LLM London Sessions #11
LLM London Sessions is back!
This is for people actually building with AI.
We’ve got speakers from Meta and engineers with decades of experience across BBC, BP, and major banks, sharing what is actually working in practice and where things still break.
If you are working with LLMs, agents, or AI in real systems, this will be worth your time.
Talk: From Writing Code to Making Business-Aligned Decisions
Great engineers do not just build things. They help decide what should be built.
Oana May, ex-Meta leader, FAANG product coach, and founder of MOATCRAFT, breaks down product sense, the skill behind how product leaders make decisions in ambiguity, and translates it into practical thinking tools for engineers.
Drawing from her experience coaching 500+ product leaders, she will cover:
What product sense means for engineers and why it is becoming non-optional
How businesses decide what problems are worth solving
Why great ideas often get ignored
How AI is reshaping engineering roles and decision-making
A framework for identifying the right problems before writing code
This is for engineers who want to understand how decisions get made and contribute ideas that actually gain traction.
Talk: Using AI to Improve Delivery Without Rewriting Your Stack
Tools like Claude Code, LangChain, and GitHub Copilot Agent Mode are powerful, but most companies cannot simply replace their existing systems.
Pete explores how teams can instead use AI to build better testing and delivery tooling, helping to:
Speed up delivery
Reduce rework
Improve quality
He will walk through practical examples of why applying AI in test tooling is often more effective than embedding it directly into core application logic.
Pete has worked in software development for almost 30 years across startups, banks (NatWest, Lloyds, BNP), energy (BP, NESO), and media (BBC, Financial Times, Telegraph). He has built, tested, and led teams, focusing on maximising quality and speed using modern tools.
Talk: Building Vector, the Enforcement Layer I Wish Claude Had
Claude is getting better at building things. It is not getting better at admitting when it has not.
This talk explores the gap between capability and reliability through Vector, a coding agent harness designed to make AI agents actually finish what they start.
Talha will walk through how this approach addresses the limitations of current agent behaviour and why reliability is still one of the biggest challenges in applied AI.
Talha is an AI Engineer who spends his days helping teams adopt AI in production. When he’s not doing that, he’s on his Brompton building Vector — a harness that makes AI agents actually finish what they start.
This event is hosted and sponsored by Tessl.