For the best experience on desktop, install the Chrome extension to track your reading on news.ycombinator.com
Hacker Newsnew | past | comments | ask | show | jobs | submit | history | ZitchDog's commentsregister

I've been having good luck with fairly autonomous LLM coding with the following rules:

  * TypeScript everywhere with extreme enforcement of the type system.

  * No "as" casts, no "any" declarations, all code must understand the shape of its data

  * All boundaries validated using a typed validation library. Many use zod, I prefer tjs. I also have strictly typed pg and express wrappers.

  * No files longer than 300 lines

  * All of these rules are enforced by an eslint configuration that runs in a pre commit hook.

Global state and classes could also be removed via eslint rules, that would be interesting, though I haven't found it to be an issue in practice once the types are strictly enforced.

How do you enforce the use of validation library with eslint?

Can't fully enforce that one, but I require only my specific HTTP router / SQL library to be used, and disallow the import of express / pg libraries via eslint. I also disable the use of JSON.parse via eslint. Combined with strong language in the CLAUDE.md, I haven't seen it get around my guard rails.

You can’t trade options on an IPO.

You can post-IPO - depending on liquidity. I don't think that'll be an issue here.

And if the thesis of "it's going to look good for the first 15 days" holds, you can indeed be very profitable by e.g. buying ATM puts. (The problem being that markets don't like sticking to time tables just to accommodate your investment thesis ;)

So yes, you'll be able to take a bearish position fairly shortly after the IPO.


A Musk joint immediately after an IPO. ATM puts will be trading at what, 250% IV?

Probably. Hey, you picked a risky thesis heading into large uncertainty, that's not free.

But for the folks who are dead convinced it's all just a scam and it will quickly collapse, that's a way to put their money where their mouth is.


> they have no incentive to do this at a loss

Are you sure? Surely there is a lot of interesting data in those LLM interactions.


Many of them are promising not to store any of this. Of course we have to trust them, for all we know they are all funded by various spy agencies

Saying that becoming an EM is "moving away from tech" is crazy. As an EM you will be steeped in tech, just as you would be as an IC. It just may not be the tech you want to be steeped in. Again, same as an IC. In either case, unless you are working in AI, you will need to "play" with things like OpenClaw in your spare time.

The real reason not to become an EM in 2026 is because AI makes our jobs 10x harder.


> The real reason not to become an EM in 2026 is because AI makes our jobs 10x harder.

This is true, but our job was getting kind of boring anyway. Time to lead, not manage. We should be having just as much fun as the ICs, and the best I know are having the time of their lives.


I've created my own "claw" running in fly.io with a pattern that seems to work well. I have MCP tools for actions that I want to ensure human-in-the loop - email sending, slack message sending, etc. I call these "activities". The only way for my claw to execute these commands is to create an activity which generates a link with the summary of the acitvity for me to approve.


Is there a risk that the summary doesn't fully match the action that actually gets executed?


Side note: Just like with a human employee asking for permission to do something.


Except for the accountability if they screw up; and the human brain thinking through what they are doing.


Nope! The summary is presented to the user via a link and once the user follows the link and approves, the action is implemented entirely outside of the agent, on a separate server.


Any chance you have a repo to share?


The approval-link pattern for gating dangerous actions is something I keep coming back to as well, way more robust than trying to teach the agent what's "safe" vs not. How do you handle the case where the agent needs the result of the gated action to continue its chain? Does it block and wait, or does it park the whole task? The suspend/resume problem is where most of these setups get messy in my experience.


I haven't needed that yet! But it seems like the agent could easily be notified of completed activities.


I had the same question recently! So I did an experiment to see if I could create something of value using agentic coding.

I made the worlds fastest and most accurate JSON Schema parser.

https://github.com/sberan/tjs

But, nobody seems to care. The repo only has 18 stars and my Show HN post got no upvotes. I'm not sure what to take away from that.


> In the domains where I'm an expert, it's a nice productivity boost. In the domains where I'm not, it's transformative.

Is it possible that the code you are writing isn't good, but you don't know it because you're not an expert?


No, I'm quite confident that I'm very strong in these languages. Certainly not world-class but I write very good code and I know well-written code when I see it.

If you'd like some evidence, I literally just flipped a feature flag to change how we use queues to orchestrate workflows. The bulk of this new feature was introduced in a 1300-line PR, touching at least four different services, written in Golang and Python. It was very much AI agent driven using the flow I described. Enabling the feature worked the first time without a hiccup.

(To forestall the inevitable quibble, I am aware that very large PRs are against best practice and it's preferable to use smaller, stacked PRs. In this case for clarity purposes and atomicity of rollbacks I judged it preferable to use a single large PR.)


I had the same question recently! So I did an experiment to see if I could create something of value using agentic coding.

I made the worlds fastest and most accurate JSON Schema validator.

https://github.com/sberan/tjs


Why do you say it's the fastest and most accurate? I don't see any raw stats, just AI generated readmes.

you also don't compare it to the top result on Google https://github.com/Stranger6667/jsonschema was that intentional?


Here are benchmark results vs AJV: https://github.com/sberan/tjs/blob/main/benchmarks/results/B...

There are raw stats in the main graph image of the readme (op/s)

I don't compare it to that validator because these are JS only - I'll update to specify the language preference. Thanks for the feedback.


Hi HN!

I built TJS over my holiday as a way to experiment and get better at agentic coding. I was astounded at how quickly software can be built when you have a full test spec like JSON Schema has.

The thing I'm most proud of is the TypeScript type inference. It automatically deduces the types from the JSON Schema. Try it out in the playground to see what I mean! https://sberan.github.io/tjs/#playground


I used similar techniques to build tjs [1] - the worlds fastest and most accurate json schema validator, with magical TypeScript types. I learned a lot about autonomous programming. I found a similar "planner/delegate" pattern to work really well, with the use of git subtrees to fan out work [2].

I think any large piece of software with well established standards and test suites will be able to be quickly rewritten and optimized by coding agents.

[1] https://github.com/sberan/tjs

[2] /spawn-perf-agents claude command: https://github.com/sberan/tjs/blob/main/.claude/commands/spa...


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search:

HN For You