GAEA Talks

This week on GAEA Talks, the format flips. Graeme Scott, the man who normally asks the questions, is in the hot seat. Guest host Georgie Barrat - one of the UK's leading technology journalists - interviews him on the topic he is most associated with: the real meaning of sovereign AI.
Recorded live in the new London studio, this is a long-form, single-topic deep dive. Sovereign AI is one of the most-used and least-understood phrases in modern AI, and Graeme has been making the case for over a year that almost everyone in the public conversation is talking past each other. In this episode he sets out what he believes sovereign AI really means, why the token economy is a trap, why local and private AI is already shipping inside Apple and Google's stack, and why the UK is uniquely positioned to lead the next AI wave.
Topics covered:
What sovereign AI means to a person, a company, a multinational and a country
The recent US restriction on a major foundation model and what it taught the industry
Why the token economy is costing the UK more than people realise
The Apple, Google and AMD hardware play that nobody is talking about
The three stakeholders shaping the AI narrative and what each one is incentivised to obscure
The cognitive decline research starting to emerge from heavy LLM use
The three questions every leader should be able to answer about their AI architecture
The worst case and best case scenarios for the average person in the next three years
Why the UK's innovation DNA from Stephenson's Rocket to Alan Turing makes this the country's moment

Why does drug discovery need to go to space? In this week's episode of GAEA Talks, the second of season five and the second filmed in our new London studio, Graeme Scott sits down with Dr Toby Call, founder of Mass Balance, former co-founder of Chronomics, and alumnus of the International Space University.
Toby explains why over forty percent of the proteins in the human body have no fixed structure, why this "dark proteome" includes some of the most important targets in cancer and Alzheimer's, and why AlphaFold and the rest of the current AI drug discovery stack cannot model them. He then makes the case for microgravity in orbit as the next great forcing function in biology, places that argument inside the wider story of the second space race, and brings it back down to earth with a clear-eyed view of sovereign AI, edge compute, and where the UK has to position itself over the next twelve to twenty four months. Essential listening for anyone building, funding or regulating the future of AI, life sciences or space.
Topics covered:
The dark proteome and why current AI hits a brick wall in drug discovery
Microgravity as the next great forcing function in biology
Space data centres, cosmic ray bit flips and the radiative cooling debate
Sovereign AI through the lens of a country, a company and an individual
The UK's innovation track record from Stephenson's Rocket to DeepMind
The AGI fear moment and the AI autonomous kill chains already coming out of Ukraine
Dr Toby Call on LinkedIn: https://www.linkedin.com/in/tobycallMass Balance: https://www.massbalance.bioGAEA AI: https://gaealgm.ai

This week on GAEA Talks, Graeme Scott sits down with May Habib - co-founder and CEO of Writer, one of the most consequential enterprise AI platforms in the world today, and one of the clearest voices in the industry on what it actually takes to make AI work inside the global two thousand.This is the season five opener, and there is nobody better to set the tone. May co-founded Writer in 2020 and has built it into the enterprise agentic platform that serves Mars, AstraZeneca, GlaxoSmithKline, UnitedHealthcare, Boots, Clorox, Metro Bank, Currys, Monzo, BNY Mellon, Edward Jones and many more of the most demanding enterprise buyers in the world. The company has raised four hundred million dollars to date, was valued at around two billion dollars at its Series C, employs close to five hundred people across six cities, and has built a reputation for being the rare AI company that actually understands what regulated, brand-critical, global enterprise looks like from the inside.In this episode, May lays out a complete framework for how enterprise AI is going to play out over the next six to twenty four months. She argues that individual productivity does not rewire an organisation for productivity, that frontier labs have left a huge gap by behaving like celebrities rather than partners, and that the next wave of enterprise AI will be defined by voice and mobile, by the complete rewrite of the sales and marketing tech stack, and by whether enterprises own their means of intelligence or rent it from the labs. She also lays out the most rigorous tokenomics conversation we have recorded on the show, and offers the single best three-step playbook for leaders that I have heard this year. Essential listening for any executive, founder, CIO or board member making decisions about enterprise AI in 2026.What you'll take away from this conversation:• Why individual productivity does not rewire an organisation for productivity• The "labs as celebrities" problem - why frontier providers leave the room before the work starts• The brand as code thesis that became Writer's product North Star• Why voice and mobile become the primary AI interface in the next six months• The coming complete rewrite of the sales and marketing tech stack• The operating agents concept - agentically constructed databases as the system of record• Why the iconoclasts inside the enterprise are not who you think they are• The C-suite power-law inside companies - and why the elite of AI power users matters more than headcount• Sovereign AI for enterprise - owning your means of intelligence, not renting it from the labs• Multi-jurisdiction compliance from day one - Mars, Cartier, Monzo, BNY Mellon, Edward Jones•Tokenomics as the new Econ 101 - and why shared context cuts token consumption twenty to thirty percent• Why frontier model pricing is going up the moment the IPOs close• The "you don't get fired for buying IBM" trap, now applied to a five million dollar monthly Anthropic bill• Why the squad-based approach beats incremental change every time• Why the future of work is not fewer people, but vastly more agents

This week on GAEA Talks, Graeme Scott sits down with Dr Ewan Kirk - founder of Cantab Capital Partners, former Goldman Sachs partner and Head of Quantitative Strategies, chair of the Isaac Newton Institute and non-executive director of BAE Systems.A mathematician by training, Ewan spent thirteen years at Goldman Sachs, rising to partner and leading a 120-strong European team of quants. In 2006 he founded Cantab Capital Partners, growing it from a team of two to roughly £4.5 billion under management before selling to Swiss asset manager GAM in 2016. Today he lives a "portfolio life" as philanthropist, board member, Royal Society Entrepreneur in Residence at Cambridge, and adviser to early-stage companies.In this episode, Ewan brings a quant's discipline to the AI conversation and refuses to let the hype slide by. He draws a hard line between deterministic systems and probabilistic LLMs ("complete the sentence - the cat sat on the - and almost all the time it'll say mat, but one day it'll say roof. Are you okay with that?"), explains why benchmark testing is fundamentally flawed once the test suites leak into the training data, and dismantles the boardroom reflex to "squeeze some AI in and hope magic happens." He's sharpest on the gap between commercial bets, where capitalism lets firms be wrong, and government bets on AI and quantum, where "there's no opting out." This is essential listening for any leader being told AI is coming and they'd better get on board or lose out.What you'll take away from this conversation:• Why an LLM is not "AI" - and the distinction between machine learning, data science and the chatbot layer everyone is actually buying• The deterministic vs probabilistic divide - why Cantab's backtesting and risk systems gave the same answer every time, and why LLMs fundamentally cannot• "The cat sat on the roof" - Ewan's one-sentence demonstration of why he'd never let an LLM run his bank account• Why benchmark testing is broken - public test suites end up in the training data, so beating the benchmark proves almost nothing• The boardroom trap - "it's not enough to say I'm going to squeeze some AI chatbots into my business and magic will happen. What magic are you looking for?"• Commercial risk vs government risk - why a firm being wrong is just capitalism, but a government going all-in on AI or quantum means "there's no opting out"• The marginal cost problem - why the LLM economy is not like the early internet, and why economics, not technology, may be what trips it up• The attention economy decoded - how emotional state drives behaviour, and how Facebook and Instagram monetised the thin layer on top of the open web• The three boundary conditions for AI's future - doom and mass unemployment, a productivity boom like the rise of computing, or trillions in capital vaporised• The decision-maker's toolkit - ask for the concrete not the abstract, demand a testable prediction, and run the randomised controlled trial before betting the business• "Is it a big number or a small number?" - the single heuristic Ewan reaches for every time he hears a statistic on the news• Why AI is ultimately a human problem - and why, stripped of the human element, "nearly everything would fall apart"

This week on GAEA Talks Live from HumanX, Graeme Scott sits down with Emil Eifrem - co-founder and CEO of Neo4j, the creator of the world's leading graph database, and one of the architects of the knowledge graph movement that is now powering explainable AI inside governments, banks and Fortune 500 enterprises.Emil started Neo4j more than twenty years ago after his team hit the limits of tabular databases while building a content management system in Sweden. Their prototype graph engine was a thousand to a million times faster at traversing relationships. He has since built Neo4j into the category-defining graph database, used by organisations from one of the world's top five banks to every major research institution working on AI. He is a veteran of the Swedish open source community and one of the most cited voices in the industry on explainability and trust in AI.In this episode, recorded live at HumanX 2026 in San Francisco, Emil takes us inside the single biggest bet in enterprise AI right now - that knowledge graphs will become a default box in every serious AI architecture. He explains why vector search alone gives you similarity with no context, why mechanistic interpretability only solves half of the explainability problem, and why rag without a knowledge graph is a dead end for mission-critical decisions.What you'll take away from this conversation:- Why the last fifty years of tabular databases cannot model the real shape of enterprise data, and why graphs can- The "empirical versus subjective" framework for deciding what AI can and cannot be trusted to own- Why every AI decision still needs an accountable human - and why that makes explainability the critical constraint- How knowledge graphs complete the explainability story that mechanistic interpretability starts- Why vector similarity scores without context are dangerous for rag retrieval in the enterprise- The real lesson from the surgeon-and-hospital insurance rabbit hole - accountability has to be architected in from day one- Why platform shifts are the only moment database companies ever get built - and why the AI shift is the biggest yet- How one top five global bank went from three percent graph adoption pre-AI to twenty percent today- The Goldilocks rule for enterprise AI - don't start with self-driving or with trivia, pick the meaningful middle- Why "AI ready from a data perspective" is the single biggest five-year survival question for enterprises- Let the business problem drag the AI, and let the AI drag in the data - Emil's single best piece of adviceAbout Emil Eifrem:Emil Eifrem is the co-founder and CEO of Neo4j, the company behind the world's leading graph database. Born and raised in Sweden, he started programming as a child, became CTO of a Swedish startup at twenty, and co-invented the property graph model that is now the foundation of the entire graph database category. Neo4j is used by seventy five percent of the Fortune 100 and powers some of the most sensitive AI, compliance, fraud and intelligence workloads in the world.GAEA Talks is the enterprise AI podcast for leaders navigating the age of artificial intelligence. Subscribe for weekly conversations with the people shaping the future of business, technology and society.Emil Eifrem on LinkedIn: https://www.linkedin.com/in/emilefremNeo4j: https://neo4j.comHumanX: https://www.humanx.coGAEA AI: https://gaealgm.ai#AI #ArtificialIntelligence #GAEATalks #GAEATalksLive #HumanX #HumanX2026 #EnterpriseAI #Neo4j #KnowledgeGraph #GraphDatabase #ExplainableAI #RAG #LLMs #AIReliability #ResponsibleAI #AIGovernance #AIPodcast #GAEAAI