Peter Hraska

Product × design × engineering. 10+ yrs Slido (acquired by @Cisco). Burning tokens with agents, posting what actually works.

[Peter Hraska @virpo]

here's the skill: https://t.co/w9gEuksdFq

(the video was made with a single prompt and Opus 5.5)

[Peter Hraska @virpo]

from 0 to €1,059 revenue with AI!
well, by selling stuff I didn't use :)

sharing it as a skill: you take the photos, agent figures out what each thing is, prices it, writes the listing and posts it on Craigslist (or similar)

bookmark it for when you clean out the basement!

[Peter Hraska @virpo]

and review got rebuilt too. their new tool, Antique, knows a comment tweak from a sensitive logic change, so people spend their attention where it counts.

at 30,000 a day, nobody reads every diff anyway.
full post ↓
https://t.co/xqSCoM5kG6

[Peter Hraska @virpo]

so 7 people built Scion in five months. the code lives on the server and files show up only when you open them. checkouts went from minutes to seconds

[Peter Hraska @virpo]

git assumes you work offline, so every machine keeps a full copy of everything

[Peter Hraska @virpo]

huge. after going from ~2,500 to 30,000+ PRs a day, Anthropic broke git and vibecoded their own alternative

🧵 read more interesting bits

[Peter Hraska @virpo]

one checkout of their repo is 25+ GB. CI does that 25 million times a day. git was simply not made for this

[Peter Hraska @virpo]

@simonw your card ended up as the cover of the site, pelican on a bicycle included :)
Showboat is your trick on it.
let me know if anything's worth fixing

[Peter Hraska @virpo]

only 8 out of 94 internet-famous engineers still read every line their agents write.

I went through all their setups and made each one a card, with the trick worth stealing 👇

[Peter Hraska @virpo]

👉 https://t.co/8OchENRYQe

you can make your own card too, just paste the prompt into your agent

[Peter Hraska @virpo]

we all learned by reading each other's PRs.

now most of the work happens before the PR: the prompt, the skills, which model. you don't see any of it.

so I went through how 94 engineers actually set up their agents. 37 use both Codex and Claude Code.

[Peter Hraska @virpo]

@thdxr INVARIANTS.md. stuff that has to stay true now and in a year. agents check every merge against it, and if they want to change one they have to ask me. works surprisingly well

[Peter Hraska @virpo]

@ChanduThota and next up: Gemini in Docs reads the folder's AGENTS.md and follows the team's rules? :)

[Peter Hraska @virpo]

Stop your agent from writing AI slop.

The cure was written in 1979, for aircraft mechanics: ASD-STE100.

No, not a new virus. But I hope it spreads.

Paste this into your agent:

Add to my global AGENTS.md: "Write in ASD-STE100, about 80% strict."

[Andrej Karpathy @karpathy]

We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks:

Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent. But even better:

Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand. But even better:

Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc. But even better:

Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work!

In summary:
- As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding.
- Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you'll be surprised.

[Peter Hraska @virpo]

@goando Such a great idea! Made it work for my Stream Deck Mini as well.

Code is here: https://t.co/1IaL0sKncu

[Peter Hraska @virpo]

@LysandreJik @art_zucker @remi_or_ When can I visit? :)

[Peter Hraska @virpo]

This might be my favourite use of AI agents so far.

You can give @openclaw a messy sketch + pegboard dimensions, and a few minutes later you have printable STLs. If your printer is on the network, it can send them there too.

https://t.co/1qNrtGVK3Y

[Peter Hraska @virpo]

"is this OpenClaw thing actually useful for something?"
literally improving my health.
gym classes near me open at midnight and sell out by 7am
told my claw to sign up for me, took literally 2 minutes
woke up to this 👇🦞

@OpenClaw @steipete

[Peter Hraska @virpo]

3 engineers. 5 laptops. 17 AI agents.
We won an investigative-data hackathon by turning scanned PDF contracts into structured data + agent workflows that cross-check entities across datasets in minutes, not hours.
Demo: https://t.co/XedT9rtIXY
@mareksuppa @jakubkratky_ + Andrej

[Peter Hraska @virpo]

@pvesterbacka Glad we could have you, Peter!