Hacker News: zmj

New comment by zmj in "Ask HN: How is AI-assisted coding going for you professionally?"

zmj — Sun, 15 Mar 2026 20:51:44 +0000

It's great. I'd guess 80-90% of my code is produced in Copilot CLI sessions since the beginning of the year. Copilot CLI is worse than Claude Code, but not by a huge amount. This is mostly working in established 100k+ LOC codebases in C# and TypeScript, with a couple greenfield new projects. I have to write more code by hand in the greenfield projects at their formative stage; LLMs do better following conventions in an existing codebase than being consistent in a new one.

Important things I've figured out along the way:

1. Enable the agent to debug and iterate. Whatever you'd do to test and verify after you write your first pass at an implementation, figure out a way for an agent to do it too. For example: every API call is instrumented with OpenTelemetry, and the agent has a local collector to query.

2. Make scripts or skills to increase the reliability of fallible multi-step processes that need to be repeated often. For example: getting an oauth token to call some api with the appropriate user scopes for the task.

3. Continually revise your AGENTS.md. I'll often end a coding session by asking the agent whether there's anything from this session that should be captured there. That adds more than it removes, so every few days I'll compact it by having an agent reword the important stuff for conciseness and get rid anything obvious from implementation.

New comment by zmj in "President Trump bans Anthropic from use in government systems"

zmj — Fri, 27 Feb 2026 22:06:13 +0000

This is the happy ending.

New comment by zmj in "Claws are now a new layer on top of LLM agents"

zmj — Sat, 21 Feb 2026 23:25:06 +0000

I also like the callback - not sure if it's intentional - to Stross's "Lobsters" (short story that turned into the novel Accelerando).

New comment by zmj in "Cloudflare outage on February 20, 2026"

zmj — Sat, 21 Feb 2026 22:23:46 +0000

Testing the "whole system" for a mature enterprise product is quite difficult. The combinatorial explosion of account configurations and feature usage becomes intractable on two levels: engineers can't anticipate every scenario they need their tests to cover (because the product is too big understand the whole of), and even if comprehensive testing was possible - it would be impractical on some combination of time, flakiness, and cost.

New comment by zmj in "Infrastructure decisions I endorse or regret after 4 years at a startup (2024)"

zmj — Fri, 20 Feb 2026 16:41:44 +0000

Separate! You lose the flexibility to move logic between the application and the database when the database is its own API.

New comment by zmj in "Ask HN: Why is my Claude experience so bad? What am I doing wrong?"

zmj — Sun, 15 Feb 2026 21:53:33 +0000

Try this:

* have Claude produce wireframes of the screens you want. Iterate on those and save them as images.

* then develop. Make sure Claude has the ability to run the app, interact with controls, and take screenshots.

* loop autonomously until the app looks like the wireframes.

Feedback loops are required. Only very simple problems get one-shot.

New comment by zmj in "Beyond agentic coding"

zmj — Sun, 08 Feb 2026 14:55:27 +0000

I like this thought. Scaling review is definitely a bottleneck (for those of us who are still reading the code), and spending some tokens to make it easier seems worthwhile.

New comment by zmj in "Claude Code is your customer"

zmj — Sat, 31 Jan 2026 21:39:45 +0000

Yesterday I had it using an internal library without documentation or source code. LSP integration wasn't working. It didn't have decompilation tools or the ability to download them.

I came back to my terminal to find it had written its own tool to decompile the assembly, and successfully completed the task using that info.

New comment by zmj in "AI’s impact on engineering jobs may be different than expected"

zmj — Fri, 30 Jan 2026 02:09:28 +0000

Paying money to abstract over lower level concerns is civilization.

New comment by zmj in "How I estimate work"

zmj — Sat, 24 Jan 2026 22:53:42 +0000

I was prepared to disagree with the thesis that estimation is impossible. I've had a decent record at predicting a project timeline that actually tracked with the actual development. I agree with the idea that most of the work is unknown, but it's bounded uncertainty: you can still assert "this blank space on the map is big enough to hold a wyvern, but not an adult dragon" and plan accordingly.

But the author's assessment of the role that estimates play in an organization also rings true. I've seen teams compare their estimates against their capacity, report that they can't do all this work; priorities and expected timelines don't change. Teams find a way to deliver through some combination of cutting scope or cutting corners.

The results are consistent with the author's estimation process - what's delivered is sized to fit the deadline. A better thesis might have been "estimates are useless"?

New comment by zmj in "Claude's new constitution"

zmj — Thu, 22 Jan 2026 05:32:28 +0000

Mid-level scissor statement?

New comment by zmj in "The assistant axis: situating and stabilizing the character of LLMs"

zmj — Tue, 20 Jan 2026 04:42:35 +0000

I wrote something fiction-ish about this dynamic last year: https://zmj.dev/author_assistant.html

New comment by zmj in "Dev-owned testing: Why it fails in practice and succeeds in theory"

zmj — Sat, 17 Jan 2026 03:07:35 +0000

At scale, every test is flaky.

New comment by zmj in "A deep dive on agent sandboxes"

zmj — Wed, 14 Jan 2026 01:23:18 +0000

devcontainers, devcontainers, devcontainers

New comment by zmj in "Don't fall into the anti-AI hype"

zmj — Mon, 12 Jan 2026 22:49:06 +0000

My experience with agents in larger / older codebases is that feedback loops are critical. They'll get it somewhere in the neighborhood of right on the first attempt; it's up to your prompt and tooling to guide them to improve it on correctness and quality. Basic checks: can the agent run the app, interact with it, and observe its state? If not, you probably won't get working code. Quality checks: by default, you'll get the same code quality as the code the agent reads while it's working; if your linters and prompts don't guide it towards your desired style, you won't get it.

To put that another way: one-shots attempts aren't where the win is in big codebases. Repeat iteration is, as long as your tooling steers it in the right direction.

New comment by zmj in "Start your meetings at 5 minutes past"

zmj — Sat, 10 Jan 2026 14:14:13 +0000

A friend of mine built something for this: https://howmuch.poiesic.com/

New comment by zmj in "Go away Python"

zmj — Tue, 30 Dec 2025 22:09:28 +0000

I wrote my first one of these today (making a tool for agents in a c# codebase). Pretty good experience, though AOT does still have some rough edges.

New comment by zmj in "Karpathy on Programming: “I've never felt this much behind”"

zmj — Tue, 30 Dec 2025 03:35:52 +0000

Yes-ish. It's worth keeping up with the rising tide of model capabilities, but it's not worth stressing over eliciting every last drop. Many of the specific techniques that add value today will be wasted effort with smarter models in a month or two.

New comment by zmj in "Ask HN: How are you sandboxing coding agents?"

zmj — Sat, 27 Dec 2025 15:05:24 +0000

devcontainers, without credentials to the git remote.

New comment by zmj in "Comptime – C# meta-programming with compile-time code generation and evaluation"

zmj — Thu, 25 Dec 2025 22:43:13 +0000

Those collections are more like copy-on-write than actual immutable. System.Collections.Frozen is the real thing.