The AI Summit London, 10th & 11th June 2026. Facts below were checked against public sources where possible (see the reference list at the end). Company-specific numbers are attributed to the executives who shared them on stage.
The AI Summit London 2026 marked its 10th anniversary this year as the flagship event of London Tech Week, drawing over 5,000 attendees and 300-plus speakers to Tobacco Dock. One message ran through both days. The industry has moved past “should we use AI” and is now wrestling with harder questions. Who governs the agents? Who pays for the compute? And what happens to the people? Here is what each session delivered…
1. The Future of Urban Mobility
- Speakers described a shift from traditional cars to smaller, lighter and autonomous vehicles, which could free road space for people and parks.
- Each company is likely to build its own mapping data rather than rely on one shared city map, which raises fresh privacy questions about sensors collecting images, Wi-Fi and Bluetooth signals.
- A project with the EU and the city of Helsinki shares sidewalk condition data for better maintenance, and the BlindSquare app, built for visually impaired users, uses such data to warn of obstructions.
2. Demystifying AI Agents: Build One Live
- Agents are simpler than people think. A live demo built a “wine scout” agent that tracks the UK wine market for good deals and explains what each bottle pairs with.
- The hard part inside companies is not building the agent. It is the digital plumbing: data access, API keys and IT security approvals.
- Break big tasks into smaller pieces. Agents perform far better that way.
3. IBM: Governing the Agent Explosion
- The speaker said 80% of organisations now use AI agents, up from around 50% four years ago, and the result is fragmentation, with every department building its own.
- Seven out of ten executives now name governance, not agent building, as their top problem, according to figures shared on stage.
- IBM demonstrated Project Bob, its AI coding assistant, working with an orchestration layer that gives one control plane for visibility, permissions and testing. Pearson added that AI in education must guide learning, not act as an answer engine.
4. Ericsson: Scaling AI to 90,000 People
- Ericsson’s speakers said the company has onboarded 60,000 employees to its AI platform and is targeting 90,000, splitting the strategy into “broad AI” (everyone uses it) and “deep AI” (automating key workflows for around 20% impact).
- Staff complete an internal “AI driving licence” so they understand basics and guardrails before using the tools.
- Measure value per agent, not the number of licences bought, and use local champions to drive adoption.
5. UK Government: A Package of New Announcements
- An extension of the Spärck AI Scholars programme, the £17.2 million scheme named after computer scientist Karen Spärck Jones that funds master’s degrees at nine leading universities, with new support to route scholars into startups and entrepreneurship.
- The RIBA x DSIT Data Centre Design Challenge, the first government-backed design competition for data centres, so they become civic assets with public engagement and strong environmental outcomes rather than grey boxes.
- New joint guidance from the Regulatory Innovation Office and the Health and Safety Executive on how robots can safely work alongside people, giving industry regulatory clarity.
- A Pro-Worker AI Adoption Prize. An expert panel chaired by Nobel Prize-winning economist Simon Johnson will shortlist the top 50 UK organisations using AI to boost workers, and business school professors will write case studies on the three winners.
- An Open-Source AI Builder Fund worth over £500,000, giving 160,000 GPU hours from the UK’s AI Research Resource to the strongest projects from the Hack for Impact hackathon run with NVIDIA, plus mentoring and a developer board for young open source builders.
6. Defence: Task Force RAID Goes Live
- The Chief of the Defence Staff announced the Rapid AI Delivery Taskforce (RAID), launched with the Prime Minister’s backing during London Tech Week. It reports directly to him and can bypass normal procurement to move fast.
- RAID targets four jobs: machine-augmented intelligence fusion, a recognised electromagnetic environment picture, automated operational planning and AI-enabled drone swarms.
- His core argument: “Warfare is at its core competitive... The side that is able to diffuse and adopt technology faster than an opponent will win.” But the UK keeps humans, not machines, accountable for decisions on lethal force.
7. Pandora: Two Agents, Two Very Different Results
- The world’s largest jewellery brand runs two AI agents: Clara for customer service and Gemma, a sales assistant built to bring the in-store “magic” of emotional gift conversations online.
- Clara gets strong traction. Gemma struggles because shoppers are stuck in habit-driven click-and-browse flows, so Pandora is exploring WhatsApp, messaging and voice channels instead.
- Pandora built its own conversation analytics engine and reads real chats to judge agent performance, and credits a fully composable architecture, with a service layer over every e-commerce function, for making the agents scalable.
8. Elsewhen: Run Operations at a Third of the Cost
- The consultancy’s pitch: most enterprises see no meaningful ROI from AI because projects are disconnected from their data, while custom agentic systems can run expensive operations at roughly a third of the cost.
- The speaker argued agentic engineering shifted gear in late 2025, citing companies like Shopify and Stripe generating large amounts of code autonomously, and said about 80% of their own platform code is written by agents under senior engineer review.
- Their KYC onboarding example promises measurable value within four weeks.
9. Cybersecurity Panel: Attackers Are Winning the Speed Race
- Panelists argued AI has not changed cybersecurity fundamentals, but it has made attacks cheaper and faster. Attackers adopt new tech instantly while defenders wait for regulatory and procurement approval.
- High AI token costs make security tooling expensive at scale, a real concern raised for bodies like the NHS where adoption is mandated but budgets are tight.
- Keep humans in the loop. Panelists shared cases where human intuition caught what AI missed, and warned against over-reliance on automated judgment.
10. Formula 1: AI as Co-Pilot, Not Replacement
- AI gives race engineers real-time insight and cuts data-crunching time, but the trust between driver and engineer stays human.
- Massive race-weekend data is useless without rigorous cleaning and physics-informed human judgment.
11. AI Governance: The Million-Dollar Lesson
- The most striking numbers of the summit, drawn from EY’s Responsible AI Pulse survey of 975 executives: 99% of large organisations reported financial losses from AI-related risks, 64% lost more than 1 million dollars, and the average loss about 4.4 million dollars per affected company.
- The flip side, per the same EY research: companies with well-developed responsible AI governance reported stronger sales, cost savings and employee satisfaction. Governance pays.
- Agentic AI raises the stakes. Speakers warned of goal hijacking, tool misuse and “accidental insider threats”, and called for agent inventories, risk tiering, zero trust and continuous runtime controls. Their advice: do not wait for regulation, because technology will always outrun it.
12. The Compute Crunch in Science
- Scientific AI is booming. AlphaFold has made structure predictions available for over 200 million proteins, and the speaker cited Recursion taking a cancer drug candidate to trial in about 18 months, a fraction of traditional timelines.
- GPU scarcity is a real bottleneck. The speaker described a company that lost six months of research waiting for chips.
- Nebius, the AI infrastructure firm, is targeting more than 3 gigawatts of contracted power by the end of 2026, with 800 megawatts to 1 gigawatt active.
13. Measuring ROI: From Productivity to Growth
- One short but sharp lesson: stop measuring AI by productivity gains alone. Put every use case through an investment risk committee and judge it on top-line growth, transaction volume and hard financial forecasts.
14. Future of Work: Skills That Do Not Exist Yet
- The most valuable job skill in five years may not have a name yet. Today’s must-haves, per the panel: AI literacy, data fluency, creativity, critical thinking and adaptability, with skills now needing a refresh every 18 to 24 months.
- AI literacy is not just technical. It includes the domain knowledge to judge whether an AI’s answer is actually right, and the confidence to use it at work.
- Women remain underrepresented in the AI workforce. The panel argued that without diverse builders, AI products will fail half the population, and flagged older women and customer service workers as groups most exposed to automation.
15. Virgin Atlantic: The AI Concierge Earning Its Keep
- Virgin Atlantic launched its AI Concierge in December 2025, built with OpenAI and design firm Tomoro, letting customers plan and book travel by text, voice or image across its websites.
- On stage, Virgin Atlantic executives said the concierge has handled around 500,000 conversations, generated about £1 million in revenue, and resolves roughly 80% of queries without a human, handing the rest to live agents.
- A side benefit nobody expected: the concierge keeps exposing errors and unclear content on the website, improving it for everyone. The airline also outlined plans for Starlink connectivity toward a fully connected fleet by 2027.
16. AI Filmmaking: A Movie in Two Weeks
- A creative director made a short film in two weeks using around 100,000 AI video generations,.They also claimed AI is now used in over 90% of VFX storyboarding and animatics work.
- AI still cannot deliver natural dramatic dialogue or sustained emotion. Humans had to add the drama by hand.
- High-concept sci-fi that once needed Marvel-scale budgets may become possible for under 1 million dollars, opening the door to thousands of untold stories.
The Five Trends That Defined the Summit
- Governance overtook building. From IBM to the EY-backed governance session, the loudest message was that ungoverned AI loses money, an average of 4.4 million dollars per affected company.
- The plumbing problem. Almost every enterprise speaker, from Ericsson to Pandora, said data access and IT approvals slow AI down more than the technology does.
- Agents went mainstream and fragmented. High adoption sounds like success until every department runs its own incompatible agent.
- Compute is the new oil. GPU shortages delayed science, defence wants AI on the frontline fast, and the UK government is now handing out GPU hours as policy.
- Humans stay in the loop, by choice. F1 engineers, cyber defenders, filmmakers and the military all drew the same line: AI assists, humans decide.
References and Further Reading
The event
- The AI Summit London (official): https://london.theaisummit.com/
UK government announcements
- Open source AI fund, data centre challenge and robotics partnership (GOV.UK): https://www.gov.uk/government/news/new-backing-for-open-source-ai-builders-data-centre-design-challenge-and-robotics-partnership
- Pro-Worker AI Adoption Prize (GOV.UK): https://www.gov.uk/guidance/pro-worker-ai-adoption-prize-how-to-nominate
- RIBA on the Data Centre Design Challenge: https://www.riba.org/news/riba-and-government-design-ideas-competition-for-data-centres/
- Spärck AI Scholarships, £17.2m (Computer Weekly): https://www.computerweekly.com/news/366625617/Labour-commits-172m-to-support-Sparck-AI-scholarships
- Minister’s data centre speech coverage (Computer Weekly): https://www.computerweekly.com/news/366644212/UK-minister-of-AI-calls-for-more-attractive-datacentre-builds
- Open source builder fund coverage (Computing): https://www.computing.co.uk/news/2026/uk-backs-open-source-ai-builders
Defence
- Taskforce RAID launch (GOV.UK): https://www.gov.uk/government/news/new-taskforce-to-put-ai-on-the-uks-frontline
- Chief of the Defence Staff speech at London Tech Week (GOV.UK): https://www.gov.uk/government/speeches/chief-of-defence-staff-speech-at-london-tech-week
- “Adopt AI or lose future wars” (City AM): https://www.cityam.com/uk-defence-chief-adopt-ai-or-lose-future-wars/
Company stories and data
- EY Responsible AI survey, the 99% and 64% loss figures: https://www.ey.com/en_gl/newsroom/2025/10/ey-survey-companies-advancing-responsible-ai-governance-linked-to-better-business-outcomes
- Virgin Atlantic AI Concierge launch with OpenAI: https://openai.com/index/virgin-atlantic-oliver-byers/
- Pandora’s Clara and Gemma agents (Digiday): https://digiday.com/marketing/pandora-is-betting-on-ai-agents-to-scale-service-and-emotional-selling-during-the-peak-holiday-season/
- Pandora case study (Salesforce): https://www.salesforce.com/customer-stories/pandora/
- Nebius capacity plans for 2026 (Quartz): https://qz.com/nebius-pipeline-growth-supports-strong-2026-revenue-outlook
- AlphaFold protein structure database: https://alphafold.ebi.ac.uk
Submit your review | |
Journalism News Network
Average rating: 0 reviews