Pythonics

[Pythonics @pythonics00]

@RhysSullivan @JosephJacks_ @Atlassian @planepowers Terrible look for plane tbh

[Pythonics @pythonics00]

If you’re doing hard comp sci stuff and not web dev, then sometimes by hand is the way to go. I don’t write any web dev code by hand but I do write some small subset of firmware by hand. The frontier models will get most of the way there today, but even now gets tripped up, doing things like changing the buffer size and breaking camera input processing downstream. This is mostly due to a lack of verifiability because the hardware side is so specialized and I don’t have adequate sims.

I’m sure some anthropic folks doing kernel dev need to act with known precision as well or training environment construction. There is a very small subset of tasks where it’s still faster and more reliable to do it yourself. Luckily that’s an increasingly small area.

[Pythonics @pythonics00]

@michaelflux @peer_rich Give it a few years. We are still extremely early in adoption.

[Pythonics @pythonics00]

PSA: If Spectrum Security Shield decides to destroy your business on a random Saturday by blocking your site for all Spectrum customers with a scary (false positive) security warning they can’t click through…. do not panic. They’ll tell you it takes 5 business days for your unblock request to process. They’ll tell you there’s nothing you can do. They won’t provide any reason for the flagging. That’s all wrong and avoidable!

Go to their upstream provider ( Cujo AI) directly and submit a website unblock request on their contact page. Within 60 mins you’ll be clear!

[Pythonics @pythonics00]

@Ask_Spectrum Thanks, conacted via DM!

[Pythonics @pythonics00]

@Ask_Spectrum Spectrum decided to randomly block our site on Spectrum Security Shield. Tens of thousands of customers are impacted and I haven’t been able to escalate for resolution! No changes have been made to the site recently. This just began happening. Please help!

[Pythonics @pythonics00]

@Ask_Spectrum I’ve already submitted. It says that it will take up to 5 business days. This is catastrophic for our business and customers.

[Pythonics @pythonics00]

@coreyepstein @turkishtea__ @IzzyatHQ Tbh I have seen HQ around for a while but your messaging is really unclear. I’m still not sure what you do but I’m really clear about glass from just a few mins of reading.

[Pythonics @pythonics00]

@zeeg Damn their website pisses me off.

[Pythonics @pythonics00]

Ok but if you’re only eating one slice you’re definitely not shipping enough; so I appreciate the parallelism but let’s get serious.

[Lenny Rachitsky @lennysan]

Tired: Two-pizza teams

Wired: Two-slice teams

@danshipper on the best team structure in the AI-era

[Pythonics @pythonics00]

We do this exactly. Pleased with the results! It’s a great filter for candidates familiarity with agent usage as well. We time constrain it pretty heavily and those who can’t leverage agents properly are revealed pretty quickly, complain about its difficulty and opt out quite readily.

[Pythonics @pythonics00]

@GeoffreyHuntley @tetsuo_cpp Literally an intern that we converted to a full hire while he’s still in college outships senior SWEs from TopTal who just began adopting this year and is nearly our top performer. In fact, he has done quite a bit of education for them and is managing their projects.

[Pythonics @pythonics00]

Although a shortage of capital may starve most startups, a failure to properly deploy capital is actually worse.

[Pythonics @pythonics00]

@karrisaarinen @linear Enable gang prompting please

[Pythonics @pythonics00]

Can’t save everyone. I’ve spoken to 8 seniors in the past week while hiring (seeking $250k+ comp btw!) who still believe agents can’t write code beyond basic boiler plate.

Many SWEs entire value was built around writing code. They were never truly technologists and never fell in love with engineering. The world has moved out from under their feet with this massive increase in agent capabilities and they can’t comprehend how useless their ability to write code has become.

To me, that is surprising, because it has always been the case that you need to stay current in SWE and learn rapidly. The second you stopped learning, your career began to die.

There are plenty of motivated juniors who haven’t been able to find employment yet who will learn & will take their place.

[Pythonics @pythonics00]

@RhysSullivan Purely for “oh shit prod is down please read logs and fix”

[Pythonics @pythonics00]

Re: OpenAI usage cuts.

Anyone who has purchased enterprise SaaS or bought a significant amount of API credits knows you *never* pay list price. In fact, 50-80% discounts off of list are common. If they’ve made it a company priority to increase API spend, a natural lever to poor is to simplify pricing structure.

Less one off discounting to move API credits = more usage & more spend.

The other side of this is the in-bulk discounted Pro 20x plan. This is the first time usage cuts have been publicly addressed… but long time pro subs know that usage per dollar has always been trending down with API prices.

GPT-6-sol & Luna (although, I think these are clear losers compared to opus 5.5) will still feel unlimited.

GPT-6-Astra is however totally unaddressed & the only model that holds a candle to Opus 5.5 & Fable 5.1. I’d expect a 6.1 increment across the board with 50% or more cuts to API prices for 6.1 Astra as part of dev day announcements… otherwise the pain will be felt for a little while.

[Pythonics @pythonics00]

Usage cuts will only hold if there’s truly a supply constraint. This is overall a good sign for AI adoption and progress.

The more interesting question: is that supply constraint due to private / consumer usage or is OpenAI setting aside more inference for the US Gov & OpenAI?

[Pythonics @pythonics00]

@thorstenball Exactly.

https://t.co/3aqvZ4gW8Q

[Pythonics @pythonics00]

Tired: “Claude/Codex: Write unit tests, integration tests, and E2E test to cover these scenarios”

Wired: “After each step of implementation, leverage computer use and/or Chrome browser automation to verify across all auth scenarios and to protect against edge cases. Keep a report as you collect the useful test suites to add!”

Tests go from tautologies to actual verifiers.

[Pythonics @pythonics00]

@JoshKuechly @nvidia @Meta @OpenAI TypesafeAI is way ahead of this bad take. They’ve been in stealth for two years. You don’t think they’re keenly aware of incumbents, the game theory of Jev, and what winning their GTM looks like?