GoKawiil
Tech News
A software engineer writing under the 'clanker' label describes insisting on fully reviewing code generated by AI coding tools before shipping it, rather than trusting it blindly. The author says accountability for production failures rests with the human whose name is on the pull request, not the AI tool that wrote the code.
FurtherAI engineering leadership shared internal metrics from building what they call an AI-native engineering org over the past nine months. The company reports a roughly 5x increase in code changes merged per engineer since January, with revert rates under 0.5%, driven by adding AI agents to coding, test writing, code review, CI, bug triage, and rule enforcement at each stage of the shipping process.
Cursor has released the Agent SDK, the same framework it uses internally to build managed agents like Bugbot, Security Reviewer, and Rollouts. Developers can define agents in code, starting from a template or using Cursor's documentation, then deploy them to Cursor's infrastructure with a single command. An internal Cursor team used the SDK to build 'kb-reviewer' in about a day, an agent that checks every pull request against the team's knowledge base of past lessons.
Anthropic engineer Addy Osmani described how the Claude Code team restructured code review after merged changes in their monorepo jumped from roughly 2,500 per day in May to over 30,000 per day in September. The team now runs changes through six sequential checks—pre-PR verification, AI review, security review, CI, risk-scaled human approval, and staged rollout with revert—rather than relying on a single reviewer. Merged pull requests on Claude Code repositories more than doubled between June and September while reverts dropped by over 40%.
A developer has released Farol, a free open-source CLI and GitHub Action that analyzes a pull request after AI code review and generates a risk score (0-100) along with a verdict of fast track, quick review, or careful review. It maps which pages, APIs, jobs and scripts are reached by the changed code using static analysis rather than AI-generated guesses, aiming to show human approvers exactly where risk lives in a diff.
A Hacker News 'Ask HN' post cites a LinkedIn message from a Meta engineer, Ali Hamie, who says he increasingly ships AI-written code without fully understanding every line. Hamie describes feeling uneasy that reviewing all AI-generated code in detail would be too slow, leaving him with only a high-level grasp of systems he builds. The HN post asks other engineers how they are coping with this shift.
A personal website post by Will Larson lists his published books, including 'An Elegant Puzzle,' 'Staff Engineer,' 'The Engineering Executive's Primer,' and 'Crafting Engineering Strategy,' alongside ways readers can reach out to him. The post does not contain substantive technical detail about the referenced 'small-change' code review tool.
Describing what you want and accepting the code that comes back is genuinely productive for some work and quietly expensive for other work. The dividing line is fairly predictable.
A blog post recounts a framework shared by an engineer named Emily for working with AI coding tools, dubbed the 'Brain Sandwich': think through the problem yourself first, delegate implementation to AI, then personally review and refine the output before sending it for review. The approach is offered as a counter to engineers fully outsourcing tasks to AI, a practice some reportedly call a 'Meat Proxy'.
Anthropic's Claude Code team published an account of how it adapted code review practices as AI-generated code volume surged. Merged pull requests to the Claude Code repository more than doubled between June and September, while the share of PRs later reverted fell by over 40%. The team says it shifted most initial code reading to automated Claude-based checks that run continuously, reserving human review for higher-level judgment calls.
The PostgreSQL Global Development Group has delayed PostgreSQL 19 from its planned late September/October launch to around November 19, cutting between eight and ten planned features. EDB's Peter Eisentraut, a core PostgreSQL committer, said many of the dropped features had architectural flaws and that newly adopted LLM-assisted code review surfaced a long list of smaller edge-case defects that developers couldn't fix in time for beta.
A new analysis argues that although AI tools now let developers write, refactor, and document code far faster than two years ago, overall software shipping speed has barely improved. It attributes this gap to software development lifecycles still built around sprints, manual quality gates, and review handoffs designed for an era when writing code itself was the slowest step.