Engineers are late to AI coding tools until they use models to map ERDs and bugs
GabGarrett · x · 2026-07-29
Why some strong engineers adopt AI later, then use it as a workflow multiplier
The post argues that highly capable engineers often became late AI adopters because early tools were worse than they were at coding and the hallucinations caused them to discount the value.
The key point is that they missed the second-order effect: even if models are weaker today, they can let you ship something faster now and make it much better later as models improve.
The author says smart adopters now use models as a steering layer rather than a replacement for understanding:
- ask the model to generate ERDs and sequence diagrams instead of reading code line by line
- use those outputs to spot bugs and validate data models
- keep enough architecture understanding to direct the model effectively
The thread's thesis is that AI becomes a superpower when the engineer can combine model output with strong systems intuition.
More from coding & agent
- Claude Code can build the same landing page twice, but only one version looks designed — PrajwalTomar_ · 2026-07-29
- Munin shows how MCP Apps keep human approval inside the chat UI — k_man9 · 2026-07-29
- Cline signs the open-weights letter and makes GLM 5.2 free — TheZachMueller · 2026-07-29
- A Weaviate extension now manages vector databases inside VS Code — CShorten30 · 2026-07-29
- ‘Prepare your personal vectorbooks’ turns note-taking into an AI meme — wavefnx · 2026-07-29
- A CLI tool now ranks multi-agent runs by cost and flags repeated context tokens — rrk059 · 2026-07-29