Dev.to AI 🤖 Ai 👁 0 📖 1 min read

A One-Weekend AI Build Stack (and the day I finally made peace with IAM)

I spent this weekend finally joining the vibe coding revolution: not by writing much code — but building the thing that writes it. The shift: I stopped treating my AI tools as a smarter autocomplete and started treating

A One-Weekend AI Build Stack (and the day I finally made peace with IAM)

I spent this weekend finally joining the vibe coding revolution: not by writing much code — but building the thing that writes it.

The shift: I stopped treating my AI tools as a smarter autocomplete and started treating them as a small team that hands work to each other. One prompt in plain English, and the "how" runs itself.

The Weekend #1 stack

Here's the actual wiring from two sessions (~10 hours, 158 shell runs, 12 tools stitched together):

  • You → Claude Code (orchestrator). Plans, runs the shell, writes files, and drives a real browser to verify its own work.
  • Local agent mesh. A local Qwen 2.5 model via Ollama, Cline inside VS Code, and a GitHub MCP server bridging the repo over Docker.
  • Code & cloud. A Vue web app on a staging environment, backed by AWS Amplify + DynamoDB (and, yes, IAM).
  • Data platform. A Python pipeline → DuckDB → Metabase dashboards.
  • Cross-cutting. An always-on monitor + architecture-hygiene dashboard (LaunchAgent), and persistent memory files so context carries from one session to the next — I'm not re-explaining myself every morning.

The funny part: IAM

I have always hated IAM. Users, roles, policies, access keys — historically my cue to throw my hands up and quietly give up on whatever I was building.

This weekend I ran amplify pull and the whole thing — user, profile, policies, access keys — came up correctly in minutes. Even the scary "owners may reassign ownership" warning didn't send me running. After years of that being my wall, it just… wasn't. A genuine 0→1 for me.

That's the real story of "productivity" here. It's not that I type faster. It's that the tasks I used to avoid — the ones that quietly killed projects before they started — are now approachable. The ceiling moved because the floor did.

Up next

Autonomous agents that build and self-improve on these projects, so the loop closes without me holding every thread. Curious, and a little cautious, about what that changes.

What's the task you've always avoided that AI finally made approachable?

📰 Read the original article on Dev.to AI

Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.