@fxnn@hachyderm.io

Felix Neumann @fxnn@hachyderm.io · Mar 19, 2026

Replying to @sebastian@phpc.social

@sebastian@phpc.social Very important problem indeed. What I couldn't read from your article is this: did you try to use AI to verify the implementation? I can think of multiple ways how tools like Claude Code etc. can help you gain confidence in the implementation -- or to find flaws. (1) Ask it for a written proof (whatever kind of "proof" it will generate, maybe you can work with it). (2) Ask it for a written walkthrough. @simon@fedi.simonwillison.net built a tool for that: https://simonwillison.net/guides/agentic-engineering-patterns/linear-walkthroughs/ (3) Go interactive, enter a conversation with your agent, guide it through your algorithm, and let you show step by step how the implementation matches it. This kind of problem will be very important in future, so I'm looking forward to any insights here!

0

Felix Neumann @fxnn@hachyderm.io · Mar 07, 2026

It's fascinating to see how many experienced developers use Claude Code and similar tools.

AI does the same tasks as those devs, in a fraction of the time, and often enough, the result is of better quality.

Sure, all guided by humans, beginning with human thought, in processes designed by humans, with a human being fully responsible for the outcome.

But they do it. They can run dozens of minutes autonomously, so that we can easily run multiple agents in parallel on different tasks.

Try it out. Reach that point. It's mind boggling.

1

0

Felix Neumann @fxnn@hachyderm.io · Mar 07, 2026

Replying to @paulshryock@phpc.social

@paulshryock What makes you think that it's zero substance? I see experienced developers every day applying tools like Claude Code to do the same task the developer would have done, in a fraction of the time, and often even in better quality. I wonder what's "all hype" in such a technology.

0

Felix Neumann @fxnn@hachyderm.io · Feb 28, 2026

If there's an actor interested in the common good with the resources to develop advanced AI models: then it's a state.

Governments must invest in responsible AI models.

(Let's dub a German transformer "GerT".)

0

Felix Neumann @fxnn@hachyderm.io · Feb 27, 2026

Replying to @oli@social.tchncs.de

@oli@social.tchncs.de Jau, und mit dem Rest weiter unten gehe ich auch völlig mit -- sobald der Mensch das Bottleneck wird, klingelt der Burnout an der Tür... @johnnythan@tuebingen.network @javahippie@mainz.social

0

Felix Neumann @fxnn@hachyderm.io · Feb 27, 2026

Replying to @oli@social.tchncs.de

@oli@social.tchncs.de Tatsächlich war das bei schwächeren Modellen so. Bei bspw. Claude Code besteht der Trick darin, eine gute Feedbackschleife zu definieren. Beispiele: "Hier ist die OpenAPI-Spec. Erweitere die Anwendung so, dass sie die Spec erfüllt." Oder "hier sind die Zugangsdaten zu einem Kubernetes-Cluster. Baue ein Helm-Chart und installiere es auf dem Cluster, sodass dann dort XYZ funktioniert". Mithilfe so einer Feedbackschleife kann der Agent längere Zeit autonom arbeiten und das gewünschte Ergebnis meist wirklich erreichen. @johnnythan@tuebingen.network @javahippie@mainz.social

0

1

0

Felix Neumann @fxnn@hachyderm.io · Dec 27, 2025

Replying to @webbop@tech.lgbt

@webbop@tech.lgbt Actually, that was not the point. @simonw@bird.makeup wrote a technical analysis [1], and the task the agents had was simply this: "Do random acts of kindness." The AI agents themselves came up with the idea of sending e-mails, and with a list of recipients, by themselves. So the actual problem here is giving AI agents so much power with so little supervision. And btw. Anthropic is not affiliated, it was a non-profit which did this, they simply used Anthropics APIs. [1] https://simonwillison.net/2025/Dec/26/slop-acts-of-kindness/ @nixCraft@mastodon.social

1

0

Felix Neumann @fxnn@hachyderm.io · Nov 06, 2025

Replying to @organicmaps@fosstodon.org

@organicmaps@fosstodon.org For live traffic information, there's quite a number of resources for individual countries or cities (https://github.com/graphhopper/open-traffic-collection) Combined with some funding-based opt-in open source life data collection and aggregation, it could be quite useful. But that would be a big initiative and its own non-profit organization...

0

2

0

Felix Neumann @fxnn@hachyderm.io · Jul 26, 2025

Replying to @juandesant@mathstodon.xyz

@juandesant@mathstodon.xyz Oh it does better than a human. LLMs are prediction machines. They can predict good continuations of any kind of document or chat, for a much broader range of topics and languages than any human could. Hallucinations are just predictions that haven't been as good as others. @climagic@mastodon.social

0

Felix Neumann @fxnn@hachyderm.io · Jun 03, 2025

Replying to @jcoglan@mastodon.social

@jcoglan

By far no dumb question. Choosing the right step size is utterly important. With older / smaller models, you'll need to take pretty small steps to get useful results, like the "add this param" example you quoted. By far too small to be useful, if you ask me.

Then there's the people who came up with a really involved workflow, composed of multiple tools and prompts. I believe this one can be really powerful, but it just cries for automation and I don't want to feel like a trained monkey. But definitely have a look, this one is really interesting in terms of the prompts used: https://harper.blog/2025/02/16/my-llm-codegen-workflow-atm/

I'm mostly going for simpler prompts, but that's only possible with advanced models like Sonnet 3.7, o4-mini or (the one I'm using) Gemini 2.5 Pro. Real-world example from my history:

Add a separate mode to the application, in which it does not print the downloaded e-mails to stdout, but instead starts an HTTP server with a RESTful API.
We will add the actual API implementation later, some dummy plain-text response is enough for now.

This added the necessary commandline switches and the HTTP server boilerplate.

The next on was this:

Now we want `startHttpServer` to receive the same input as `processEmails`.
 It shall also fetch e-mails, and then provide the aggregated stories from all e-mails under a `/stories` REST endpoint. As usual, start with proper unit tests.

And there I had my REST API.

Typically, I switch between such requirements-oriented prompts, and then more lower-level improvements/refactorings, making sure that the code quality and architecture is right, like this one:

In `main` func, handle `fetchAndSummarizeEmails` only once, for both `printEmails` and `startHttpServer`, because both need it in the same way. Also, handle errors directly, no need to pass them on to e.g. `startHttpServer`.

@alanpaxton

0

Felix Neumann @fxnn@hachyderm.io · Jun 03, 2025

Replying to @jcoglan@mastodon.social

@jcoglan I can relate. While I get that people want to earn money with their blogs, I believe that there must be a better way than to feed Substack or Medium's pockets. Anyways, the one good free blog on the general LLM topic (I'm reading regularly) is that from @simon, https://simonwillison.net Also I got recommended a blog post from https://harper.blog/posts/ which seems to feature some interesting AI coding posts lately. Nothing more specific at the moment, unfortunately. @alanpaxton

1

0

Felix Neumann @fxnn@hachyderm.io · Jun 03, 2025

Replying to @jcoglan@mastodon.social

@jcoglan Well, other people made different experiences. I, with my admittedly tiny toy project https://github.com/fxnn/news I'm currently working on, using Aider and Gemini 2.5 Pro. I'm curious to explore (soon) how AI does with larger legacy code. Then there's Aider, the tool which I'm using for AI augmented coding, I believe it usually has 80-90% of a release's code written by AI: https://github.com/Aider-AI/aider Then recently one of Cloudflare's repos was discussed, whoch they developed using Claude: https://github.com/cloudflare/workers-oauth-provider @alanpaxton @akahn

0

1

0

Felix Neumann @fxnn@hachyderm.io · Jun 03, 2025

Replying to @jcoglan@mastodon.social

@jcoglan Yes, the steps we make shouldn't be too coarse, and also AI augmented coding is still somewhat adventurous. One of the important things it needs is a good "system prompt", some rules which tell it e.g. to never fix a test by deleting it. I can recommend @kentbeck's recent blog posts on that matter, paywalled, e.g. https://open.substack.com/pub/tidyfirst/p/persistent-prompting @alanpaxton @akahn

0

1

0

Felix Neumann @fxnn@hachyderm.io · Jun 03, 2025

Replying to @alanpaxton@urbanists.social

@alanpaxton From my observation, as of today, AI can write relatively simple source units errorless in one shot and extend existing source units with smaller variations. It can often fix issues like failing tests or compiler errors on its own. Hence, in a typical dialogue, the human can focus on code review, evolving the architecture and keeping the direction towards the requirements. Tests are crucial, as they allow AI to automatically detect and fix errors, which saves lots of time and nerves on the human side, and reduces the effects of review oversight. Refactoring is possible, but I'm awaiting the integration of well-known automated refactorings as AI assistant tools, which would increase the efficiency and effectiveness. Currently, I would say that the gain of AI augmented coding lies in speed (at the expense of quality) and, to some limited extend, automation. But as I said, given the current exponential growth, this will improve and it's possible that there will be unforseen use cases. @jcoglan @akahn

0

1

0

Felix Neumann @fxnn@hachyderm.io · Jun 03, 2025

Replying to @alanpaxton@urbanists.social

@alanpaxton Very good and important questions. We're talking about a technology that's growing exponentially: whatever the answer is today, it will be different next year. All we can do today is experiment with all the new tools they release, in order to understand how they work and learn what they can do for us. @jcoglan @akahn

0

16

0

Felix Neumann @fxnn@hachyderm.io · Jun 02, 2025

Replying to @jcoglan@mastodon.social

@jcoglan Actually depends IMO. Writing production code first, having AI write unit tests afterwards? Terrible idea. Using AI to generate missing unit tests in a years old legacy system? Better idea. Doing AI coding in TDD style? Great idea. Turns out that writing the test first really improves AIs ability to come up with good production code. TDD fans won't be surprised. @alanpaxton @akahn

3

1

0

Felix Neumann @fxnn@hachyderm.io · Dec 26, 2024

Replying to @amyipdev@transfem.social

@amyipdev@transfem.social @lynnesbian@fedi.lynnesbian.space Why? If the important safety feature would be a not-equals-zero check, then 258 axis would be stored as two axis, and therefore be safe (as two does not equal zero).

1

0

Felix Neumann

Posts

Remote instance