What Coding Agents Teach Us About Building AI Systems

​Why have coding agents like Claude Code improved so quickly over the past year, and what can we learn from them to build better agents in other domains?

In this event, you’ll hear from people actively building agent systems, and we’ll break down the core primitives of the agent loop: retrieve relevant context, reason, act with tools, verify the result, and repeat. With that loop, we’ll go deep on what an agent harness does and why retrieval and memory have emerged as core problems in building useful AI systems.

Coding is the clearest example because code is highly verifiable: agents can write code, run builds, execute tests, inspect diffs, and close the loop automatically. That verifiability has made coding the ideal environment for reinforcement learning, and frontier model companies have leaned into it heavily.

Coding may be the first domain where this pattern is working at scale, but is it the only one? We'll explore what it takes to bring the same closed-loop verifiability to other domains, and share what we've learned building our own agents and agent harness at Hornet.

​Following a run of sold-out events in SF, London, and Stockholm, Hornet is now hosting our first event in our home base, Trondheim.

18:00 - Doors open
18:15 - Talk by Jo Kristian Bergum, CEO Hornet.dev
19:00 - Panel discussion
19:30 - Q&A and mingle

​Food and drinks will be served

​If you want to read up on Hornet's approach, read our blog posts: https://hornet.dev/blog