Jonathan Leahy
Full-stack developer, based in Bristol, UK. Builder of things.
Staff Engineer at Pismo, A Visa Company, building payments infrastructure at scale. On nights and weekends I'm running an experiment: can a whole software company run on AI agents? I've tried versions of this before — this one's working well enough that I'm pushing it further, to see what an agent-driven company actually becomes.
→ Hero visualisation prototypes · exploring animated elements for this section
Machine-generated overview
Own the blueprint, rent the code
Two synthetic hosts read the book and talk it through: whether a software company can run on AI agents, and why the codebase is the part you rent. They get the argument right, which is either reassuring or the whole problem, depending on where you sit.
Featured Projects
Gyrum.ai
An ongoing experiment in agent-run software — agents that plan, build, review, and deploy. 93+ projects shipped so far.
About me
Decades shipping software — now a passion project: automating the entire product workflow with AI agents working Agile, build to ship to support. More about me →
Gyrum.io
Another experiment — Kubernetes-based SaaS services, multi-tenant and self-serve from the ground up. The same dark-factory idea as the book, run as a platform.
HipHip
Team recognition and kudos board. Give shout-outs, track streaks, celebrate wins.
Buzzy
Family chore bounty board where kids earn pocket money. Multi-currency, QR invites.
Cannon Drop
Physics-based browser game — idea and build by one of my AI agent companies, just for fun. Aim, fire, destroy targets.
Days Out For The Family
Find family-friendly attractions and days out across the UK — search, favourite, and discover places near you.
The Rented Codebase
The Rented Codebase
The codebase is rented. The blueprint is owned. The tests are the proof.
A working paper on the AI dark factory behind Gyrum — why the code you ship is becoming a disposable instance, why the precise specification is the only asset left, and how a factory that regenerates its own codebase measures that claim instead of asserting it.
Edition 2.0 · 30,000 words · three hours Read two chapters free
What I work with
Go, TypeScript, SvelteKit, React, PostgreSQL, Docker, Kubernetes, CI/CD pipelines, and AI across the stack — Claude, Codex, Gemini, GLM 5.2, Kimi K3 and Qwen 2.8, with ElevenLabs for voice and Imagen for images.