The enterprise builds here come from my day job, so they're anonymised: no organisation named, no app names, no screenshots. The rest are tools I built for my own business and use every day.
01 · The rebuild
Two years of developer time, rebuilt in six weeks.
The original events platform took one to two developers about two years. I rebuilt it to the same spec in six weeks, with a squad of eight AI agents I designed and ran. The squad did the development, the application lifecycle setup, and the build and deploy pipeline. I set the spec, the direction, and the review gate.
It shipped as a governed application: 12 automated workflows, a virtual agent on the front door, and source control, automated builds and environment promotion behind it. Somewhere between 50 and 200 people across the organisation use it.
Rebuilding to a spec that already exists is easier than working out what the thing should be. The original team had to discover the requirements. I already had them. That is part of why six weeks was possible.
A Power App that clinical teams use to manage quality and risk across their staff, with 25 automated workflows running behind it.
It also has a separate cut-down version for guest users. Rather than just hiding the admin screens, it leaves them out of the app entirely, so people outside the core team never load them in the first place. If you go looking, there's nothing there to find.
25 automated workflows · guest variant
clinical platform · schematic
03 · Ticket triage agent
Nobody sorts the queue by hand any more.
Half the tickets in any support queue arrive without the information needed to act on them. So someone reads each one, works out what's missing, emails the person, and waits.
I built an agent that screens every ticket on arrival. It pulls the detail it needs, goes back to the user directly when something is missing, and routes what is complete to the right technical specialist based on the subject.
every ticket screened · in production
triage · schematic
04 · Finance agent network
Repetitive finance work, carried by a network of agents.
Finance teams carry a lot of work that is exacting but repetitive. Checks that happen every month, in the same order, against the same rules.
I built a set of Copilot agents to carry it: a parent agent that understands the workflow, and child agents that each handle one narrow job. One agent trying to do everything drifts and starts making mistakes nobody catches. Narrow agents under a coordinator stay accurate, and when something goes wrong you can see exactly which one it was.
parent + specialist children · in production
agent network · schematic
05 · Flightdeck
The control room behind the six-week rebuild.
Running one AI agent is easy. Running a squad is a management problem: who is working on what, who is blocked, and who is about to overwrite whose changes.
Flightdeck is the desktop app I built to run them. Each agent works in its own isolated copy of the project on its own branch, so four agents can build in parallel without collisions. Approval prompts are detected and ranked, a review screen shows every change as a side-by-side diff before it merges, and a board dispatches work straight to agents.
The video is the real application, recorded end to end. Only the project data in it is simulated.
From bug report to deployed fix, about 80% faster.
An agent built against the clinical platform above. It knows the application's structure, so when something breaks it traces the problem, points at the cause, and the fix goes from reported to deployed in a fraction of the old time.
It's the only build on this page where I have a clean before-and-after measurement, and there's a full write-up of how it works.
