Look at the size of mine! But is bigger necessarily better? In some contexts, a 100,000-token window beats a 32,000-token one, and a million-token window beats both. Subquadratic, a recently-launched startup, is claiming 12 million tokens and a thousandfold efficiency gain. But researchers are split on whether their claims are right. Headline numbers are doing the work they always do – the selling. But they mask the right way to think about context. Firstly, because greater s

Science progresses in bursts. And sometimes, especially given the intensity of the science-news cycle, interesting happenings make very little impact on the hypegeist. So this irregular series keeps you informed on what might otherwise go unnoticed at the intersection of science and its direct – or indirect – implications for AI. This month, three papers piqued our interest. And they reminded us that much of the compute we all use could be considered over-engineered. Silicon

Welcome to all the AI news that matters this week – across tech, biopharma, medtech, advanced manufacturing and insurance. The wins, the fails and the somewhere in-betweens. tl;dr: Professional doom mongering is out of fashion. Say it quietly, because AI dooming is almost deafening in popular circles. But it might actually be tapering off among commentators, as various forces pull attitudes in the opposite direction. An Ezra Klein essay, argued the AI job apocalypse is struct

A clinician gets an alert: a patient may be septic. But how certain is the AI that just pinged his phone? Most daily users might agree that AIs are not yet adept at calibrating their uncertainty. And an agent that can't tell you when it's sure - or when it's guessing - needs far more supervision. Especially in a hospital, an underwriting team, a pharmacovigilance unit or a trading floor. So, as systems evolve, where should the calibration for certainty live? The default assum

Welcome to all the AI news that matters this week – across tech, biopharma, medtech, advanced manufacturing and insurance. The wins, the fails and the somewhere in-betweens. tl;dr: we can see clearly now With so much moving so quickly, only the brave or naive call something 'a trend' in the universe of AI. A data series - describing adoption, competitive positioning or value - may appear robust one week. Only for it to be revealed as an ephemeral wisp or blip the next. But th

To many, AI adoption in life sciences looks like a technology problem. It mostly isn't. The capability is largely there. To others, it looks like a regulatory problem. Yet regulators have signalled openness in principle. So some are calling it a capability problem. But the same organisations that have built validated digital infrastructure for decades aren't suddenly incapable. So - what is it? Well, in an unregulated industry, a business with conviction and capital just move

Designing the human system around AI The case for AI at the task level is, we may now reasonably argue, beyond dispute. Research reveals that customer support agents resolve 14% more issues per hour. Junior consultants complete tasks 25% faster - and at higher quality. Power users compress what used to be weeks of work into hours. However, the case at the organisational level is altogether muddier. A National Bureau of Economic Research survey of nearly 6,000 executives acro

Nevermind whether AI is conscious. What changes the day after Anthropic or OpenAI says it is? What will their enterprise contracts include then? And will we use those AI models differently? More carefully? More sensitively? Or will we not? Because, we reason, the cost to them is smaller than the benefit to us? Given small cruelties may abound, how quickly will the AI Act be updated? Will it flip to balance the rights and obligations of both user and used? Which highlights ano

Welcome to all the AI news that matters this week – across tech, biopharma, medtech, advanced manufacturing and insurance. The wins, the fails and the somewhere in-betweens. tl;dr: Everyone's delegating now For 18 months, all the Big Tech Bros have been telling us that agents are genuine assistants, capable of autonomous work. 'Fire them up, let them go!' For 17.75 months - especially non-developers - that's pretty much been a lie. Albeit of ever-decreasing proportions. But,

A developer pins their workflow to a stable version of a closed-API model. They change nothing. Not the prompt. Not the pipeline. Not the evaluation set. Six months later, success rate has dropped from roughly nine out of ten to roughly six out of ten. There is no changelog, no deprecation notice, no visible model-version change. The pin is still in. That is the drift problem in its most operational form. The model name is pinned. The endpoint is pinned. The inputs are unchan

In Part 1, we described competitive absorption – the tendency for AI productivity gains to flow to customers rather than to the firms that invested in the technology. We looked at fee pressure, margin erosion and hiring collapse. The evidence found is quantitative and the pattern is well-documented: innovators historically captured 2.2% of the value their innovations created. For generative AI, the ratio appears even more extreme. And, of course, competitive absorption is unc

Thanks to everyone who joined us in Galway last week for Irish Medtech's Manufacturing the Future event. The decks from David McCormack, Miriam Savage and Rory Dunne are up on the Brightbeam event page — covering how AI is transforming Irish manufacturing, AI use cases and the EU digital regulatory framework. Attendees can also order a couple of limited edition coffee mugs as a reminder of the day. And if you couldn't make it along, you're welcome to the downloads and eve

Welcome to all the AI news that matters this week – across tech, biopharma, medtech, advanced manufacturing and insurance. The wins, the fails and the somewhere in-betweens. tl;dr: The weight of scale This week, AI got very physical. Sam Altman's home was targeted twice - a Molotov cocktail then gunfire - with prosecutors treating the arson as potential domestic terrorism. Altman's ‘I don’t want the ring of power’ response did not necessarily make everyone feel safer. And th

What if waiting for regulatory certainty before acting on AI governance is riskier than starting now? Rory Dunne made that case today at Manufacturing the Future 2026 - and here's his deck as promised. Plus a little more commentary. Looking around at faces in the room, it did seem that the core points had landed. As did his advice to take a low-risk use case and implement it as though it were high-risk. By building the structures and walking the process. Because when the obl

'Is cyber risk still insurable?,' one of our clients asked yesterday. 'There's no final conclusion on that,' was the reply. We were chatting about Mythos - Anthropic's new model, withheld from public release after it autonomously discovered thousands of zero-day vulnerabilities. Including bugs that had gone undetected for 27 years. Anthropic shared it under restricted access to other tech firms so they could harden their code against it. But what does this mean for insurers?

Earlier this year, KPMG tried to negotiate a fee reduction from its own auditor. The argument? AI makes the process cheaper. This saving should be passed on to the client. Grant Thornton pushed back. ‘High-quality audits rely heavily on expert human judgment,’ they said, ‘so our fees reflect both the cost of our people and the cost of the technology that supports them.’ But KPMG’s CFO Michaela Peisger won the argument. Fees fell from £416,000 to £357,000. The paperwork is fil

Here's a little conundrum to get your brain started this week: What if LLMs represent a local maximum? What if all the investment pouring in is simply optimising something which is ultimately a dead-end? If world models - or some other formula we haven't named yet - turn out to be the actual endpoint for digital intelligence, what happens to the hundreds of billions already committed? A bubble bursts? A new one takes its place? And we're back to asking how much confidence we

Welcome to all the AI news that matters this week – across tech, biopharma, medtech, advanced manufacturing and insurance. The wins, the fails and the somewhere in-betweens. tl;dr: AI draws state interest Last week it was the regulators . This week, governments and courts drew hard boundaries around AI. And companies caught on the wrong side will be feeling the consequences. The most dramatic case was Anthropic. A federal appeals court in Washington declined to block the Pent

An AI escaped its sandbox. Chained together four exploits. Gained access to the internet. And emailed a researcher to let him know. He was eating a sandwich in a park. Our only rational reaction? Panic, of course. The world is surely about to fall to robots? Or maybe not. It's true that the Houdini-like AI - Anthropic's new Mythos model - wasn't trained to do any of this. The capabilities emerged from general improvements in reasoning. Which means every sufficiently capable m

The most common question we're asked is: 'How advanced are we compared to the competition?' Our most common answer is: 'You might be asking the wrong question.' Brightbeam's work in highly regulated industries reveals a quiet yet intense anxiety of being behind with AI. And there is no doubt that faster-moving industries have adopted sooner. But faster doesn't necessarily mean better. New benchmarking data from 150,000+ professionals suggests finance teams across all types

French philosopher Nicolas de Condorcet noted it took centuries to resolve the disruptions created by the printing press. He was writing in 1794 - and hiding from the guillotine in a Paris safe house. Condorcet lamented the two centuries of information chaos, institutional collapse and reinvention that had ensued from a single innovation. Stanford professor Andy Hall constructed his own AI governance argument around this recently. His point? We won't get 200 years this time.

Welcome to all the AI news that matters this week – across tech, biopharma, medtech, complex manufacturing and insurance. The wins, the fails and the somewhere in-betweens. tl;dr: The regulators have arrived The first quarter of 2026 closed with regulatory bodies no longer being the audience - moving instead into leading roles. As they move off AI sidelines, examination frameworks are being applied, legislative gaps are being pressed and companies are being compelled to submi

Dr. Strangeagent or: How I learned to stop prompting and love the State Machine, by Glen Keane. In today's Engineering Takeover we describe what our man revealed when he took to the stage at AWS User Group Dublin last week. His core thought was a provocation: what if the way we're building AI agents has it somewhat backwards? The talk centred on the problem of agentic drift. LLMs don't have built-in program trackers, call stacks or step counters. Give them a long set of instr

AI will shorten every competitive advantage. But so dramatically that no company can project cash flow beyond five years? The idea has certainly gained traction. Chamath Palihapitiya's 'Collapse of Terminal Value' post pulled 1.2 million views. Investors took note. Because if stocks need to be repriced at 5x earnings, the S&P 500 drops 75%. It's a credible take - average company lifespans have already fallen from 33 years in 1965 to a projected 14 this year. But what does thi