Antoine Sauray

SWE @blablacar

[Hedgie @HedgieMarkets]

🦔Microsoft canceled its internal Claude Code licenses this week after token-based billing made the cost untenable, even for a company with effectively infinite cloud resources. Uber's CTO sent an internal memo warning the company burned through its entire 2026 AI budget in just four months. American AI software prices have jumped 20% to 37%, and GitHub (owned by Microsoft) is dropping flat-rate plans for usage-based billing across its products.

My Take
The AI subsidy era is ending in real time. The same company that put $13 billion into OpenAI and built the Azure infrastructure powering most of Anthropic's compute just looked at the bill from a competitor's coding tool and decided it was not worth paying. That is not a productivity failure on Anthropic's end. Token-based pricing is forcing every enterprise customer to confront the actual cost of running these models at scale, and the number turns out to be far higher than the flat-rate experiments suggested.

This ties directly to my Gemini Flash post yesterday. Anthropic, OpenAI, and Google all raised effective prices in the last six months. Enterprises that built workflows assuming AI costs would keep falling are now watching annual budgets evaporate in months. Two outcomes look likely from here. Either enterprises scale back AI usage to fit budgets, which slows the revenue ramp the labs need to justify their valuations ahead of IPOs, or the labs cut prices and absorb the losses, which makes the unit economics worse at exactly the wrong moment. Both paths land in the same place, the numbers stop working, and somebody has to take the writedown.

Hedgie🤗

[Aurélien Véron @aurelien_veron]

Il suffit de voir le nombre de parlementaires présents à l’audition du co-fondateur de Mistral IA pour comprendre que la France sera absente de cette révolution majeure.

[Antoine Sauray @asauray]

@kimmonismus It cannot pickup new languages it hasn’t seen, it’s just good at what it’s seen during training. That’s a plain good lie right there.

[Nick Huber @sweatystartup]

AI about to get 20x expensive.

These $200 / month claude subscriptions are burning $5,000 worth of credits.

The bubble is going to pop and it will pop soon.

[Legendary @Legendaryy]

Your Claude subscription is massively subsidized and it won't last forever.

A $20/mo Pro plan burns through ~$180/mo in API-equivalent tokens. Heavy Max users hit $5K/mo on a $200 plan.

Actual compute cost is roughly 10% of API pricing. Venture capital covers the rest.

Anthropic just 2x'd Claude Code limits for "Spring Break'" during off hours. Enjoy it while it's here.

At some point the math has to math.

Clip from the @modernmarket_

[Pranit @Pranit]

Anthropic just pulled the oldest trick in SaaS pricing.

I pay $200/mo for Claude Max. My limits have been noticeably worse this past week.

Now they announce 2x off-peak usage for two weeks. Sounds generous.

But here’s what actually happens: limits quietly drop, a temporary 2x makes the reduced limit feel normal, the promo ends, and you’re left at a baseline lower than where you started. You just didn’t notice the downgrade because the 2x absorbed the transition.

These AI plans are massively subsidized. The raw compute behind a heavy user costs multiples of the subscription price. Every move like this is the subsidy quietly correcting.

Very sneaky, Anthropic.

[Rohan Paul @rohanpaul_ai]

New Harvard Business Review research reveals that excessive interaction with AI is causing a specific type of mental exhaustion ( or AI brain fry), which is particularly hitting high performers who use the tech to push past their normal limits.

A survey of 1,500 workers reveals that AI is intensifying workloads rather than reducing them, leading to a new form of mental fog.

While AI is generally supposed to lighten the load, it often forces users into constant task-switching and intense oversight that actually clutters the mind.

This mental static happens because you aren't just doing your job anymore; you are managing multiple digital agents and double-checking their work, which creates a massive cognitive burden.

The study found that 14% of full-time workers already feel this fog, with the highest impact seen in technical fields like software development, IT, and finance.

High oversight is the biggest culprit, as supervising multiple AI outputs leads to a 12% increase in mental fatigue and a 33% jump in decision fatigue.

This isn't just a personal health issue; it directly impacts companies because exhausted employees are 10% more likely to quit.

For massive firms worth many B, this decision paralysis can lead to millions of dollars in lost value due to poor choices or total inaction.

Essentially, we are working harder to manage our tools than we are to solve the actual problems they were meant to fix.

---

hbr .org/2026/03/when-using-ai-leads-to-brain-fry

[Antoine Sauray @asauray]

@DamienToscano @DFintelligence @SNCFVoyageurs C’est probablement pas la même chose: tu peux utiliser http ou ssh pour git, de nombreux réseaux public bloquent ssh. Ici c’est http qui est utilisé, ce qui veut plutôt dire qu’il y’avait soit des problèmes réseaux, soit une mauvaise configuration.

[Antoine Sauray @asauray]

@cryptopunk7213 Pause on hiring will have the opposite effect five years from now, when lack of seniority will be an issue to keep systems running and debugging hard problems that llm struggle with (and likely will)

[Antoine Sauray @asauray]

@lennysan @bcherny Even coding is not solved, and it’s actually noticeable on projects where documentation exists but with low training data. Try to write a https://t.co/ERrXonx6ps pipeline with any llm, and you’ll see.

Shows that llm have a hard time generalizing outside of training data

[Benjamin Code @benjamincode]

J'adore le vibecoding.
Mais il perturbe complètement notre relation avec notre travail.

Il faut tout réapprendre.

Avant vous étiez fier de vous quand vous aviez produit en une journée même pas un dixième de ce que vos agents vous produisent en 1h.
On pouvait fermer les ordis pour la soirée, la pause était méritée.

Aujourd'hui, on s'arrête plus.
On produit 100x plus vite avec la sensation qu'on peut toujours plus.
Il n'y a tellement plus de friction que ne pas faire bosser vos agents semble être du temps perdu.
Ça prend 3min de de la lancer sur une tache. Puis 3 min de checker que tout va bien. Puis encore 3 min de lancer une deuxième tache en parallèle.

Avant on produisait jusqu'a la limite de nos capacités, on était satisfait du ratio effort/output, et on était aussi épuisés. C'était trois conditions qui nous faisaient converger vers un arrêt naturel.

Le vibecoding a fait sauter les trois en même temps. Notre signal d'arrêt naturel a disparu.

Dernièrement je sens au fond de moi que quelque chose déconne.

Quand continuer ne coûte plus rien, comment on mérite encore de s'arrêter ?

[Antoine Sauray @asauray]

@raphael_lg_ @NVIDIAGeForceFR Pas eu le temps

[Antoine Sauray @asauray]

@raphael_lg_ @NVIDIAGeForceFR Le pc démarre pas donc pas besoin de gpu pour l’instant 😂

[Léαlinux 🐧 @lea_linux]

Gamin de 15 ans : fait un téléphone, du hardware au software, donne le tout en opensource.
Les gens dans le thread : mouais bof, tel bidule est pas forcément au point, et il manque ça et puis cela.
... c'est désespérant...

[Antoine Sauray @asauray]

@mckaywrigley Yes like the agent is coding on its own 24/7

[Antoine Sauray @asauray]

@dodorm69 @sebi2706 @gchampeau Les deux ont quand même un intérêt à prétendre être en bas de la courbe pour attirer des investisseurs et du public.

[Antoine Sauray @asauray]

Docker no long sunsetting free plan. As we approached the deadline, many probably uninstalled Docker to avoid the paid plans. Those are likely not returning though.
https://t.co/TMNfERzMc8

[Antoine Sauray @asauray]

And suddenly software became reliable around the world as most people stopped deploying

[Ty Smith @tsmith]

"Kotlin is now the recommended programming language for server-side JVM usage at Google, set to replace Java while still providing access to a large existing Java ecosystem."

https://t.co/KpIqEwYzpR

[Nikita Bier @nikitabier]

Making fun of BeReal for slow load times during user spikes is just another form of American cultural imperialism. French code just runs slower.