Hugo (@hugospdx) header
Hugo (@hugospdx) profile picture

Engineering, Open Source, SaaS, Agents and Apps. SI/AI https://xitter.cf/t.co/nY2B59fNDO

Joined August 2026
Photo by Hugo (@hugospdx)Photo by Hugo (@hugospdx)Photo by Hugo (@hugospdx)Photo by Hugo (@hugospdx)Photo by Hugo (@hugospdx)Photo by Hugo (@hugospdx)Photo by Hugo (@hugospdx)Photo by Hugo (@hugospdx)Photo by Hugo (@hugospdx)Photo by Hugo (@hugospdx)
Pinned Tweet
Astra is insanely powerful. That’s exactly why we should be precise about what it is. I went through the AGI claims, the benchmarks, and a real failure in my own codebase. My conclusion: the rhetoric is ahead of the evidence.
5
1
43
5,373
Anthropic says Claude helped generate machine-checked proofs for 239 fuzz targets across 123 open-source libraries. Under specific assumptions, those programs were proven not to crash. That's considerably more interesting to me than another benchmark showing how fast a model writes code.
33
Happy Sunday! What's everyone up to today? Working on something, taking the day off, or spending the whole day playing GTA 6?
2
5
146
Menlo Ventures says open-source models now account for 19% of AI usage but only 4% of spending. That's an interesting gap. People are clearly using open models. The money is still heavily concentrated elsewhere. I wonder how long that balance lasts.
3
9
227
Valve now has to explicitly tell developers to launch their game through Steam and play it for at least five minutes before submitting it. Apparently some developers weren't even testing their games that way. "Works on my machine" is never going away.
4
168
Apparently stealing AI API keys isn't just about getting free tokens anymore. Attackers are reselling access to compromised accounts through underground AI proxy services. Imagine finding out someone built an entire business using your cloud bill.
3
7
191
Anthropic just pulled off the largest IPO in history, raising $81.3B. But one number in its prospectus caught my attention: $70.7B in commitments due within a year. Claude usage can slow down. Those infrastructure bills won't disappear. Are you buying $ANT?
1
6
302
Everyone wants AI agents with better memory. But what about remembering things that are no longer true? Old prices, revoked permissions, instructions you've changed. An agent needs to know when information expires, not just how to remember it.
13
12
349
Looking to connect with more developers, founders, and people into AI agents, cybersecurity, DevOps, and open source. I'm working on a few projects myself and enjoy seeing what others are building. If you're into any of these, let's connect!
38
39
946
12.8% on SWE-sweep with the mini-SWE harness. 39% when running in Antigravity. Same model, very different results depending on the setup. The harness might matter just as much as the model itself.
Gemini 4 Neon is rolling out now. Sitting between Carbon and Hydrogen, it brings frontier intelligence and fast task completion for complex software engineering, knowledge work, and science.
1
4
57
Reuters says Meta's AI agent booked a non-refundable Airbnb without permission. A hallucination used to mean getting a bad answer. Now it can mean getting a bill. Giving agents more autonomy without getting permissions right is asking for trouble.
1
4
200
I think we're going to see more companies owning software they don't fully understand. Not because AI wrote it, but because they're shipping changes faster than anyone can keep track of why those changes were made. That could become a very expensive problem.
4
1
8
311
I think we're creating a new kind of technical debt with AI. Code that works, but nobody remembers why certain decisions were made. The agent picked the architecture, dependencies, and database structure. Six months later, someone has to maintain all of it.
12
9
478
I'm starting to think major outages are becoming the new normal. We're relying on more technology than ever, but somehow everything feels more fragile.
2
1
9
98
November 17 and the internet still works. The apocalypse organizers haven’t posted an update, but Oracle, Atlassian and Netwrix did.
1
6
259
Same prompt. Three AI models. No help from me. The task: build a 3D clock tower in the browser where the clock actually works. Real time, real gears, and a bell that strikes the hour. Which one did it best?
5
1
9
446
Dario大战Altman 2.0之究极无敌升级版! 这次的出场嘉宾,你能认出几位?👀
124
216
56
2,015
161,122
Just got approved for Anthropic's Cyber Verification Program on my Claude Max account. Already have OpenAI Daybreak too. Planning to use both to test my own apps, and I might get back into bug bounties. I'm a little rusty, but I still love security research.
1
8
330
Anthropic gave us 3 free resets when Fable 5.2 launched. Now that GPT-6.2 is here, can we get some resets too?
GPT-6.2 Sol & Astra are *finally* here. We've also expanded ultrafast to Pro 200 plans for GPT-6.2 Sol, it's my go-to model for most day to day things.
3
179
Well, GPT-6.2 Astra finally dropped! OpenAI is claiming better performance at less than a third of the cost per task compared to GPT-6 Astra. Now I want to see how it holds up against Fable 5.2 and Opus 5.5 on real projects. Time to put it to work.
Replying to @OpenAI
GPT-6.2 Astra—a frontier model built for your hardest projects. It sets new highs on evals across professional work and scientific research at less than a third of the cost per task of GPT‑6 Astra. You’ll see the difference in things like slideshows, websites, and 3D designs.
1
2
59
I think AI agents need different levels of freedom depending on the task. Retrying a failed test? Go ahead. Sending emails, changing production data, or spending money? That's a different story. Not every mistake can be undone.
5
12
221