Learned a ton hosting this month’s Coding from Beach meetup. The pizza order stood no chance with over 60 builders turning up for the session on “Examining the Agent Brain”.
For context, Coding from Beach is the Hub’s monthly meetup for coders, engineers, founders, students, teachers and AI-curious builders. This one was about the “agent brain”: the knowledge, memory and skills an AI agent needs, whereas tools, MCPs and CLIs are more like the agent’s hands.
Lots of alpha with multiple CTO-level talks:
Witek Strzelczyk gave a great presentation on the self-improving company and how he is building a closed loop customer service system for his company, with a graph database as agent “brain” for accessing company knowledge and CLI as agent “hands” for performing useful work in the company systems.
Chris Lang shared how he is thinking about the integration of agents into a large existing software engineering operation. Thought provoking ideas about using agent councils to weigh in on key decisions, and the notion of organisational “beliefs” about how things should be done, and how these should impact agent behaviour.
Greg Freeman showed how his team at NextPlay are building the talent intelligence data layer for the startup and scaleup ecosystem in Australia. Their approach makes use of agentic search to surface talent profiles from their knowledge system, to find nuanced matches with a particular job opening.
Leo Alves took a different angle and showed how you can fine tune small open source models on your own hardware, with specific knowledge, to give them the ability to do more advanced reasoning and perform more complex tasks.
Bert van Brakel made the case for using a graph database as the core engine for an agent brain and how he is constructing a self-improving brain for his projects. Oscar Lloyd showed how the same approach can be used for creating a personal assistant brain with knowledge about your life, gleaned from historical chatbot transcripts which contain a lot of this info.
I shared how you can use Google Docs as a shared brain layer between team members because of its built-in markdown capabilities. The team can edit files in a shared folder and an agent can read those files as markdown via curl in one command. This is handy for knowledge that is changing regularly and lets the team use a familiar tool without having to learn git etc.
Thanks to those who presented and attended.
If you are serious about building with AI and curious about best practices, then you should come to these monthly meetups.
Coding from Beach is part of the Hub’s Tokenizer program, a multi-year AI skills program to help the Noosa region build practical AI capability.
