HNHacker News
TopNewBestAskShowJobs

stefanhoelzl

136 karma · joined February 10, 2019

submissionscomments
stefanhoelzl··on Ask HN: What have you built privately with AI that you'd like to share?
actually with AI my private projects get to a usable state for the first time :)

started with CodeHydra to work more efficiently with multiple agents: https://codehydra.stho.net/

The second one is a App to share photos during an event: https://snapsync.stho.net/

stefanhoelzl··on Ask HN: Any Software Engineers here who enjoy their AI-native dev workflow?
Per-AI I was pretty picky about the right code „esthetics“ and had joy in finding better patterns and solutions to have understandable and maintainable code. This led to me being often slow and never finishing my side projects as soon as the boring parts started.

Since I code with Claude (using codehydra[1]), I have accepted that the code itself is much worse. But I found joy in working at a much higher architecture level. This led me to finish my side projects and being much faster.

It is very different, but I can find statisfaction in both. Since I do not think AI will go away, I fully went the AI way and rarely looking at any code anymore. But changing my workflows to work on architecture, design and most important verification strategies for the agent generated code.

[1] https://github.com/stefanhoelzl/codehydra

stefanhoelzl··on Ask HN: What do you do when your AI agents are working?
Working on multiple tasks/projects in parallel.

The crux is to know where your attention is needed and quickly get the full context of the task when you come back.

That’s why I built Codehydra https://github.com/stefanhoelzl/codehydra

stefanhoelzl··on Ask HN: How are you staying productive using LLMs?
For me the most important part is to do multiple jobs at once, so while the agent is doing anything I can work with another agent which needs input. For this to work, I must be able to have the complete context for a task in one place, to get a up to date when I switch back to a task. That’s the tool I built helping me to achieve this https://github.com/stefanhoelzl/codehydra
stefanhoelzl··on Ask HN: What are tools you have made for yourself since the advent of AI?
https://github.com/stefanhoelzl/codehydra

Allows me to efficiently work on multiple tasks in multiple repositories concurrently.

stefanhoelzl··on Ask HN: How to think in terms of parallel Claude agents
Thanks for giving it a try!

If you have any questions or problems just open a issue on github!

stefanhoelzl··on Ask HN: How to think in terms of parallel Claude agents
I had primarily the issue, that when working only on one task at a time, I had nothing to do, when my AI was thinking or implementing a plan.

Just launching multiple AI instances does not scale because you loose the overview an context of each task. Also you need git worktrees to isolate the agent from each others work.

My solution was a tool around VSCode, handling worktree creation and notifications for idle agents. A MCP which can interact with VSCode allows the agent to prepare any outcomes for preview or show a notification what to test or to do next. That way I cycle through agents answer the questions it has or review its work. I regularly work with ~5 parallel task (sometimes up to 10).

https://github.com/stefanhoelzl/codehydra

stefanhoelzl··on Ask HN: What is your dev set up like?
Until October 2026 I used mainly PyCharm.

Since then I do all my dev work with Claude Code. And currently pulling in more and more workflows into Codehydra including PRs and Issue management.

stefanhoelzl··on Ask HN: What are you working on? (April 2026)
a IDE to work with multiple AI agents in isolated workspaces. https://codehydra.dev / https://github.com/stefanhoelzl/codehydra

It is not only isolating with a git worktree but also all other context related to this task (previews, reviews) and allows the agent to interact with this context via notifications and so on.

stefanhoelzl··on Ask HN: How do you offload all coding to AI?
Do not wait for your agent, launch parallel tasks when your agent is working. Then it does not matter anymore how long your agent takes, as long as you always have at least one idle agent.

Codehydra (https://github.com/stefanhoelzl/codehydra) can help you with this workflow.

stefanhoelzl··on Ask HN: What Are You Working On? (March 2026)
Solving my issue with context switching when working with multiple coding agent sessions.

https://github.com/stefanhoelzl/codehydra

Always looking for feedback and ideas how others handle this!

stefanhoelzl··on Show HN: CodeHydra – Run multiple isolated AI coding agents in parallel
I was building Codehydra because since I started using coding agents, my workflow changed dramatically. I mostly run multiple sessions at once and I am mostly reviewing and testing the changes.

The context switching and keeping track of my running tasks got very demanding. Thats why I built Codehydra, it easily helps to keep track of my sessions and notifies me, which session need attention. It also allows me to automate the integrated VSCode in a way that it immediately presents the results when done with a task, so as soon as I switch to the session, I can immediately start the review/testing and see what was done. By using git worktrees I can test/review all changes in isolation.

Happy to answer questions and hear feedback. And also curious to hear what workflows/tools others would use this for.

stefanhoelzl··on Ask HN: What are you working on? (February 2026)
CodeHydra (https://github.com/stefanhoelzl/codehydra or https://codehydra.dev) a multi-workspace IDE for parallel AI agent development.

I've been using AI coding agents for more and more tasks, but didnt have the right tool to orchestrate. I'd kick off a task, then sit there waiting for the agent to finish before I could start the next one or had multiple terminals open was constantly checking which agents are busy and which are idle. Running multiple agents against the same repo also isn't optimal, and manually creating git worktrees added a lot of friction to my workflows.

Thats why I built CodeHydra. Just it at one or multiple local or remote git repos and it creates lightweight workspaces using git worktrees. Each workspace gets its own embedded VS Code and AI agent. You see all your workspaces in a sidebar with real-time status indicators showing which agents are idle or busy. I typically run 1-3 projects with up to 6 workspaces in parallel — review one agent's workwhile others keep going. I also put some effort into good keyboard navigation. So creating, deleting or switching workspaces is very easy by keyboard.

And to really automate our whole workflow, agents can control CodeHydra itself via an MCP server. An agent can close its workspace when it's done, spin up new workspaces for sub-tasks (with an initial prompt), or automate VS Code — like opening a browser preview tab or specific files.

Built with Electron + Svelte 5 + TypeScript. Supports Claude Code and OpenCode as agents. MIT licensed, works on Linux/macOS/Windows.

Try it with `npx codehydra` or `uvx codehydra`. I appreciate any feedback!

stefanhoelzl··on Framework (2nd Gen) Event is live on February 25th
What are your guesses about what they will show?
stefanhoelzl··on Ask HN: Is “Distributed CI” Possible?
You could try Bazel with a remote cache https://bazel.build/remote/caching

But it requires some work to move all your build an test targets to bazel.

stefanhoelzl··on What happens to the hardware (MacBooks, Monitors, Chairs) of failed startups?
If the company went bankrupt, the liquidator tries to get as much money as possible back.

I know from one case where all the IT equipment was sold to an auction house.

stefanhoelzl··on Running C unit tests with Pytest
This similar project could also be interesting https://github.com/mrh1997/headlock

It is used at my company to test our firmware written in C with python and was developed by a colleague.

stefanhoelzl··on Air France missed to register support domain
btw CNIL is the french national data protection authority https://en.wikipedia.org/wiki/Commission_nationale_de_l%27in...
stefanhoelzl··on Air France missed to register support domain
currently this page says:

Dear Air France, After filling in your Web form for a refund of my flight ticket, your infrastructure generated an e-mail to this domain which was then promptly bounced back to me since you failed to register it. I then spend 1h listening to bad quality music in your support hotline, before giving up. As you failed to register this domain, I registered it for you. Once you have processed my support request we can discuss me not forwarding all received emails to CNIL anymore. Have a good day! Christian

stefanhoelzl··on Ask HN: What do you think about type coverage in Python?
Yes exactly, mypy will catch mismatching types. But in my example I do not have mismatching types, but still can detect an issue when testing a function which expects `Optional[float]` only with an `int`.

Regarding literals or annotated variables: mypy will detect mismatching types in both cases.

stefanhoelzl··on Ask HN: What do you think about type coverage in Python?
mypy accepts much broader types then the concept I was proposing. Lets take this example:

  def foo(bar: Optional[float]) -> Optional[float]:
      return bar

  def test_foo():
      assert foo(5) == 5
It has 100% code coverage and mypy will accept it.

What I proposed will complain in this case, that you never tested `foo` with `None` or `float` as input parameters and that it never has seen `foo` returning `None` or `float`.

So even with 100% code coverage and mypy not complaining 'type coverage' could give you some hints that a) your tests are not extensive enough or b) your annotated types are too broad

stefanhoelzl··on About the Security Content of iOS 12.2
Looks like a lot of them would not have happened, if they had used Rust in the First place!
stefanhoelzl··on Show HN: Share your Git hooks and config
one goal of my approach was to not have any dependencies you have to install to be able to use it.

It is certainly not as mighty as pre-commit. But it doesn't add an extra layer around the stuff you know (hooks and gitconfig) and it can be enabled by setting just one git config value