1# London Software Guild
Details
About the London Software Guild
The London Software Guild gathers developers, architects, and teams across London who care deeply about the craft of building great software. No fluff, no vendor pitches — just people who take the work seriously, sharing what they've learned the hard way.
***
## Tonight's Speakers
## 🎤 Kieran — Founding Engineer, Demand-Genius
Kieran is a founding engineer at Demand-Genius, working to close the gap between how humans and machines perceive your brand versus reality. Background in electrical & electronic engineering, previously working on real-time payments systems, high-scale distributed dynamic rules engines, and reverse proxies. Third Start-Up/Scale-Up (because I'm dumb and have never learnt my lesson).
Talk: The Statistical (Un)Certainty That Our AI Co-Founder Ruins Everything
We have handed the most powerful text prediction machines ever built the keys to our codebases. They write code that is statistically likely to be correct. But "statistically likely" and "correct" are not the same thing. As we move toward autonomous multi-agent systems writing production code at machine speed, that gap becomes the central engineering problem of our time.
Code is not text. It is a formal, deterministic, mathematical structure that happens to be serialised as text. Every tool in the agentic coding stack — retrieval, editing, version control, coordination — treats it as text.
This talk argues that the answer isn't better frontier models. It's better systems around models. Specifically, the engineering disciplines we've used for a century to control uncertain systems — feedback loops, formal verification, dependency analysis, and control theory — are the missing layer between "AI can write code" and "AI can write code you can trust."
***
## 🎤 Andy Roberts — Senior Engineering Leader
Andy Roberts is a senior engineering leader with 25+ years of experience — the kind where you've actually written the code, run the teams, and been in the room when things went wrong at scale. Previously Head of EMEA Enterprise Solutions at Apollo GraphQL, where he helped companies like [Booking.com](https://booking.com/), PayPal, and The New York Times stop arguing about their APIs. Before that, Head of API Engineering at RS Components, wrangling 11 teams. Has strong opinions about federated architecture and isn't afraid to use them.
Talk: Built for Browsers, Billed by Tokens
Your microservices solved for latency. AI agents are solving for context. Nobody told your APIs.
Every era of software development has been defined by a constraint that reshaped architecture. In the database era we optimised for round trips. In the web era we obsessed over latency. Mobile forced us to think about bandwidth and battery. Each shift required new instincts, new patterns, new defaults. We're in the middle of another one.
The AI era introduces a new constraint: context. Language models have finite working memory, and every API response your services return goes into it. The kitchen-sink response payloads that work fine for browsers are quietly making your AI agents more expensive and less capable — and the cost is almost entirely invisible in your existing dashboards.
This talk maps the constraint transition from latency to context, shows what it means for the microservice APIs you have today, and offers five concrete things you can do about it without touching a line of existing backend code.