Pedro Teixeira

Big news: Graphy’s charting infrastructure now powers charts in @GammaApp

😬

@onel0p3z Yes, having had complex things exploding in your hands is also a big factor 😁

@pgte And experience ... You've been there. You've made that mistake. After over engineering a solution, you end up going back to keeping it simple and decouple as much as possible and adding checks and levels and alarms where it makes sense. Then you go and remove the noisy ones 🫠

I think my low tolerance for complexity comes from learning that the art of good design is achieving an elegant solution to a hard problem. A the most elegant solution is (at least to me), the least complex solution.

It's been a great first week at @GraphyHQ! Been head-down all week trying to get up to speed, but now it's time to educate myself more on charts and visualisations 📈. Starting with this one amazon.com/Good-Charts-Up…

Personal news: I just joined @GraphyHQ as a Software Engineer! 🎉 Super excited to get to work with the team.

@ordepdev what about automating merges coming from upgrades at least? Once I have a robust-enough e2e tests, smoke tests and a easy way to roll back if needed, for some projects I do that.

Merging pull requests is always the main bottleneck. I keep pushing for some required upgrades, and people suggest automating things. But I always say, just look at all the open PRs waiting to be merged. Automation won't help if we don't actually merge them...

When debugging, I make them "suffer" instead of suffering myself. Ask them to create an E2E test that replicates the problem you're having, and then let them linger for the time it takes to fix the issue, autonomously. It's just that it works, mostly, but this is a bit too brute-force to me, but that's what we have right now... 🤷‍♂️ Oh, and have them save the lessons learned at the end of it.

I really hate when scientists refer to our skill learning as memory. To me, memory is an imprint of a perception that lingers in our brain and that can be recalled or activated. A very different thing, to me, are the behavioural potential changes that we call learning, that rewire our cognitive and reflexive brain.

For when you feel itchy and you don't know why

Reverse-Engineering the Wetware: Spiking Networks, TD Errors, and the End of Matrix Math 👉 metaduck.com/reverse-engine…

Just dropped a new blog: Reverse-Engineering the Wetware: Spiking Networks, TD Errors, and the End of Matrix Math 👉 metaduck.com/reverse-engine…

@heyblake helpmaton.com

Drop your project URL Let’s drive some traffic

I dearly believe that exploring how consciousness can be constructed will be brought to us by software engineers. Hubristic? Perhaps, but SWEs are used to absorbing and applying cross-disciplinary principles all the time, and they usually have (the good ones) a fairly well developed mechanistic intuition.

@auchenberg Sorry but these are terrible IMO, they have no puns, and you know puns are the highest form of humor 😅 My favorite one still is Deci -> Decimate Also don’t discount the mascot-ability of the name: Deci - a ten-legged octopus 🐙 Big decisions.

Choosing a company name that has a great name for your employees is underrated: 𝗦𝘁𝗿𝗶𝗽𝗲 → Stripes 𝗔𝗻𝘁𝗵𝗿𝗼𝗽𝗶𝗰 → Ants 𝗚𝗶𝘁𝗛𝘂𝗯 → Hubbers 𝗧𝗼𝗮𝘀𝘁 → Toasters

Loading states: When an app has at least one button that stays active after I clicked it and while it’s loading, it instantly gives me the ick. It’s astonishing the amount of heavily used apps that do this.

Recently I've been using a combination of different models to define and refine a high level spec when the task at hand is critical. I use gemini 3 pro in high reasoning mode to help me start thinking about a subject, based on a *lot* of context about the current product. Then I derive a high level spec fit to give to a coding agent where I iterated for clarity and accuracy. Then I use ChatGPT 5.3 codex to create a plan to implement this. Review, and then implement. Then I use a bunch of macros I call: - ensure the plan is completely implemented - ensure there is enough test coverage to make the use of this new feature generally save. - review and improve (I have detailed instructions for each of these). Then I create a PR on Github and ask for a review from Copilot. Then I use a script and instructions so that the coding agent can retrieve and reason about the review. I repeat the review process. Done.

It was such a shit-show for me I had to uninstall the thing. Probably a skill issue on my part... 🤷‍♂️

I finally have some use for my dormant openrouter credits in my personal account

Official version 1.0 of @helpmaton is out! 🥳 (Fixed a bunch of things, reviewed the UI, but mostly just to make semantic versioning work for me).