David Ackerman

yeah, that's right. i let gpt-5.4 have enough permissions to wipe one of my hard drives and reformat it had to type my sudo password like 5 or 6 times but it worked flawlessly

so annoying that my phone and desktop can't both be properly connected to my bose headphones at once. i wonder if a small physical device could be an intermediary to do this perfectly

@Shpigford i wonder if people would quibble about what "breaking" means? these would probably be bad customers but still, could be a headache

idea: you pay me $500 to record a video of me signing up and trying to use your service. if nothing breaks, you get your money back. my track record says almost nobody gets their money back.

@keysmashbandit imagine how uncomfortable you are right now compared to 300 years in the future!

The thing I feel most cognitively dissonant about when thinking about history is how much time everyone spent being physically uncomfortable. Heavy itchy wool, greasy hair, too cold or hot. If I try to imagine myself doing daily life circa 1704 I get immediately blocked by that

@steipete This is why Google puts "AI can make mistakes" on every gemini-related component

This guy emailed me asking for a *token session refund* because his claw made mistakes. 🙃

@thekitze @noisemakerjon No way. You're lying

@noisemakerjon cabinet ofc. if it's up to me zero machines should be on the kitchen counter, even coffee machine that gets used once per day

just realized that the only thing that belongs on a sink is soap. everything else that u use 1-2 times per day should be hidden in a cabinet. it look so much cleaner

@thekitze Now do the kitchen sink. Do you have A sponge and/or dish soap out?

@thdxr They'll just resort to using the chipotle customer support bot

almost ready to dog food a new product and i can't wait to turn these on make these zoomers type some code

always needing to use 5.4 xhigh is a skill issue

if @opencode could support /fast mode for openai models i would use it exclusively

another candidate for the mental model that is in my head

the IQ/think quadrants with gpt are interesting gpt-5.3-codex-spark low: low IQ, yolo it gpt-5.3-codex-spark xhigh: low IQ, think really hard gpt-5.4 low: high IQ, yolo it gpt-5.4 xhigh: high IQ, think really hard

> Specs are low fidelity they can be as low or high fidelity as you want > They aren't a common shared DSL why is this needed? > Have very low signal to noise I actually think the reverse. you can convey your point without a bunch of unnecessary syntax noise > Not verifiable or deterministic I sort of agree with this but don't think it matters too much. it can convey "intent" better than code (unless you include comments, which is just spec) and you can refine it until it ensures the properties you want (without overspecifying) > They don't encourage iterative work why not? the spec can evolve just like the code can evolve > Almost always written by an agent, thus prone to more slop sure, but it's easier to audit than the code because it's a compressed representation > Lose their inherent value as soon as they are converted to code. > Specs are throwaway not if you keep it up-to-date I'm not saying we don't need code (of course we do), but it's starting to feel like an amazing pseudocode layer to me. I think most of the things you are disagreeing with /reacting to are not totally invalid, but they are examples of it being done wrong. I think we're still figuring out this different way of working so I'm unsurprised we haven't nailed it yet. But it has lots of potential.

I don't know. Hard for me to agree with the spec-is-code argument. Specs are low fidelity. They aren't a common shared DSL. Have very low signal to noise. Not verifiable or deterministic. They don't encourage iterative work. Almost always written by an agent, thus prone to more slop. Lose their inherent value as soon as they are converted to code. Specs are throwaway. They are a way to temporarily express behavior intent - an agent translates that to a version of code. That's the thing that lives forever. If you are convinced that spec is code then you should be able to confidently delete all code and regenerate from specs.

just compared the same design doc and prompt for gemini 3.1 vs. claude to prototype a complex frontend from scratch and gemini won by a mile gemini is not perfect, but claude's looked like a 10th grader coded it up after school

I don't really agree with this take. Writing actual code comes with tons of accidental complexity that isn't really making your product better or helping you ship faster. Sure, a "spec" exists on a continuum of detail that eventually bottoms out at actual code. But most programs do not need specification at the level of current programming languages, just as most programs don't need manual memory management. haskellforall.com/2026/03/a-suff…

@unclebobmartin yeah i think our prior lack of ability (or desire) to handle arduous refactorings provided a good pressure to keep things structured well at all times now that we can manage drastically worse code, most will unfortunately write drastically worse code

The Slog. We all know about the slog. We've been postponing a bit architectural refactoring because we know it's going to be a slog. But eventually the pressure builds and we heave a great sigh and begin the long arduous process of making a thousand dangerous changes and running the test suite as often as possible. Along comes the AI and suddenly the slog doesn't seem like such a big problem anymore. We just tell the AI to slog through, and twenty minutes later it's done; and it's right! And so off we go, confident that slogs are relegated to an ancient past. We'll never have to slog again! And then comes some deep systematic flaw that we must correct. And the AI simply cannot deal with it without hours of constant babysitting and monitoring. And there we are, slogging again.

Concretely I'm actually in the weeds designing the architecture and then actually looking at the code to make sure it's doing what I want. Plus directing the test strategy I'm honestly still not writing much code by hand though. fighting with syntax is just inefficient

Currently challenging myself to write a "perfect" android app that is high performance and bug free (e.g. using AI superpowers on quality instead of quantity). It feels like eating celery

@zeeg I think humans need to catch up to working effectively with LLMs. It's harder than ever to be disciplined in the face of the AI slot machine Unfortunately AIs will probably change again before we figure out the current situation

im fully convinced that LLMs are not an actual net productivity boost (today) they remove the barrier to get started, but they create increasingly complex software which does not appear to be maintainable so far, in my situations, they appear to slow down long term velocity

@kentcdodds I'm going to do it too Probably won't work but you have to try

The urge to build my own agent with my own memory system is overwhelming.