Adopting Agentic: Software Engineering for the AI Age

Our 30 June 2026 event has taken place. Watch the recorded talks below.

Techspace, London EC2

The moment everything changes

AI is transforming the speed and nature of software delivery — not just as a tool that assists developers, but as an active participant in the engineering process. The focus must shift from augmenting developers with AI tools, to building the processes and environments where AI-assisted development can truly succeed.

We need to reshape our processes: away from ones designed around human strengths and weaknesses, towards new ones shaped around the strengths and weaknesses of AI. There are a great many important questions that we don't have answers to – yet.

We welcomed practitioners and peers for an evening of talks and conversations to share experiences, learn together, and start forming a picture of what our future looks like.

Watch the talks

Colin Eberhardt, CTO, Scott Logic

Introduction to Adopting Agentic Software Engineering for the AI Age
Colin traces the evolution of AI-assisted development from GitHub Copilot’s line-by-line completions to agentic, asynchronous workflows, arguing that software development has shifted from writing code to directing systems that produce it. The economics of code have changed: generation is approaching zero cost under the right conditions, but creating the right code remains as hard as ever. Every process built around human strengths and weaknesses now needs rebuilding around the strengths and weaknesses of AI.

Rhea Sam, AI Adoption Specialist, Valtech

Build Boldly: AI Adoption in Engineering
Rhea argues that the gap between headline AI adoption statistics and meaningful organisational change is almost entirely a people and culture problem: 41% of workers receive zero guidance on AI use, which drives unsafe shadow AI behaviour. She outlines a Champions framework for closing the fluency gap, where peer advocates run honest enablement sessions, document what works, and build reusable workflows from real projects. Her key shift: measure outcome metrics (confidence scores, time saved, productivity gains), not activity metrics like PRs shipped or licence usage.

Chris Price, Principal Architect, Scott Logic

Why You Shouldn’t Add MCP to an Arcade Machine
Chris uses a live demo of an AI agent controlling a real coin-pusher arcade machine (connected via an MCP server) to explore a spectrum from low-level generic tools to a fully algorithmic solution. The central insight: “if all you have is an AI budget, everything starts to look like an agent.” He cautions against incentivising AI utilisation over solution fitness. The punchline: the reason not to add MCP to an arcade machine is the Gambling Act 2005.

Richard Thorpe, Head of Engineering, FE fundinfo

Hard, Honest, In Progress: Agentic Development in the Trenches
Richard gives a candid account of 24 months of agentic AI adoption across a 200-person engineering organisation, including building and deliberately retiring a bespoke multi-agent tool when GitHub Copilot shipped equivalent functionality. He frames that decision as one of their biggest wins. Adoption curves plateau repeatedly, he warns; it takes deliberate new triggers to re-energise teams. Unexpectedly, QA engineers and BAs have led the most creative AI tooling work, now driving SDLC improvements.

Chris Parsons, CTO & Founder

Vision for an AI-First Product Organisation
Chris applies the Theory of Constraints to explain why individual developers reporting 2–3× speed improvements rarely translate into equivalent team-level gains: coding was typically not the bottleneck. He distinguishes two responses: hybridisation (everyone does everything, outsourcing judgment) versus amplification (smaller, smarter expert teams each enhanced by AI). AI-first organisations should amplify small teams, not blur roles. AI handles artefact creation; humans focus on judgment and customer understanding.

Panel Discussion, moderated by Colin Eberhardt

Adopting Agentic Software Engineering for the AI Age
The panel covered three topics: the future of junior engineers, measuring value vs. velocity, and AI sovereignty. On juniors: a short-term hiring dip is likely, but juniors will adapt. Seniors who refuse to change are more at risk. On metrics: PRs work as a proxy, but never as an individual target. On sovereignty: diversify across providers, avoid single-model dependence, and seriously evaluate open-weights alternatives. The gap between open-weights and frontier models has narrowed to roughly two months.