Code Review Is Hard. For People, and for AI Too
Code review is still hard
16 POSTS
Ai 1
◆ Improving number-flow · Part 4How Far Will Agents Fix number-flow?
I compared number-flow as fixed by five coding agents against my fork v0.2.1 once more. Performance reports changed with the font, the code went through a review with the authors hidden, and a follow-up experiment ended with keeping the mask default.
◆ Swapping the Model Behind a Coding Agent · Part 4Wrapping Up the Coding Agent Model Swap Experiment
Looking back on 1,164 runs and the process of building an internal benchmark. Cost per task, the Claude Code run environment, what I had hoped for from open-weight models, and the conditions I would change next time
◆ Swapping the Model Behind a Coding Agent · Part 3What Happens When You Swap the Model in Claude Code for GLM 5.3
I solved 36 internal tasks three times each with Claude Sonnet 5.5 and GLM 5.3. Comparing resolved rate, limit stops, cost, and patch defects across 216 runs, looking at how the conditions differ from public benchmarks, and deciding whether to swap the model
◆ Swapping the Model Behind a Coding Agent · Part 2Building 36 SWE-bench Tasks from 20 Internal PRs
Turning merged PRs from an internal WebSocket package into coding-agent evaluation tasks, validating the grader, and isolating personal instructions from the Claude Code run environment
◆ Swapping the Model Behind a Coding Agent · Part 1How to Compare the Models Behind a Coding Agent with SWE-bench
How do you measure the model performance of a coding agent?
Where Frontend Came From, and Where It Goes After Agents
Why did the frontend stack pile up in layers, and why did rendering move back to the server? This essay weighs whether the same stack survives once agents write the code, and asks whether, if it does, the value of the developers who know it survives too.
◆ Judgment in the AI Era · Part 3When Was the Last Time You Read Code Seriously?
As code review and interviews stop reading code closely, there are fewer chances to build the judgment that matters more in the AI era. This essay checks the evidence from studies such as METR and GitClear, and considers why the need to learn judgment went away.
◆ Judgment in the AI Era · Part 1If the Code Meets Spec and the Bugs Get Fixed, Does It Matter That You Can't Read It?
If the spec is satisfied and the bugs get fixed, do you still need to be able to read the code? This essay weighs the argument that you do not and the weak points in its premises, then lays out what you need to understand when you stop reading code directly and what juniors should aim for.
◆ Judgment in the AI Era · Part 2If AI Erases the Boundaries Between Planning, Development, and Design, What Remains?
With AI tools, people can now produce the deliverables of other roles, and so we hear that the boundaries between job titles are collapsing. Separate the ability to make something from the ability to know whether it is right, and where do the boundaries get redrawn?
Bun rewrite Exposed Something: OSS Could Only Stop External AI
Bun moved roughly 960,000 lines from Zig to Rust in six days using Claude Code. This post works through what that means for OSS governance and resource asymmetry.
AI Only Amplifies Me to the Level I Can See
I'm definitely coding faster, but why are my codebase and skills staying the same? Examining the gap between perceived and proven benefits.
In the Age of AI-Generated Code, Where is Frontend Engineering Headed?
How does the work of a frontend developer change once AI coding tools arrive? Thoughts centered on writing specs, code review, cognitive debt, decision speed, and the role of senior engineers.
Complete Guide to Core Coding Agent Concepts
Complete overview of core coding agent concepts: Rules, Commands, MCP, Sub-agents, Hooks, Skills, and Plugins
WebGPU Is Finally Available in Every Browser
In July 2025, the browser GPU API changed generations for the first time in 14 years
In an Era Where AI Writes Code, What Should I Do as a Frontend Developer?
I don't know what the future holds, but shouldn't we just keep developing diligently?