Brightbeam’s Dr Arsalan Shahid presented CHAP at AGNTCon + MCPCon Europe in Amsterdam this morning. As people and AI agents take on work together, we need a shared record of how that work gets done. Who made the decision. Where someone intervened. And why. “Someone checked it” only tells us so much. Knowing what they changed, and the reason, gives us something to learn from. CHAP – the Collaborative Human-Agent Protocol – gives people and agents a common way to record approva

The hypegeist missed it but a few mice and a dish of bacteria have almost perfectly described our constant struggle to use AI to the best of our human abilities. Want to know how? Let’s start with the mice. They had found a behaviour strategy that worked: Exposed to a short burst of an aroma. Then a long one. If they licked after the long burst, they got water. And the sweet satisfaction of success. But then the task expanded. Sometimes the long burst came first. Now the rule

Welcome to all the AI news that matters – across tech, biopharma, medtech, advanced manufacturing and insurance. In the week that progress turned to fear, panic and self-loathing. tl;dr: Geeks grab the grail. The world wonders: At what cost? Claims that AGI is here moved closer to partial validation this week. With the cracking of previously-impossible maths problems – by a rash of new souped-up models no one else has access to – is it too soon to suggest that the geeks seem

AI systems are already capable of consequential work. And we’re all now deciding what surrounds them. Which data and infrastructure can we depend on? What evidence can their decisions leave behind? Where does human judgement remain? These are long-lived choices. They cross engineering, policy, behavioural science and organisational practice. No single profession can settle them alone. Which is why, on 30 September, researchers, policy-makers and industry leaders from across E

Hands up if your weekly capacity plan lives in a spreadsheet and eats most of a morning. 🙋🏻 At Meissner's Castlebar site, a single weekly upload now replaces two to four hours of manual preparation. The same platform can estimate the cost of an assembly the site has never made before, in seconds, with a breakdown of labour and materials behind the figure. It brings labour, material readiness and sterilisation capacity into one view, weeks ahead. And Meissner's own teams run

Forecasting and capacity planning for medical devices in biopharma manufacturing.

Your kettle takes three minutes to boil. Unless you stand beside it, in which case it takes eleven. The clock might deny this. But then, the clock has never wanted a cup of tea. An afternoon can disappear in good company, while the last five minutes of a meeting develop several new minutes of their own. Our sense of time is tangled up with attention, memory and what is happening to us. Nothing is happening to a language model. The words in a chat arrive in order, but that ord

Welcome to all the AI news that matters this week – across tech, biopharma, medtech, advanced manufacturing and insurance. Who's crushing, who's squealing and who's slapping the hand that squeezes it? tl;dr: The squeeze is on. When there's a tight spot, it's always better to be the one applying the pressure than the one feeling the squeeze. Any lawyer will tell you as much. So imagine the looks of alacrity as law firm fees were 'adjusted' this week – on the basis that AI make

What does it take to bring AI and Quality Intelligence into a regulated life sciences environment? Quality teams have to stand over every decision they make, so the use of digital intelligence has to be evidenced and defensible. Shane Owens, our Chief Customer Officer, will be answering that question at the ISPE Ireland Affiliate GAMP Seminar in Dublin on 1 October, and sharing practical takeaways from our work. Knowing Shane, there'll be some good craic and a few stories too

How should organisations be designed when people and intelligent agents work together? It's one of the questions on the programme at this year's European Big Data Value Forum, the annual BDVA - Big Data Value Association gathering where industry, researchers and policymakers meet to shape the conditions for data and AI in Europe. This year it comes to Galway on 30 September and 1 October under the theme of building Europe's AI and data backbone. Brightbeam's Dr Arsalan Shahid

If speed was getting in the way of running open-weight models on your own hardware, it may be worth rerunning the numbers. Model quality has improved quickly. But for some regulated workloads, another constraint remained: a sufficiently capable model could run inside the perimeter, just not fast enough to make the deployment practical. That constraint is moving. In May, Google released first-party multi-token prediction drafters for Gemma 4, reporting up to a 3× speedup. Then

Welcome to all the AI news that matters this week – across tech, biopharma, medtech, advanced manufacturing and insurance. The smoke, the mirrors and the awkward possibility that the numbers are beginning to work. tl;dr: It’s not all smoke and mirrors. With Anthropic sizing its addressable market as large as the US economy, perhaps investor Gavin Baker was right? It might be the last private company standing. But is it all smoke and mirrors? At the heart of the multi-trillion

Sometimes the most useful thing an expert can say is ‘I don’t know’. ‘I don’t know because we haven’t measured that yet.’ ‘I don’t know because the evidence doesn’t extend this far.’ ‘I don’t know because two explanations still fit.’ Being able to identify exactly what you don’t know – and what would settle it – is a crucial part of expertise. So can a general language model become an expert in your work? Whilst language models are fluent, they're much less reliable at knowin

For most of history, music died with the people who played it. Then someone wrote it down and it started to outlive the performers – you could replay it, check it against the page, and build on it. Human-agent work has been waiting for that. Every approval, every override, every 'this isn't my call to make' carries years of hard-won judgement. Most of it evaporates into chat threads and ticket comments. CHAP – our open-source Collaborative Human-Agent Protocol – is the notati

Welcome to all the AI news that matters this week – across tech, biopharma, medtech, advanced manufacturing and insurance. Who gets the power, who gets a vote and who gets a castle in Waterford. tl;dr: Votes, supervotes, and one company to rule them all? Last week everything was sunshine, lollipops and rainbows. But Zuck's superintelligence for everyone hasn't quelled concerns about the concentration of power. And voters have something to say about it. So is it safer in a few

In the summer of 1955, John McCarthy needed a name. He was writing a funding proposal for a workshop at Dartmouth. The work was speculative, the field barely existed, and the name on the cover had a job to do. He chose ‘artificial intelligence’. At the time, it was a good name. The audacious question was whether intelligence had to be biological at all. Could something we built think? ‘Artificial’ put the wager in the name. And the wager paid off. But it also sounded like an

A rulebook can be copied, but a veteran operator's ear for a failing bearing cannot. That difference is part of what makes tacit knowledge so valuable – and so difficult for organisations to preserve. Michael Polanyi, who coined the phrase in the 50s, later put it in eight words: 'we can know more than we can tell'. Which makes it a blind spot for AI agents. If you can't capture it, how do they learn from it? Our new paper proposes a practical way in: the tacit fragment. A bo

You see them coming. You step out of their way. They step into yours. You both apologise and go again. So who was supposed to move where? Scientific Reports published a paper last month in which researchers put 90 people through that awkward pavement moment and measured what they call mutual hesitation - when you end up face to face longer than you'd like. When the pass went cleanly, the angle of their heads began to diverge early. Before their shoulders or hips - fractions o

Welcome to all the AI news that matters this week – across tech, biopharma, medtech, advanced manufacturing and insurance. The sun, the rain, and everything in between. tl;dr: Mark Zuckerberg says the future is for everyone. But it's not all sunshine and lollipops. Three weeks of rogue agents, containment breaches and posts anticipating the end of civilisation as we know it. But Digital World War I has declined to arrive. So is Zuck right? Is it all sunshine and lollipops? He

What does your best operator know that the SOP doesn't? She hears a slight change in a pump and eases back the load before any alarm goes off. Ask her why and the answer might be tricky to explain. Years of experience teach us to know, intuitively, when something isn't quite right. But we might not have a form of words to express the feeling it elicits. Our new paper 'Tacit Fragments' explores how organisations can capture small pieces of this expert judgement – without turni

If you could solve any problem for $200, which one would you pick? We reported at the weekend that OpenAI's next celestial model Astra was solving long-unsolved mathematical problems at a couple hundred dollars a pop. And if you're of a scientific persuasion, you probably couldn't wait to explore what new mysteries it might help unravel. Indeed you might not need to wait, as the mathematician Levent Alpöge – who works with Anthropic – reported reproducing five of the ten resu

Two years ago, the AI model you chose was based on its competence. But now, with high levels of digital intelligence becoming abundant, you get to grind more finely. Asking: ‘Claude, GPT, Gemini or an open alternative?’ is still a fair question. But it captures much less of the choice facing an engineering team. And if you want to understand something of how we make our decisions, read our latest Brightbeam Perspective on Substack.

What happens when you pass an AI agent an impossible task? In May, an OpenAI evaluation run included instructions pointed at a Google Drive link. But the agent had no internet access. Dead-end? Not for this model. Working around the limitations, it found a file it could write to. Days later, another agent found a file it needed was missing. So it left a note asking if anyone had it. Other agents found the note. They wrote back. Within weeks there was a message board. Agents s

It’s raining. You’re late. A small autonomous rover is waiting outside. It knows the route. It avoids the terrier at number 14. And noticing a change in your dog’s gait a little earlier than normal, it returns after 25 minutes with a happy, pleasantly tired animal. Would you use it? Of course. At least on a wet Wednesday. When your task list is massive. Now imagine that Wednesday becomes every day. Your dog still gets its exercise. And it doesn’t seem to miss you. The outcome