惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

博客园 - 【当耐特】
N
Netflix TechBlog - Medium
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
雷峰网
雷峰网
MongoDB | Blog
MongoDB | Blog
有赞技术团队
有赞技术团队
Engineering at Meta
Engineering at Meta
M
MIT News - Artificial intelligence
Google DeepMind News
Google DeepMind News
罗磊的独立博客
Hugging Face - Blog
Hugging Face - Blog
WordPress大学
WordPress大学
T
Tailwind CSS Blog
小众软件
小众软件
J
Java Code Geeks
人人都是产品经理
人人都是产品经理
博客园_首页
MyScale Blog
MyScale Blog
博客园 - 聂微东
V
Visual Studio Blog
The Cloudflare Blog
月光博客
月光博客
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
U
Unit 42

Martin Alderson

Have the frontier labs mixed up AI safety and security? What GLM-5.3 Flash running on Chinese hardware actually means The summer of open weights How I think about reducing AI costs Watch out for cache read costs I'm (mostly) picking models on speed now, not intelligence The first known runaway AI agent - or a very bad marketing stunt? Winners and losers in the coming AI margin collapse (part 2) GLM 5.2 and the coming AI margin collapse (part 1) Expert-aware quantisation: near-Q4 quality at near-Q2 size? A brief history of KV cache compression developments xAI is looking more like a datacentre REIT than a frontier lab Is datacentre sovereignty really that important? I went on the Built for Turbulence podcast What's going on with Gemini? Managed agents are the new Lambda Open weights are quietly closing up - and that's a problem 29th August 2026: a scenario Figma's woes compound with Claude Design A little tool to visualise MoE expert routing Has Mythos just broken the deal that kept the internet safe? What next for the compute crunch? Telnyx, LiteLLM and Axios: the supply chain crisis Using agents and Wine to move off Windows Why Claude's new 1M context length is a big deal How to use the Qwen 3.5 LLMs to OCR documents No, it doesn't cost Anthropic $5k per Claude Code user Is the AI Compute Crunch Here? Why on-device agentic AI can't keep up Using OpenCode in CI/CD for AI pull request reviews
Reducing codebase cognitive debt through... quizzes?
Martin Alderson · 2026-08-31 · via Martin Alderson

Like many, when I'm building with coding agents, one of the "worst" problems is the codebase feels like it is often evolving faster than I can keep up with it. This cognitive debt builds up quietly - nothing is broken, but you gradually have this feeling of not being totally on top of what is going on.

One remarkably effective technique I stumbled on is asking the agent to quiz you on the code base. A simple prompt like:

hi, please quiz me about this codebase.

i want 5 questions of increasing difficulty - then explain to me at the end what i got wrong about my understanding. use the askuserquestiontool

Produces a simple quiz you can answer.

Claude Code quizzing the user about the llama.cpp codebase

The best bit is chatting to the agent about the ones you got wrong. It has the context of what you thought versus what the code actually does, so it can explain the gap far better than it could if you'd just asked it to describe that part of the codebase itself.

You can of course use this for non code tasks too, e.g. complex spreadsheets or document sets.

I've had a lot of success with this - if I'm working on a complex refactor with a difficult plan, asking the agent to quiz me about the plan or part of the project in question before starting seems to make me catch issues more than trying to read pages upon pages of Claudish.

I've tended to find my perception of the cognitive debt is (usually!) far more than the reality, and this helps nullify the sinking feeling that the codebase/project is running far ahead of my understanding.

You could go a step further and enforce you passing a quiz about a PR too before you allow the agent to create one.

Sometimes with agents the best options are the simplest.