Speaking

I speak about AI deployment in production, engineering leadership, and the trust infrastructure that financial systems need before they can absorb AI's mistakes cheaply. The perspective comes from building these systems, not studying them from outside.

Trust, not capability: measuring whether AI actually helps in production engineering

Most reports of AI making engineers faster are self-reported, and self-report is the least reliable instrument we have. People who like a tool report gains whether or not their output changed. So the interesting question is not whether AI is capable. It plainly is. The question is whether you can show, with evidence, that it helped, and whether your systems can absorb its mistakes cheaply enough to trust it in production.

This talk is an account of doing that inside a large regulated fund: driving AI adoption across an engineering group without the authority to mandate it, and building the measurement to know whether it worked. It covers the baselines worth setting before you change anything, why attendance is not adoption and adoption is not impact, the safety layer that makes AI errors cheap to catch, and treating token spend as a measured input rather than an overhead. One configuration change cut per-session cost by roughly two-thirds. A second team adopted the approach without being asked, which turned out to be the only adoption signal that could not be gamed.

Some of it worked. Some of it did not, and telling the difference is the point.

What you leave with

• How to baseline a developer workflow before introducing AI, so you are measuring change rather than sentiment.
• Why self-reported productivity is unreliable, and what to measure in its place.
• A safety pattern that makes AI-generated mistakes cheap to catch before they reach production.
• How real adoption spreads by demonstrated value rather than mandate, and how to tell the two apart.

Format and audience

30 to 40 minutes, or 45 with questions. A 20-minute version works for a shorter slot. It suits engineering leaders, heads and VPs of engineering, and platform or developer-productivity teams: intermediate to senior, and it assumes engineering leadership rather than AI research. I am comfortable with keynote, panel, and fireside formats, including unscripted conversation and extended audience Q&A.

Other talks

• Ten paradigm shifts: what programming since 1981 teaches you about AI
  From ZX81 BASIC in 1981 to AI agents in 2026. Pattern recognition across every major technology transition. What the previous nine shifts tell us about this one.
• Bainbridge's ironies, forty years on
  A 1983 paper about factory automation predicted the exact problems we now face with AI coding assistants. What it means for how we build teams and protect the cognitive work that matters.
• The suspicion tax: when AI erodes trust faster than it creates value
  Why self-reported AI productivity gains are structurally unreliable, what to measure instead, and why the organisations that get measurement right will compound the advantage.
• Building engineers when AI writes the code
  Junior developer training in the age of AI. What skills still matter, what formative work looks like now, and why optimising for today's velocity breaks tomorrow's capability.

Bio

One line: Ivan Cronyn leads AI enablement at a major London hedge fund manager and is a hands-on principal engineer, writing on AI in production engineering at cronyn.co.uk.

Short (50 words): Ivan Cronyn has built software in regulated finance since 1998, and now leads AI engineering enablement at one of the world's largest listed hedge fund managers. He builds the tooling himself - agents, evaluation harnesses, CI safety layers - and drives its adoption across engineering and the business. He writes at cronyn.co.uk on why trust, not capability, decides whether AI pays off.

Long (150 words): Ivan Cronyn leads AI engineering enablement at one of the world's largest listed hedge fund managers in London, where he builds the tooling - agents, evaluation harnesses, CI safety layers - and drives its adoption across engineering and the business. He has built software in regulated finance since 1998, at Barclays Capital, Merrill Lynch, GLG Partners, Brevan Howard, and his current firm, across risk technology, quantitative research, and client solutions: the full lifecycle from signal generation through risk management to client delivery. He has written code since 1981, starting on a Sinclair ZX81 in Durban, South Africa, and has shipped production systems through every major paradigm shift from structured programming to AI agents. His writing argues that trust, not capability, is the real bottleneck in AI-assisted engineering: the models are good enough, but the systems around them must be designed to absorb their mistakes cheaply.

A fuller career record is on the CV.

Headshot

Download high-resolution headshot (JPEG)

Get in touch

For speaking enquiries: ivan.cronyn@gmail.com or LinkedIn