A completed London Agentic AI evening with Amazon on harness engineering: the runtime, tools, memory, security and observability layer around models that turns them into production-ready AI agents, featuring talks from Amazon, Databricks and Pydantic.
Hosts, sponsors and speaker organisations represented at this London Agentic AI stop.
This London Agentic AI stop dug into harness engineering, the practical discipline of building, deploying, testing and operating production-ready AI agents. If an agent is a model plus a harness, the evening focused on the harness: the runtime systems, tool integrations, memory management, security protocols and observability frameworks that make agentic systems reliable.
Amazon hosted the community at its London office for a technical evening. Anisha Malde opened with an introduction to harness engineering framed as building a winning multi-agent team, Sultan Al Awar from Databricks presented Omnigent, an open-source meta-harness for composing agents with security and cost controls, and Samuel Colvin, CEO of Pydantic, covered harness engineering with Pydantic AI.
A concise briefing on the technical context behind the event: why the topic mattered to the London Agentic AI community and what sponsors, speakers and attendees came to discuss.
The Luma and Meetup pages confirm the full title, Amazon LHR16 London venue, 15 July 2026 timing and approval-based Luma registration.
The published programme lists three talks: Anisha Malde (Amazon), Sultan Al Awar (Databricks) on Omnigent and Samuel Colvin (Pydantic).
The event recording is now available on the London Agentic AI YouTube channel.
What harness engineering is and why the layer around the model decides production reliability.
How to orchestrate multiple agents into a coordinated, reliable team, with verification and human-in-the-loop checkpoints.
How Omnigent works as an open-source meta-harness for composing agents with security and cost controls.
How Pydantic AI approaches harness engineering for typed, dependable agent behaviour.
How runtime, memory, security and observability come together to run agents in production.
An introduction to harness engineering framed through multi-agent orchestration: how to coordinate agents like a winning team, with verification and human-in-the-loop checkpoints around runtime, tools, memory and observability.
A technical talk on Omnigent, an open-source meta-harness for composing AI agents with built-in security, policy and cost controls for production workloads.
Pydantic CEO Samuel Colvin on harness engineering with Pydantic AI: building typed, dependable agents with structured validation around the model. Full abstract to follow.
See how the event unfolded: the room, the community and the atmosphere from this London Agentic AI gathering.
Open on YouTube →Original registration, archive and recording links for this London Agentic AI event.
The UK's original, high-signal Agentic AI community for AI engineers, agent builders, researchers, founders and technical leaders building production agents.
Subscribe on Luma →
