Watch the video on YouTube.

I took Anthropic's SDLC playbook and tried to build my own CRM. One that I can actually use, not just agent slop.

This is what happened when I put the process to work: intent, design, planning, building, testing, deployment and maintenance. What helped, where I got lost, and what I changed along the way.

The experiment also sent me into visual planning with plan-html, feedback loops with bubble-fix, and a separate comparison of Codex, Claude Code and DeepSeek on Pi.

Read the full breakdown in the blog post.

Want to get more out of AI in engineering?Want to build something?Building something, or want more from AI? See how →

More posts

Stop Approving Plans You Didn't Read

How many AI plans have you scrolled past and approved this week without really reading them? As coding shifts from prompts to loops, the plan becomes the last cheap moment to steer the agent. So I built plan-html: an interactive plan built for humans, not agents.

Read

Protect Your Context. Kill MCP. Use Agent Tools.

MCP shoves full API schemas into your agent's context on every interaction, burning thousands of tokens on documentation your agent doesn't need. Agent tools load on demand. 300 tokens vs. 3,000+.

Read

AI Orchestration - Prove - What the AI team Delivered

Read

AI Orchestration - Deploy: Agents in Action

Read

AI Orchestration - Build: Team of AI Agents from Scratch

Read