Right now, the team is me.

I've spent 10+ years across startups, product, strategy, and marketing — a generalist who's lived in the personal-knowledge-management space for a long time. Early adopter of Roam, then Logseq, Tana, Obsidian, Notion, Coda.

At every company I've worked at I've been "the Notion guy" — the one who built the operating system everyone else ran on. I've done the same on a consulting basis, pairing those systems with AI-powered workflows. Being non-technical in the traditional sense turns out to be an advantage: I know what an ordinary knowledge worker actually wants from an AI workspace, because I am one. Today I'm AI lead at Dragonfly Thinking, building multi-agent strategic-intelligence systems and teaching teams to go AI-native.

At the risk of sounding like a tosser: I'm a pretty high-agency guy. Every job I've had I've gone in through the side door rather than traditional channels — cold-emailed my way onto Peter Diamandis' private plane for a job I wasn't qualified for (I didn't get it), while I was still at uni. Started a podcast in my early twenties because I was reading books and wanted to ask the authors my questions.

Alongside the product work, I studied philosophy with honours, working across cognitive science, ethics, thermodynamics, and complex systems. My thesis, On What Is and Should Be, sat at the intersection of the Free Energy Principle, active inference, and entropy-based ethics. A second essay, A Selfish & Computational Justification for Global Wealth Redistribution, reframed economic growth as evolutionary search through possibility space.

I had long-form conversations with people working in these fields — Kevin Kelly, Geoffrey West, Aubrey de Grey — on the Talk of Today podcast. I seriously considered a PhD to keep pulling on the threads. I chose to build instead — these ideas land harder inside a product than inside a department.

Four ways the work above shapes Calyx:

1. A model can only act on a world it can see. Agents are systems that model the world and act to minimise prediction error — they work badly in opaque substrates and well in legible ones. That's why Calyx is a folder of plain markdown files: deterministic, inspectable, low-entropy. The file-first call isn't aesthetic. It's the condition under which agentic reasoning actually works.

2. Capability democratisation as a product implication. If humans are the distributed computational substrate of the economy, then unlocking under-tapped cognitive capacity compounds across the whole system. The implication for Calyx: an agent-native workspace cheap and learnable enough that blossoming entrepreneurs in places the AI tooling boom skipped can run a real business out of it.

3. Portfolio of Experiments as a design philosophy. Businesses are organisms on a fitness landscape — survival means continuously shrinking the gap between your model of the world and the world itself. Calyx is shaped to let knowledge workers run many small bets in parallel, see which ones reduce uncertainty fastest, and scale the winners. The FEP applied to how you ship work.

4. LLMs can shape-rotate, but they can't write. Models are reliable on tasks with verifiable success (math, code) and unreliable on tasks with subjective success (writing, judgement, taste). For the next decade the right interface isn't "AI does the work" — it's an environment where humans steer, agents handle the laborious middle, and the artefacts those agents produce live in a substrate the human can inspect.

A few pieces of writing and work, with timestamps.