My AI workflow from zero to shipped — what I actually used and why
3 min read
I co-founded a project in 2025. Designer, PM, and first builder — all me. Here's the actual workflow, tool by tool, phase by phase. No fluff.
Research
Dumped user interview notes into Claude and GPT. Asked them to pull patterns, highlight contradictions, rank problems by frequency. Skipped the spreadsheet. Got to a problem statement in hours, not days.
Claude + ChatGPT → research synthesis
MVP build — March 2025
Designed in Figma, then built it myself in Cursor. I'm not a full engineer — but with AI filling the gaps, I could write and iterate on the frontend without waiting on anyone. Shipped a working MVP, then handed it to engineers for production.
The point wasn't to replace engineers. It was to stop the "design → waiting → feedback → waiting" loop.
Figma → Cursor → handoff
Growth analysis
No analyst on the team. So I exported Firebase data weekly and ran it through AI — retention, DAU, drop-off, activity patterns. Got a structured report every week. Made product decisions off that instead of gut feel.
Firebase export → AI report → product decision
Agentic ops — 2026
We set up an AI agent in our team Discord with multiple roles. A few things I actually used it for:
— As designer: asked it to change frontend copy or styles directly. No ticket.
— As PM: weekly competitor brief, user request summary — auto-generated.
— As remote team member: checked the agent's log to catch up on what happened while I was offline. Better than scrolling Slack.
OpenClaw agent → Discord → multi-role ops
The app hit #10 on the App Store and 80K users in 6 months. I don't think the AI tools caused that — but they let a small team move at a pace that wouldn't have been possible otherwise.
