Peter Sobot

I'm super proud to announce the release of Pedalboard, a new Python library my team just open-sourced. Use VST and Audio Unit effects from Python in just a couple lines of code!

i have this turned on when friends see it they freak i love this feature

Reminds me of Peter Naur's classic 1985 essay "Programming as Theory Building" which argues that a program is not its source code. A program is a shared mental construct (he uses the word theory) that lives in the minds of the people who work on it. If you lose the people, you lose the program. The code is merely a written representation of the program, and it's lossy, so you can't reconstruct a program from its code. If you think of total software debt as technical debt + cognitive debt, then previously, we mostly had technical debt. Now with AI we have both. Previously, when you built something, you accumulated technical debt but relatively little cognitive debt because you had to understand what you were building in order to build it. In other words: the theory came for free as a byproduct of the work. AI breaks that coupling. Now you can produce code without building the theory. So you're now able to accumulate both kinds of debt simultaneously - technical debt in the code and cognitive debt in yourself. And cognitive debt is arguably worse because you can fool yourself into believing it doesn't exist. Technical debt tends to show up in semi-obvious ways that we understand well as an industry. Cognitive debt is more insidious - it means you're unable to even reason about the program (because you possess no theory of it) - which is what Naur describes as the "death" of a program.

As an ML practitioner, this is extremely exciting for the promise of - finally! - statically typed tensor shapes. Nice work @1st1! #numpy-style-broadcasting" target="_blank" rel="nofollow noopener">peps.python.org/pep-0827/#nump…

(On that topic, @projection_lab is amazing and cannot be beat by a couple hours of Claude use)

One of the biggest dangers of using AI agents to "just write this quick tool for me" is that it'll get you 90% of the way there after a couple of hours; when you could have instead spent those hours using a much better tool that someone else has already created.

Powerful new Harvard Business Review study. "AI does not reduce work. It intensifies it. " A 8-month field study at a US tech company with about 200 employees found that AI use did not shrink work, it intensified it, and made employees busier. Task expansion happened because AI filled in gaps in knowledge, so people started doing work that used to belong to other roles or would have been outsourced or deferred. That shift created extra coordination and review work for specialists, including fixing AI-assisted drafts and coaching colleagues whose work was only partly correct or complete. Boundaries blurred because starting became as easy as writing a prompt, so work slipped into lunch, meetings, and the minutes right before stepping away. Multitasking rose because people ran multiple AI threads at once and kept checking outputs, which increased attention switching and mental load. Over time, this faster rhythm raised expectations for speed through what became visible and normal, even without explicit pressure from managers.

@sudara incredibly beautiful

razzle dazzle

research is an infra problem first and research problem second

reminder: only 10 days away. we can make it.

@bernhardsson Really good post - but please tell me this was an intentional easter egg? 😅

I just published this blog post about software companies buying more software and all its corollaries: SF being back, tech layoffs, revenue going up but margins going down, and much more: erikbern.com/2026/02/25/sof…

is the big bad AI going to take all of our jobs or is it just cold and dark and February and we’re all spending too much time on computer

Software engineering will only become more lucrative for those who love software engineering. We're experiencing a Tower of Babel moment, barbarians at the gate, so forth. The future of the profession belongs to those who loved the game before they invented the slop cannon.

Almost everyone goes on this journey: 1. "These are useless" 2. "These are interesting" 3. "I am getting more productive" 4. "I am bored and need to use 2 of these at once" 5. "I run an army of these robots, and if you don't, skill issue, you are NGMI" 6. "Wait, WTF is this code?" 7. "I was fooling myself, clarity of intent is the real bottleneck"

@spimescape I think that one was sadly not recorded - but the tl;dr was "fork() is a footgun in multithreaded programs, just use threads and GIL-free native code instead." Even more true nowadays when verifiably rewriting your Python code in GIL-free Rust or C++ is trivial with agents.

@psobot Is there a video of the talk? Was searching on YouTube, but couldn't find anything.

Didn't quite realise what a footgun multiprocessing + shared memory could be in Python. Maybe skills issue? But also confusing when same program either runs or hangs without changing the code.

surprised this button from NYT cooking hasn't caught on as a meme yet

Every engineer I've talked to says the same thing: AI made individual tasks faster but their days got harder. When a 3 hour task shrinks to 45 minutes, you don't do fewer tasks. You see yourself shipping faster, so your own expectations adjust. The baseline moves. It's like 996 culture but it sneaks in surreptitiously.

Opus 4.6 just referred to some Opus 4.5-written code as "a monstrosity." We have reached the singularity.

chat is it good or bad to get a handheld keyboard so you can continue to prompt your agents while working out on your indoor bike

looks like someone at Anthropic ran kubectl scale deployment/opus-4-5 --replicas=$HALF but has not yet run kubectl scale deployment/opus-5-0 --replicas=$HALF