Return to site

🎯🤖 AI & PR: WHEN DOES IT REALLY MAKE SENSE?

🤝 Resharing Peter Lawrey’s insights to clarify: AI can accelerate PRs, but engineering wisdom still does the hard work.

November 21, 2026

🔸 TL;DR

AI is great for PR when the change is simple, low‑risk, and easy to reverse. The harder part: spotting wrong assumptions, over‑complexity, or misaligned solutions; still needs human engineering judgment.

1️⃣ [I use AI for much of the heavy lifting in research, trial solutions, mock systems, and development admin.]

AI fits best for PRs that package research, experiments, and boilerplate into something reviewable quickly.

#mock

2️⃣ [However, when AI produces a PR, I don't feel the need to hand-rewrite much of it; it's almost always a really simple change.]

Great use case: small, scoped PRs (typos, minor refactors, tiny features) where intent and impact are crystal clear.

#typo

3️⃣ [Gen AI is better than a developer for technology I have no expertise in, on a cost basis (e.g., per unit time).]

AI can draft PRs in unfamiliar stacks, but treat them like prototypes, not production code, especially for non‑trivial logic.

#unfamiliarTech

4️⃣ [But the result for anything non-trivial is usually a good start, usually works, but it is often a mess, or isn't fit for purpose.]

For complex PRs, AI is a starter, not a finisher; humans must simplify, align with requirements, and clean up the mess.

#complexityToFineTune

5️⃣ [The difficult part of engineering is frequently recognising that the approach is unnecessarily complicated, makes the wrong assumptions, or solves the wrong problem.]

PRs for critical paths, architecture, or subtle bugs need human judgment to catch wrong assumptions and over‑engineering.

#tradeoff

🔸 TAKEAWAYS

▪️ Use AI to speed up small, clear PRs, not to own critical architecture.

▪️ Let AI handle boilerplate; humans handle ambiguity and trade‑offs.

▪️ AI can kickstart work in unfamiliar tech, but don't deploy it blindly.

▪️ The real skill is recognising when AI’s solution is solving the wrong problem.

▪️ AI is a teammate for productivity, not a replacement for engineering thinking.

#CriticalSystems #EngineeringJudgment #ProblemSolving #TechLeadership #AIAndHumans

Go further with Java certification:

Java👇

https://bit.ly/javaOCP

Spring👇

https://bit.ly/2v7222

SpringBook👇

https://bit.ly/springtify

Java Fullstack Book👇

https://bit.ly/jfstack27