Max Sumrall

@mattpocockuk That grill-me skill + pi-interview extension from @nicopreme is chef's kiss. I can stay in a grilling session for hours. I started using it mainly for fixing bugs. It always finds bad design choices and asks the right questions. It's not just for building new features.

@teslaeurope why can HW3 owners subscribe to FSD? Especially those with EAP, what is the intent of the €49 subscription to FSD?

@tvanschadewijk Interesting, there’s a minor lane shift there in the middle of the intersection which it seems to gotten confused by.

We are seeing lots of examples of FSD doing really well in the Netherlands, and that’s great. But from my own and my friends’ experience, I know that despite it doing great most of the time, it makes mistakes every now and then. I’d love for these to be shared too so we get a realistic view of EU FSD. Out of the 4 instances I experienced, I was only recording during one - and it was the most serious. The car turned left onto Wibautstraat, a big through-traffic street in Adam East. It then did something really weird: in the flow of traffic after a green light, it suddenly stopped in the middle of a crossing on the right side of the right lane (but blocking traffic from behind), despite the navigation directing it to continue on Wibautstraat. 🤷‍♂️

@screenfluent @nicopreme I just added in the /grill-me prompt “..using pi-interview..”

@nicopreme How do you combine it?

/grill-me + pi-interview is a cheat code 🔥

@sundarpichai @backlon @GeminiApp @antigravity Listen to gramps @badlogicgames

Introducing Gemini on Mac. It’s the first time we’re bringing the @Geminiapp to desktop. The team built this initial release with @Antigravity, and it went from an idea to a native Swift app prototype in a few days. More features on the way!

@lucasmeijer @onusoz “Paul Dirac’s beauty in code” only goes so far, also looking for ideas here. cc @khoiracle

@onusoz how do we integrate this in a daily workflow though? it's not yelling "now make it simple if you can" after every session. but what is it?

Agents always want to add. it's low risk. We need systems that push back. That remove. That combine, that simplify.

@xanderwasserman @teslascope @NotATeslaApp @KRoelandschap @Rustavi FSD works with HW3, on v12 right? Better than nothing. Or does it not work?

Confirmed, FSD subscription is €49/mo in the Netherlands if you own Enhanced Autopilot. PS: why can I purchase the FSD subscription on my HW3? @teslascope @NotATeslaApp @KRoelandschap @Rustavi

@badlogicgames @lucasmeijer This was so much fun seeing in person with no expectations, and realizing there’s someone else who’s excited about pi and has similar feelings about using llms. So happy to have met @lucasmeijer

recommended viewing. one more time, on it's own. this is probably yhe most practical talk on using coding agents i've watched to date. watch it. by @lucasmeijer it's also a great demo of pi and captures exactly why i built it. youtu.be/fdbXNWkpPMY?si…

The difference between the two, for serious engineering work, is simply brutal. Claude Code with Opus is, when the task at hand is very complicated, borderline useless, while GPT 5.4 can do a reverse engineering mixing: hardware knowledges, major disassembly skills, and so on.

@jjpcodes @comper_io How can I try @comper_io ?

plugged gstack into @comper_io . doesn't look that wild compared to openclaw honestly? (lots smaller! openclaw codebase is biiiig) like 80% of the code is in the skills; bunch of tests, skills do look like slop tho. not as bad as i thought tbh.

@VoxxedAmsterdam @badlogicgames @nicopreme When: 12:10 in room 10 Add it to your favorites: m.devoxx.com/events/vdams26…

@VoxxedAmsterdam Come to my talk on day 2! It's tips for coding with LLMs: "What AI agents actually need from you". If you can't make it, the tl;dr is install pi from @badlogicgames and skills from @nicopreme, and don't clank it harder than the situation allows.

@badlogicgames @ferologics Watch? Nobody’s watching YouTube anymore. Lobster boy has a cli for that: github.com/steipete/summa…

@ferologics ain't nobody got time to watch videos. or implement operational transform.

welp, vibed it in an hour. i really only needed single writer, multiple readers/commenters. easy enough. github.com/badlogic/jot

@badlogicgames I vote for “Clanker Commander”

chat, is he serious?

@hibakod Because of this potential bug, OpenAI has reset limits for users.

@onusoz Is the visualization just a one-off thing? Looks nice. Making flows seems kind of hard, looking at the examples. Obv not with an agent helping, but a built in viz could help.

acpx v0.4 ships Agentic Workflows, or as I like to call them "Agentic Graphs" It let's you create node-based workflows on top of ACP (Agent Client Protocol), to drive any coding agent (Codex, Claude Code, pi) through deterministic steps This let's you automate routine, mechanical legwork like triaging incoming PRs, bugs in error reporting, and so on... For example, OpenClaw receives 300~500 new PRs per day. A lot of them are low quality, but they still relate to real issues, so you have to address them somehow You need to: - extract the intent - cluster them based on intent - figure out if the proposed changes are legit, or whether they are slop local solutions, like trying to catch flies instead of drying out the swamp - if the PR is too low quality or the intent is not clear, close them - run AI review on them them and address any issues that come up - refactor them if the changes are half-baked - resolve conflicts - and so on... So that when the PR is presented to the attention of the maintainer, all the routine legwork is done and the only remaining thing is the decision to (a) merge, (b) give feedback to the PR author, or (c) take over the PR work yourself I wanted to build this feature since a couple months now, since Codex got so good. OpenAI models are now good at judging implementation quality, so I found myself repeating the same steps I wrote above over and over I also tried putting all this in a single prompt. But I believe there are workflows that should not be a single prompt, but a sequence of prompts in the same session That is because like humans, LLMs are prone to PRIMING. I claim that putting all steps in the same prompt at the beginning of the context will generally give suboptimal results, compared to revealing the intention to the model step by step Creating such a workflow also gives more OBSERVABILITY into the each step that an agent is supposed to take. Agent generates JSON at the end of each step, and that structured data can be used to monitor thousands of agents running at the same time in an easier way, on a dashboard Similar features have been introduced in e.g. n8n, langflow. But AFAIK they are not integrating ACP like the way I do I wanted to have a fresh approach, and to build an API that I can develop freely the way I want, so I created a new workflow API inside acpx The video is from the workflow run viewer, but that is not where you build the workflow. You build it by using the acpx flow typescript API. See examples/pr-triage in acpx repo Before building that, I started from a Markdown file with a Mermaid chart of the flow I had in mind. The Markdown file acts as a spec for the flow, and I have built the workflow through trial and error. I call this process "workflow tuning" I started working on acpx repo PRs one by one, tuning the flow, slowly scaling to more PRs. Finally, when I felt confident, I ran it in parallel over all external open PRs in the acpx repo. I believe it already saved me hours this week My next goal, if well received, is to set this up on a cloud agent so that it can process the 300~500 PRs the OpenClaw repo receives every day, in real time, as they come in I believe this will save all open source maintainers around the world countless hours and make it much easier to herd and absorb external contributions from everyone!

@jjpcodes @badlogicgames “clankin it to the max” need to be on a tshirt

thoughtful post. you're totally right, but i'd limit this to the subset of software that _actually matters_. (you sort of touch on this but not super explicitly). a lot of software doesn't matter, including a lot of the slop-projects that many of us ship. those are, well, tech demos at best, and it's fine, if not great, for them to be hastily thrown together slop. we can now validate faster - and that's a great thing! if you can clank out a slop-startup in 2-3 weeks instead of 9 months, get PMF for a tool and then re-write, thoughtfully iterate - and make a good version (that scales, has real users, etc), i'd say that on balance that's a good thing. obviously, Serious Business Software should probably not be clankin it to the max and bringing society down with them. and boasting about having INT_MAX commits per day is very silly. but at the same time, we can still have some fun clankin it for the majority of throwaway, inconsequential, toy software. as long as we recognize it for what it really is.

I'm usually not one to write thought pieces without much technical depth. But here we go. Slow the fuck down. mariozechner.at/posts/2026-03-…

@badlogicgames It gets better

i love this comment. it illustrates how we software people permantently get real-life wrong, then build systems that are supposedly helpful, when they are not. bonus points if you claim the system replaces humans. the world is more complex than it seems. domain expertise is still everything.

You can download it here: github.com/maxsumrall/pee…

What week is it? Week 14? Or 13? You can just do things. So I made a menubar app that tells me what week it is. (Maybe it already exists, but building it with codex was easier). I even spent the time to get it notarized by apple, so it just works.