’s Next?
I’m planning to build a starter kit for this concept and release it as open source. Follow updates and real-time experiments in my Telegram channel.
How do you see the future of AI-driven development? Should we aim for more autonomy, or keep strong human oversight?
These are my “thinking out loud” notes on where AI and dev workflows could be heading. Maybe it’s controversial, but this kind of experimentation is what keeps our industry moving forward.
Top comments (0)