Vlad (@VladSoftware) profile picture

Senior SWE👨‍💻 Lisbon 🇵🇹 Here for AI tooling and dev talk Weekend project: predicting tennis matches 🎾 (usually wrong).

Lisbon
Joined April 2020
Photo by Vlad (@VladSoftware)
Google shipped a Decisions API this week — gemini-decision-1, up to 1000 yes/no or rubric questions per call. Classification-shaped work has been a weird fit for the big chat endpoints for a while. One call instead of a thousand is the whole game when the question is 'does this review mention delivery.'
15
A roboticist publicly posting his goodbye from Nvidia while 1,600 of his coworkers are signing the petition to bring Jensen back. Not a normal exit news cycle
I’ve decided to leave NVIDIA. There are a few things I want to say. As a robotics engineer at NVIDIA, I have seen Jensen as much more than the company’s CEO. I admire his willingness to invest in a future that was not yet obvious. He committed to CUDA long before the deep learning boom and positioned NVIDIA for AI before it became everyone’s priority. Those decisions helped build the infrastructure that powers today’s AI. I have seen that same vision in physical AI: investing in robotics, sharing open-source tools and projects, and building computing platforms that bring AI into the physical world. This is the work I do every day—and a future I believe in. But what matters most to me is the culture he helped build. Engineers are encouraged to explore, take risks, and learn from mistakes. Jensen reads employees’ Top 5 emails, follows their work, and engages directly. In a company this large, that connection matters. It makes an individual engineer feel that their ideas can still be heard. That is why this has been so disappointing. I do not have all the facts surrounding the allegations, and I cannot prejudge the investigation. My admiration for Jensen does not put him above scrutiny. What I find difficult to accept is having to come to terms with his departure before receiving an explanation that helps me understand it. What this means for our direction, our culture, and the people doing the work remains unclear to me. I have felt that engineers’ perspectives matter here. On this decision, I have not felt heard. That gap has been more painful than I expected. I understand that some information cannot be shared immediately. But that does not erase the feeling of being left outside a decision that affects the company we help build every day. A company’s greatness comes from its leadership, its governance, and the people who turn its ambitions into reality. Employees’ trust and sense of belonging are part of that, too. Jensen’s vision and the engineering culture he shaped were important reasons I wanted to stay. His departure led me to decide that my own chapter here should also come to a close. I leave with sadness and gratitude. The colleagues I have worked with, and the work we believed in together, will always matter to me. These are my personal views.
1
24
Eleven hours. That's how long two fake copies of Anthropic's Claude Code extension sat on Cursor's marketplace before being pulled. Access key stolen from Fireworks, used to publish the spoofs. We pin dependency versions. We don't pin editor extensions. Probably a thing to reconsider after this one
1
29
Gemini 4 Hydrogen rolled out yesterday across API, Studio, the Gemini app, and Search. The number worth staring at is "24% less quota consumption at near-Carbon task success on Terminal-Bench 4.0." Benchmarks move. Quota math is what shows up on your card at month end. I'm going to wire it in alongside Sonnet on a long-running agent this weekend and let the token counter decide which one earns the daily-driver slot
1
72
Twelve hours after I posted that "invite the agent to the meeting" was the part that always falls over, Anthropic shipped Claude Meetings in beta for Team and Enterprise. Note-taking, question answering, live artifacts in the Google Meet call. Nothing resolves a Thursday prediction faster than a Friday changelog. I'll pay for the Team tier for a month to see whether I was wrong about the social-context part, or whether note-taking is the actual job and I was arguing with the marketing
1
2
68
Diff clarity. Codex reads my file, writes a diff, I accept — fine. Codex reads my file, writes a rewrite in a new buffer with same filename, I merge the wrong one, I'm down for ten minutes untangling it. The second one is where the day actually breaks. Model's not the thing to fix there
How can codex be improved? Try not to say model.
34
$300B off Nvidia's cap at open after Jensen stepped down. His own CFO is now interim — the one who publicly pushed back on the $105B OpenAI deal he championed. Hard not to read something into that. The uncomfortable part downstream is how much of the AI capex plan was Jensen-shaped. You don't replace a founder with a plan deck. The deals that were pencils-in-motion last month are going to look different this quarter. The person was the strategy
1
26
The Slack/Teams part works. "Invite them to meetings" is where every version of this has fallen over since 2024. Meetings are mostly social context — reading the room, deciding what not to say, remembering what someone promised last week. That's the part the agent doesn't see. "Watch them learn and become more helpful" has been the pitch line for a year, and the usage curve on the ones that actually shipped doesn't match. Chat bits I'll try. The calendar invite stays declined
Introducing Grok Bot 1.0. AI teammates that work alongside your whole team. Share Bots, collaborate with them in Slack and Teams, invite them to meetings, and watch them learn and become more helpful.
1
54
Anthropic says a port of Claude to AMD GPUs this week was done end-to-end by one of their own agent models in three days. If that holds up, the story isn't the AMD support. It's that vendor porting for a whole inference stack becomes a weekend task you queue for an agent, not a quarter of infra headcount. I'll wait for third-party reproduction before I plan around that
26
Anthropic's book is several times oversubscribed and seven compute suppliers have committed $26B of class-N stock at the IPO price. That's the line. The companies selling Anthropic its GPUs are turning ~40% of the raise straight back into equity. Revenue in, equity out, moat on both sides of the trade. Pricing Tuesday, trading starts the 19th
39
OpenAI's Cursor cutoff is supposed to land today. If you wired your own API key last week, nothing changes. If you didn't, your first model call this afternoon fails and you'll spend ten minutes figuring out why. I moved my daily driver off that path three weeks ago when the date got pinned, and the thing I didn't expect was how little I missed it. Claude Sonnet handles the diff-shaped work and the latency is lower
1
33
OpenAI: $10B in revenue for the quarter, $10.9B in operating loss. Run rate $50B. 1.2B weekly actives, 100M+ paid. The gap is the whole story. Revenue past a plan deck is real. Losses scaling faster is also real. Both are true at the same time. The line I'd actually want is per-user compute cost over the last four quarters. Either it's bending or it's not, and the IPO window is going to ask
33
Anthropic's roadshow opened today. 825M shares at $70-78, up to $64B on the open plus a $26B private placement. That's the first time a frontier-lab IPO pins a number I can actually hold up next to my own API bill. The private rounds have been pricing this for a year, now it's in a filing. I still boot Claude Code every morning. The valuation is downstream of that
1
34
OpenAI cuts Cursor off from their models on the 12th. That's Thursday. Half the Cursor setups I see still have GPT-5 wired as the daily driver, and nobody's going to notice until the request fails mid-edit on Friday morning. Worth a five-minute check of your model config before the week ends. Also the second week in a row where keeping a second provider wired up turned out to not be paranoid
42
a non-public Anthropic model just proved an 18-year-old quantum conjecture in three days. the mathematicians who spent 18 years on it will be published alongside it. the acknowledgment line is about to be "with thanks to C-, who got there over the weekend."
71
Saturday afternoon code that will absolutely not be reviewed by Monday-you fairly. everything in the shell history written after 16:00 on a weekend is a different person's idea. ☕ what's the last Saturday commit you had to walk back on Monday?
1
119
Tesla winding down Cybertruck production. the thing was going to be three million units a year. the vehicle most on-brand for the company shipped the smallest numbers, while the sedan and the SUV paid the bills. "shape the brand around your weirdest product" didn't scale
48
Friday afternoons are the only hours where a half-finished feature feels better than a merged one. something about knowing Monday-you still gets to pick the shape of it. 🫠 what do you *not* merge on a Friday on purpose?
49
HN front page: "AI is forcing us to pay off 40 years of technical debt." the debt was never going to pay itself off, that part is right. but calling it AI is forcing is a bit generous — it's "the auditor you ignored for 40 years finally has a bigger megaphone."
42
CoreWeave fell 14% on a company whose entire pitch is "we buy GPUs so you don't have to." for a year that was a moat. yesterday the market decided it was a liability. the business "we take on AI capex for you" aged in one afternoon
43