Lyfe Long Learner

I’m a designer, I’m a programmer, and I’m an autodidact. I like tech that lifts others, not preys on them. I like ambitious vision that’s focused on making the future brighter than the present. Finally, I enjoy thinking and sharpening the greatest asset I have - my mind.

Este artículo me encantó: "Deja de pensar en la AI como un coworker. Es un exoesqueleto" Acerca de como el uso d estas herramientas amplifican las tareas de software engineering, y un contraste de realidad donde la AI por si sola es bastante mala completando tareas correctas

@thdxr @ubax @opencode We read the code

@ubax @opencode 1. good team 2. we read the code

First time using Claude Code since I wanted to checkout Opus 4.6. But man does it's UX feels so basic compared to @opencode. How is it that a small group of people like @thdxr have been able to beat such giants with almost infinite resources? Boggles the mind!🥴

@karrisaarinen @staysaasy THIS, this, is the way! While some leaders speak of doomsday scenarios in the USA, abroad labs are forming partnerships to embed ai in learning and augment future generations of creators, makers, teachers, and scientists. techcrunch.com/2026/02/18/ope… We can have a richer future!

I’ve been thinking why are all AI labs/industry running doomsday predictions. AI replaces every industry. Video models = movie studios, cooked Cowork: white collar work, cooked Voice model: support, cooked Coding agents: swe, cooked What if this technology helps people to do better work? Like now you can get better support because AI handles more of the dumb work. When is the last time you got great support? It’s not like most of these domains operate in some golden era of quality, but more in some level of enshittification era. My hopeful view is that AI will lift all boats and will make people or companies that are very good at what they do, to make it even better and pull ahead more.

If you’re wondering why people are making these ridiculous claims - it’s because if this doesn’t happen their capex won’t pay off and they’re fucked.

@raunofreiberg @asallen @glennui @lochieaxon I’m paying close attention 👨🏾‍💻

We're teaming up with @asallen, @glennui, @lochieaxon, and more. The mission? Recreate and preserve software history.

@vanschneider Immediately saved to “My mind”. How many well educated capable individuals lead a life where none of the immense potential they posses is used towards causes or efforts they themselves have imagined or even less believe in. That’s so very unfortunate.

@joshpuckett @WillManidis Josh Puckett with the endorse, that’s a great sight 👌🏾

@WillManidis Will, please. I am running out of long blocks of time to read and sit with your brilliant writing. I am desperate for you to turn down the heat 🥵

@WillManidis Phenomenal Read 👌🏾

@asallen “Design is ultimately about values—it’s our purest expression of the world as we want to see it.” Having the ability to envision new futures that impact millions and then do the work to make those visions real, is why I still think design is the greatest profession.

@obsdmd Obsidian is the best in the game

Obsidian CLI help.obsidian.md/cli

Anything you can do in Obsidian you can do from the command line. Obsidian CLI is now available in 1.12 (early access).

@mitchellh This was a great read!

Wrote up about my personal journey from AI skeptic to someone who finds a lot of value in it daily. My goal is to share a more measured approach to finding value in AI rather than the typical overly dramatic, hyped bait out there. mitchellh.com/writing/my-ai-…

@ThePrimeagen Not related but if you’re still considering a NYC stop share the details I’d pull through. Hard skills still matter, thinking still matters, ai is an incredible tool, but the greatest force multiplier in the world times zero is still zero.

as of now i think the best use case for ai is context, my context. i programmed a feature that required a couldflare R2 guard to prevent multiple updates, part of that i wanted it the value in r2 to always be there so an upfront init to be called. unfortunately life happened and i had to return to the code ~4 days later. i forgot about the init idea i had and then started to debug the test. its super annoying to debug due to launching cloudflare wrangler as a sub process, hooking up the client, adding logs until i can find the point of failure, etc etc. integration tests amirite? anywho, i just tell my boi unbreakable kimik2.5 to run integration tests and add logging until we can find the point where the next break is and surface it to me to figure out what to do next. found the init issue within 2 minutes, saved me ~30 minutes of debugging, and i created a very fun piece of code to solve this. and then i tell it to continue to do debugging until the next issue, surface it to me, give me logs to prove it, and ill think of the next solution. this genuinely saves time

@asallen +1 on what he said!

Expect to see more of this kind of design job…

@mikeyk Can you use your Claude code subscription with Xcode? @mikeyk

Xcode 26.3 launched today with a native integration with the Claude Agent SDK, the same harness that powers Claude Code. Devs get the full power of Claude Code (subagents, background tasks, and plugins) for long-running, autonomous work directly in Xcode 🤖 anthropic.com/news/apple-xco…

@joshpuckett @darylginn I just checked my email and it’s there 🙌🏾. Thank you for the transparency and the quick follow up!

@_rcpg @darylginn You should have received an email last week! DM me, wanna see if there are spam/delivery issues

Are you currently using AI to design? Or are you still 100% Figma?

@thdxr @pvncher What ?!?! 😂💀😂

@pvncher i'm like drugs are fun but i don't want you to die otherwise we can't keep doing drugs

people get so mad when i'm harsh on vibe coding then they get really confused when they realize i work on opencode which makes them even madder

@joshpuckett @darylginn Yes, interfacecraft.dev, any updates on a release date? I pre-ordered it a while back, but wasn’t sure if any announcements had been made about an upcoming release.

@_rcpg @darylginn Interfacecraft.dev *

@joshpuckett @darylginn Will the full interface.dev offering also be available tomorrow?

@darylginn Figma's great for many things, but to really dial things in and ship nothing beats designing in the browser. So I'm making this for you: x.com/joshpuckett/st…

@8bit5_0 @MLStreetTalk “Multiplicative in base competence” is huge 999,457,389 x 0 is still … 0. People still need understanding of the domain they’re in (aka actual skills) for any of the ai tools/agents/whatever to have value that’s not some neat demo reel level trick or LinkedIn hustle bro post 💀

@MLStreetTalk There is much to be said about your second to last paragraph. For simplicity, I’ve started calling ai agents as mech suits that are multiplicative in base competence. Many people assume they are additive when they are in fact not

Dario Amodei published "The Adolescence of Technology" the other day, claiming: that AI could "displace half of all entry-level white collar jobs in the next 1–5 years", enable "10–20% sustained annual GDP growth" and is already so capable that "some of the strongest engineers I've ever met are now handing over almost all their coding to AI". Four days later, Anthropic's own study (by Judy Shen and Alex Tamkin) seemed to contradict Dario's claims. It is almost remarkable that Anthropic published this study. I do credit Anthropic's research culture for allowing such findings to see daylight though. It was an RCT with 52 software developers learning a new Python library. Half used AI assistance and half coded by hand. - The AI group scored 17 percentage points worse on comprehension tests - The largest gap was in debugging - the skill arguably most needed to verify AI code (in the olden days at least) - The AI group wasn't even faster - but super delegators saw time savings - Participants reported feeling "lazy" and aware of "gaps in understanding" Those who delegated fully to AI learned very little. Those who used AI for conceptual questions and explanations (who arguably already knew what they needed to learn), retained knowledge just as well as the control group. Amodei says that the strongest engineers delegate effectively, then extrapolates that to predict that entry-level workers will be displaced. Strong engineers can delegate because they already understand deeply. They provide the missing coherence and constraints to delegated tasks. They know which questions to ask. They can verify outputs and catch errors. The study says as much. Entry-level workers are the very folks who lack competence, and who need to learn ...and... the study shows that AI actively impairs their learning! Errors are how we learn, like the old adage that you "learn from your mistakes". It's only when we make mistakes that we rearrange our constraints. The no-AI group encountered more errors, and these errors helped them learn. Non-interactive AI removes the productive friction which leads to profitable increases in understanding! Maybe some delegated BECAUSE they didn't know enough in the first place to ask smart questions or verify the results. Whether delegation causes incompetence or incompetence causes delegation is an interesting distinction but the result is the same. I've been calling this phenomenon "understanding debt" for the last few years i.e. that when we delegate, at best we stop learning, at worst, our comprehension and competence atrophies. It's a chicken and egg problem, you need to already understand a lot to get the best out of AI. AI amplifies your existing competence - it does not substitute (much) for incompetence past some threshold of complexity. Amodei says that "AI is now writing much of the code at Anthropic" and that this creates a "feedback loop" accelerating AI development. It's possible that AI will get so good that we will stop debugging and just specify tests and other forms of function descriptions. But that moves into an abstraction layer which entirely screens off mechanistic understanding of the code. The faster you go, the less you specify, comprehend and verify. And the fact that the AI group (in the study) KNEW this was happening! They reported feeling 'lazy' and aware of their own skill gaps. It was seductive - even though most weren't actually faster, and they sensed it was harmful. But many developers will take the path of least resistance. There is a weird phenomenon "the supervisor illusion" with language models that we discount the role of human supervision and competence. When a technically sophisticated user - let's say, an Anthropic engineer makes Claude Code sing, they implicitly provide constraints that guide coherent generation (competent prompts, specifications, tests, tool critics, questions and other verification). But the supervisor is doing the heavy lifting. This doesn't generalise to incompetent developers. Not quite imminent job displacement.

@joshpuckett @RobinPlu @darylginn Thank you for providing an update. I’m patiently waiting for access as well. ✌🏾

@RobinPlu @darylginn I'm a believer that anything worth doing is doing well. I appreciate your patience. The initial set of collections will be published within ~1-2 weeks.

What are you working on right now?

@kepano @irishdrug1511 @obsdmd I also came across this for what it’s worth github.com/chocoford/Exca…, but haven’t gotten around to trying it. Again though don’t sleep on the iPad if the team was considering Apple Pencil support for drawing in the future. Maybe even on the canvas option within Obsidian.

@kepano @irishdrug1511 @obsdmd I have other solutions like goodnotes but they’re clunky, & don’t live in the same folders as my obsidian vault. I’ve decided to use an interim solution for now where I take Freeform notes with my Apple Pencil in concepts, but the workflow is still rough.

What's the best way to use apple pencil on @obsdmd? @kepano? I use ink currently, it's fairly outdated & might do claude-coding on it to make some changes, but why reinvent the wheel?