End to End AI Development — practical analysis
We’ve gone fully AI-native on purpose: AI writes code, AI reviews code, AI helps debug. Some people call that overuse. Fair.
This series is a practical stress-test of the usual objections — not a vibe-coding manifesto, and not “AI will replace judgment.”
The bet: if you steer with architecture, separate concerns for review, and keep humans on the hard paths, end-to-end AI is a workflow you can defend. If those pieces are missing, the critics are right.
Drafted with AI. Reviewed and edited by Ed Henderson.
How to read this
Each post: the common claim, what I’ve seen in practice, where the objection still bites, and what I’d change in the workflow. Start with #1 and #2. The rest follow.
Roadmap
- 01Understanding decay
- 02Review theater (AI reviewing AI)
- 03Confident wrongness — later
- 04Architecture drift — later
- 05Test theater — later
- 06Security / secrets — later
- 07Ops blindness — later
- 08Skill atrophy — later
- 09Ownership fog — later
- 10Context window lies — later
- 11Dependency / license sludge — later
- 12Team sync (design in chat history) — later
- 13Speed addiction — later
- 14Evaluation gap — later