Last week we spent two days on the floor at London Tech Week and came back genuinely buzzing. Not because everything was polished and perfect. In fact the opposite. The most exciting thing about the AI space right now is how messy and experimental it still is. People are figuring this out in real time, building things that did not exist twelve months ago, solving problems that most businesses do not even know they have yet.
Here is what we found.
The Conversation Everyone Was Having: AI Agents
If you spent five minutes at London Tech Week without someone mentioning AI agents, you were probably in the toilet.
The shift from chatbots to agents was the dominant thread running through the entire event. And once you understand the distinction, it is hard to unsee. A chatbot answers your question. An agent goes and does the thing.
Roger from Blackbox AI was one of the first people we spoke to and he put it as clearly as anyone: an agent is the same underlying model you are already familiar with, but with an additional layer of permission to actually execute actions on your behalf. It books. It schedules. It processes. It acts. Without a human managing each step.
Roger’s world is the security side of this, and he was refreshingly direct about how alarming the implications are. Agents can now launch cyberattacks at scale without any human involvement whatsoever. Before agents, a cyberattack needed people to execute it. Now it does not. The scale of what this enables is, in his words, difficult to fully comprehend. His recommendation was counterintuitive but logical: use AI to run offensive attacks on your own systems around the clock to find the vulnerabilities before someone else does. Open source models make this affordable. The businesses not doing this are the ones that will end up in the news.
He also flagged something worth paying attention to on the cost side. An Uber CTO recently burned through an entire annual AI budget in four months. As AI usage scales, spend management is becoming as important as capability.
Then we spoke to Neyla from Dust, a French company with offices in New York and San Francisco, and the conversation took a completely different turn. Where Roger’s world is enterprise security, Neyla’s is accessibility. Dust is an agent platform that lets entire teams build and share custom AI agents using nothing but plain English. No code. No developer. No technical background required. You define what you want the agent to do, what role it plays in your organisation, and it gets built.
The phrase she used that stuck with us: multiplayer AI. Not one person in a company using ChatGPT on their own. An entire team building shared tools, collaborating with agents, scaling their collective output. The gap between knowing AI exists and having it actually work for your business is closing faster than most people are ready for.
Specialist Tools Are Quietly Winning
One of the more interesting patterns across the two days was how consistently specialist tools were beating general ones in the conversations we had.
Dishank from AttifinAI is building a legal AI platform for UK law firms, and the problem he is solving goes deeper than efficiency. Law firms handle sensitive client data and the prospect of that data being processed on American servers, subject to US law, and potentially used to train future models, is a genuine compliance risk that the legal sector is increasingly awake to. AttifinAI runs entirely on-premises. Nothing leaves the organisation. Nothing gets used for training. It searches across UK jurisdiction sources in real time including legislation.gov.uk and the Supreme Court, and every single response comes with a citation so the user can verify exactly where the answer came from.
That last detail is what makes it genuinely useful rather than just interesting. In a high-stakes professional environment, a tool that says here is the answer and here is exactly where it came from is categorically different from one that just gives you the answer and hopes you trust it.
Uchechukwu Buzugbe, a Data Scientist at Analytics Intelligence, was tackling a limitation that most people using AI tools are completely unaware of. Large language models are language tools. They predict the next word. They are not built for maths. When you send a spreadsheet to ChatGPT and ask it to analyse the data, it will often get it wrong, and it will get it wrong confidently.
Uchechukwu’s contribution to solving this was architectural. Working on one of enterprise AI’s most overlooked challenges, he developed a dedicated reasoning engine that performs mathematical computations independently before passing the validated results to the language model for natural language explanation. The results are deterministic – meaning the same dataset and the same question will always produce the same answer, every single time, because the answer is being calculated rather than predicted. This architecture bridges the gap between conversational AI interfaces and the accuracy and reliability that enterprise grade data analysis actually demands. The practical result is that organisations can interact with complex business data through plain English without sacrificing analytical consistency. He described it as having a data analyst in your pocket at two in the morning. You would not ring your actual analyst at that hour. The agent does not mind.
Rohan from I-Migrator was solving a problem that affects far more businesses than most people realise. UK immigration compliance is genuinely complicated, and the rules change constantly. Government updates happen without warning. Employers who sponsor foreign workers can find themselves non-compliant over something as small as failing to report a salary change. Rohan’s platform monitors these changes automatically and flags what employers need to do before it becomes a problem. The alternative, as he put it, is paying extortionate fees to immigration lawyers to stay on top of something that technology can handle continuously and at a fraction of the cost. It is a quiet, unglamorous solution to a real and expensive business risk.
The Big Players Are Paying Attention
Not everyone on the floor was a startup. Patrik from JP Morgan Workplace Solutions was there specifically to find and support early stage founders, and their offering was surprisingly straightforward. A free cap table solution for up to forty stakeholders, designed to get companies investor-ready before a funding event forces the issue.
The insight Patrik shared that landed hardest was this: most founders do not realise they have a cap table problem until something major happens, and by then it is usually too late to fix it cleanly. JP Morgan’s play is simple and smart. Offer a genuinely useful free tool to founders who have four or five employees and build a relationship now, with the hope that when those founders become the next Stripe or Revolut, they stay in the JP Morgan ecosystem. A loss leader with serious long-term thinking behind it.
The Creative Space Was Not What We Expected
Going in, we half expected the creative side of the event to feel like an afterthought. It was anything but.
Jamie from Epidemic Sound was one of the most direct voices of the two days. Asked whether AI-generated music is the future of their platform, the answer was clear: no. Jamie’s view is that generative AI music consistently fails to capture genuine human emotion. It sounds robotic. It lacks the depth that a real musician brings. Epidemic Sound uses AI to help creators find the right music faster and to give artists better tools, but the music itself is made by people. If a brand needs something custom, the platform connects them directly with an artist who creates it for that brief specifically. You are supporting a human. You are getting something that cannot be replicated by a prompt.
In a landscape where synthetic content is becoming ubiquitous, that position is both a philosophical stance and a genuine commercial differentiator.
Anna from Visualise stopped us in our tracks entirely. We sat down in a VR headset and watched a 360 degree immersive film about the Chernobyl disaster. It did not feel like watching a documentary. It felt like being there. Anna described the core idea simply: you can transport someone somewhere they cannot go, and the story becomes more believable because the person is literally placed inside it. The creative applications of immersive content are vast and largely underused. For brands with complex or emotionally resonant stories to tell, the gap between conventional video and this kind of experience is worth taking seriously.
The Human Problem Nobody Talks About
Graham from LemonWorx builds computer vision systems for retailers, turning camera footage into hard data about how customers move through and interact with physical spaces. Dwell time, heatmaps, interaction patterns. Things that used to be decided by a manager’s gut feeling now have numbers behind them.
But the most important thing Graham said had nothing to do with cameras or retail. Asked about the biggest challenge facing the AI space, his answer was immediate: people. Not the technology. The people. There are simply not enough engineers, data scientists and AI specialists to meet current demand, and the gap is growing. Every company building in this space is competing for the same scarce talent. For the next generation of professionals deciding where to build a career, Graham’s message was direct: go into computer science or engineering. The demand is enormous and it is not going away.
It was a grounding moment in a two-day conversation that could otherwise feel entirely abstract. The most sophisticated AI system in the world still needs humans to build it, direct it and make sense of what it produces.
What the Keynote Stage Added
Beyond the interviews, the Transformation Stage provided two of the most practically useful sessions of the event.
The Hexagon presentation was the one that stayed with us longest. Their CMO walked through what happened when a small team used AI as a genuine efficiency partner across their entire brand operation. Two people managing the workload of fifteen. Workflow speed up 75 percent. Resource usage down 87 percent. £1.5 million saved over eighteen months. Brand perception up 71 percent. The numbers are extraordinary but the framework behind them is simple. Automate the repeatable work: content production, data collection, reporting, analytics. Protect the irreplaceable work: strategy, authentic voice, creative direction, original thinking. The moment you outsource your thinking to AI your brand starts to sound like everyone else using the same tools. They called it a sea of word salad and it is an accurate description of a lot of what is being published right now.
The line from their keynote slide that summarised everything: the differentiator is not the tool. It is the talent.
Sid from Pinterest added a layer of genuine practical value in a fireside chat on trends. Modern trend cycles are moving 4.4 times faster than they were in 2018. Viral moments are not trends. Durable cultural shifts are trends, and Pinterest sees them earlier than almost any other platform because its users come to plan their next decision rather than document their last one. Their annual forecast, Pinterest Predicts, has maintained an 88 percent accuracy rate over six years and is completely free to access. For anyone doing content or social planning, it is one of the most underused tools available.
What It All Adds Up To
Two days, nine conversations, two keynote sessions. Here is what we are taking back to how we work and how we advise our clients.
The move from chatbots to agents is real and it is happening now. If your business is still treating AI as something you occasionally ask a question, you are already behind the curve. The question to ask is not whether to use agents but which process to start with.
General AI tools have real limitations that most people are not aware of. They cannot be trusted with maths. They process data on servers that may not be appropriate for sensitive information. Specialist tools built for specific industries and specific problems are consistently outperforming general alternatives for exactly these reasons.
Security is the next major conversation in AI. The shift toward autonomous agents means the attack surface for every business is growing. Knowing where your data goes when you use AI tools is no longer optional.
Compliance is a quiet but growing risk. Whether it is immigration rules, data sovereignty or professional regulations, the businesses that use technology to stay on top of these things automatically will have a significant advantage over those managing them manually.
Human creativity is not under threat. It is increasingly valuable. The businesses flooding the market with AI-generated content are making authentic human voice scarcer and therefore more valuable. The agencies and brands protecting their genuine voice are building a real competitive advantage.
The talent gap is real. The technology is advancing faster than the people who can build and direct it. For businesses serious about AI, investing in people is as important as investing in tools.
The differentiator is not the tool. It is the talent.
More to come. Our next piece goes deeper into the biggest challenges facing AI right now, straight from the people building in it every day. Follow Livetech so you do not miss it.