AI2Day Daily Brief — 8 Sept 2026

Good morning. It is Tuesday, the eighth of September, and the biggest story in our feed today is a humanoid robot company with one point eight million dollars in sales seeking a two point five billion dollar stock market valuation.

Agility Robotics builds Digit, a two-legged robot designed to work in warehouses and factories. Last year the company brought in one point eight million dollars in net sales against a one hundred and forty million dollar operating loss. It has now filed paperwork to go public through a SPAC merger valued at two point five billion dollars. That gap, roughly fourteen hundred times annual revenue, tells you where humanoid robotics sits right now. The deal is expected to raise more than six hundred and twenty million dollars, including two hundred million from Foxconn. Agility says it has logged more than sixty-five thousand operating hours across nine customer sites and projects twenty-five thousand robots deployed by twenty thirty-five. Those projections are Agility's own.

A filing obligation, not a business milestone. Worth keeping that in mind. Next, a robotics startup pushing back on the conventional wisdom that robots need more data before they can be useful in factories.

Inbolt is a French startup that builds software for industrial robots, letting them use computer vision to see where a part actually is on a conveyor belt rather than following a fixed pre-programmed path. Most factory robots miss the part the moment it sits at a slightly different angle. Inbolt's system adjusts in real time. The company says it has deployed across more than one hundred factories and logged over forty million robot cycles in production at sites including Ford, Toyota, and Stellantis lines. CEO Rudy Cohen's argument, which he will make at RoboBusiness in October, is that the real bottleneck is not data collection but the time and cost of getting systems installed and running on real factory floors. The company has raised around twenty-three million dollars to date.

A useful corrective to the lab-first framing. Now to a story from the UK, where patient trust in an NHS data system is becoming a measurable problem.

Tens of thousands of NHS patients in England have withdrawn their medical records from research use, and health innovation minister James Frith has pointed to Palantir as a specific driver of that distrust. Palantir, a US technology company, holds the contract to build and run the NHS Federated Data Platform, the central system for sharing patient information across hospitals. The company built its reputation doing data analysis for US military and intelligence agencies, and privacy campaigners have raised concerns about a firm with that background holding sensitive health records for millions of people. Patients in England already have the right to opt out of non-clinical uses of their data, and the new figures suggest a meaningful number are now exercising it. Frith warned that continued withdrawals could reduce the data available to medical researchers.

Today's sponsor is Zero2Hero: an AI poker coach sits at your table, and you can ask it mid hand whether the player across from you is bluffing. Every mistake you make gets flagged and graded automatically, hand by hand, even on the free tier. Lifetime access is $99, in US dollars, until 15 September, at zero, the number two, hero, dot A I.

That is a real tension: the same infrastructure intended to advance medical research may be shrinking the dataset it depends on. Staying in health, a working GP is raising a practical alarm about AI scribes.

Dr Mary Gibbs, an out-of-hours GP in England, says she finds meaningful errors far more often in AI-generated consultation notes than in notes typed by colleagues. AI scribes are software that listens to a medical appointment and automatically produces a written summary. An NHS watchdog has already flagged the technology for getting drug names and diagnoses wrong. Dr Gibbs says AI-written triage notes, the initial assessments done before she speaks to a patient, are consistently longer, repetitive, and sometimes self-contradicting. When she questions patients directly, she regularly gets a history that differs significantly from what the AI recorded. Her concern is not theoretical resistance to new technology. It is that she cannot trust the notes in front of her, and that unreliable notes make consultations longer and riskier.

Deployment speed without reliability is not a feature. Finally, if the last four stories have thrown terms at you that you half-recognised, this one is for you.

TechCrunch AI has published a plain-English glossary of the AI terms that come up most often and get explained least clearly. A few worth knowing right now. AGI, or artificial general intelligence, is the idea of an AI that can do most things a person can do at least as well. There is no single agreed definition even among the labs building toward it. OpenAI defines it as systems that outperform humans at most economically valuable work. Google DeepMind frames it as matching human capability across most cognitive tasks. AI agents are software tools that carry out multi-step tasks on your behalf, not just answering a question but taking a sequence of actions. Chain-of-thought reasoning describes a technique where a model works through a problem step by step rather than jumping straight to an answer. Knowing a handful of these terms makes it considerably easier to judge product claims and policy debates on their actual merits.

If you want the full week in context, the Monday show AI Today Weekly goes deeper on everything that matters. That is your AI briefing for today. Every story is at A-I-2-Day dot live. That is A, I, the number two, D-A-Y, dot live. We are back tomorrow morning. If this was useful, hit the thumbs up and subscribe, so the next one finds you.