In partnership with
FROM OUR PARTNER
100+ coding prompts top engineers use to ship 5X faster
Claude Code, Codex, and Cursor are on every engineer's tech stack. Most still treat them like a search bar and get average output back. Top engineers work from a system instead. These 100+ prompts are that system.
Sign up for The Code and get:
100+ ready-to-use prompts built on real Claude Code, Codex, and Cursor workflows, so you ship in minutes instead of hours
The Code newsletter (5 min daily) to keep sharpening your AI-coding edge after the prompts land
TODAY AI BRIEF
3 things that show how AI is changing code review
1 42% of developers now let AI write at least half of their code.
That is up from 12% a year earlier, according to BairesDev's Q3 2026 Dev Barometer. The survey covers 705 developers in more than 60 countries, and BairesDev sells software development services, so read its figures as a vendor's survey. Read Techzine's summary.
2 Developers save more time coding, then spend it reviewing.
Developers report 13 hours a week saved on coding, double the 7 hours from a year earlier. Yet 67% review more AI-generated code than last year, 52% spend more time debugging problems AI introduced, and 78% of the CTOs surveyed say they spend more on review, QA and validation. Read the full article.
3 Reviewers of agent-written pull requests mostly comment on documentation and refactoring.
An academic study of 3,177 agent-authored pull requests and 19,450 review comments found that the dominant themes are documentation gaps, refactoring needs and styling. Testing and security also come up, but less often. The authors say AI agents speed up production while gaps still call for targeted human review. Read the arXiv paper.
Bonus reads: Addy Osmani's January post, Code Review in the Age of AI, sets out a practical checklist for authors and reviewers alike.
AI FORWORD
AI did not kill code review. It made the burden of proof explicit.
The shift. AI now writes a large share of the code many teams ship. BairesDev's survey found that 42% of developers let AI write at least half of it, and Techzine reports the same data as a sign of a new bottleneck. Read the summary.
The cost. Coding gets faster, and the work moves elsewhere. Techzine cites a separate study finding that experienced developers took 19% longer on tasks when using AI, and another survey finding 17% lower debugging skill test scores among AI-assisted developers. Both are cited secondhand in that article, so treat them as leads rather than settled results. Check Techzine's sources.
The evidence gap. Osmani cites figures such as logic errors being about 75% more common in AI-generated code and about 45% of it containing security flaws. He gives no primary study for them in the post, so verify them before you quote them. His full argument is clearer than the numbers he quotes.
The reframe. Osmani puts the job in one sentence: a computer can never be held accountable, and that is your job as the human in the loop. Review was never only about catching bugs. It is how a team decides who answers for code that ships, and that does not change when the author is a model.
PRESENTED BY :
7 Stocks You May Never Need to Sell
Buying a stock is easy. Finding one you may never need to sell is harder.
This free report reveals seven blue-chip companies selected for financial strength, durable demand and long-term potential to hold up when markets turn ugly.
SOCIAL SNAPSHOT
The numbers people are sharing about AI code review
13 hours, not 13 free. BairesDev's CEO, Darren Shimkus, said: "Developers nearly doubled the time AI saves them in coding, to 13 hours a week. Not one of those hours came back." Read the quote in context.
A 13,000-line pull request. Osmani says the OCaml maintainers rejected a 13,000-line AI-generated pull request because nobody had the bandwidth to review it. Read his post.
The study's sample. The arXiv paper analyzed 3,177 agent-authored pull requests drawn from the AIDev dataset. Its LLM annotator matched human labels on 78.63% of comments. Read the paper.
AI reviewers are split. Osmani notes that some teams report AI review catching 95% or more of bugs, while others dismiss its comments as noise. He suggests a well-configured AI reviewer can catch 70 to 80% of low-hanging issues, a figure he does not source. Read his section on AI review.
The survey's caveat. BairesDev is a software services firm and its Barometer is a company survey of 705 developers. Read the percentages as directional. Check the survey context
PROMPT STATION
Review a pull request like a skeptical senior engineer
AI-written code often looks clean and still hides a missing test or an unchecked input. This prompt works in Claude, ChatGPT or Gemini and turns a diff into a short list of questions for the author. Use it before you approve, not after you merge.
Here is a pull request description and its diff: [PASTE THE PR DESCRIPTION, THEN THE DIFF]. The author used an AI tool to write part of it.
Act as a skeptical senior engineer with 20 minutes. First, check the description for four things: the intent in one or two sentences, proof the change works, a risk tier, and which parts an AI wrote. List what is missing. Second, flag any change that touches authentication, payments, secrets or untrusted input, and say what a human threat-model review should check.
Finish with the three questions I should send back to the author, ranked by how much they would change the merge decision.Tip: Swap the bracket for a real pull request, such as a login change or a checkout update. For a sharper result, add the test output, so the reviewer can see which claims the tests actually back up.