James O’Hara
My work is making other people’s weeks shorter. Not as a slogan, as the actual job. I sit with a team, watch how the week really runs, find the work that eats it, and build AI that handles that work. The inbox that takes the first hour of every morning. The report someone assembles by hand every Friday. The court date that gets checked manually because nobody trusts the calendar. Small, boring, weekly things that quietly cost a business its best people’s best hours.
The short version: 20+ years in ops. 3 years at Meta. The longer arc is below, but the through-line has never changed. Every job I’ve had came down to watching how a team’s week ran, finding the hours that leaked, and building whatever stopped the leak. For most of my career that meant systems and people. Now it means AI, and the AI is honestly the easy part. Knowing which hour to go after is the job. I do both ends myself: pick the hour, write the code.
Most recently at Meta, where I built the daily-brief AI that 75+ ops people run before standup. Seventy-five people, an hour each, every working day — that’s the size of thing I go hunting for. Nobody was told to use any of it. They run it because it pays for itself before the day starts, and that’s the only adoption test I trust. I’ve also written up what I mean by applied AI, if you want the thinking behind the category.
Where the 20 years went
I started on a bank’s trainee management track at eighteen and never really left operations. Two decades of it: support floors, hosting companies, infrastructure and program teams, first in the UK and then the US. Same job at every stop — find the work eating someone’s week, build the thing that makes it stop. I helped grow companies from a hundred grand to a quarter of a billion in revenue, and one of them sold for $350M with me running operations through the build-up. The last three years were at Meta, running ops programs for a research org and building the AI its teams now start the day with. Autohmate is the same job, sold directly.
Proof of work
There are plenty of people who can advise on AI, and plenty who can build it. The overlap is thin. I’ve run operations for twenty years and I write the code myself — which means the plan and the build are the same person’s problem, and nothing gets lost in a handoff.
The clearest evidence runs every day: a production system of AI agents that handles my own operation and burns through 100M+ tokens a week on real work, not demos. It also built this site. The launch review alone ran 86 agents arguing with each other before a finding reached me, and I made the calls between rounds. What the same way of working produced at Meta: five hours back, every week, for 75+ operators, and a thirty-tool stack cut to three. The same program took procurement approvals from ten days to two, and produced the custom AI tool I product-owned and built to move the budget platform off Airtable. All of it for a 300-person org with a $120M+ budget.
The rest of the shelf, one line each:
- fleetctl, an API-first container platform that runs my infrastructure.
- failchain, a dependency-discovery engine that maps what breaks what before it breaks.
- A Telegram bridge that runs my AI sessions from my phone.
- A fitness app my household uses every day.
- And this site, built, reviewed, and shipped by the system above. The full story is an essay.
The small daily tools live in the guides.
Working together
I work in short, scoped engagements. I learn how your week actually runs, we agree what’s worth changing, I ship the build with you, and I come back a month later to see if it stuck.
Most engagements run a few weeks end to end and are scoped per project; typical builds land in the four-to-low-five-figure range, and you know the number before anything starts. If your problem is smaller than that, write anyway. Sometimes the right answer is a one-day fix, and I’ll tell you so. If the work doesn’t land, if it hasn’t stuck when I come back, you don’t pay for it. Stuck is the test we agreed in writing before I built anything. And there are things I won’t build: decks that go on a shelf, pilots designed to demo well. If it isn’t part of someone’s real week within a month of shipping, it doesn’t count.
You own everything I build. It runs in your accounts, on your stack, and I hand it over so your team can run it without me. If you want me back for the odd check after that, we agree it up front.
The longer version, stage by stage: how I work →
References
References from past work available when we talk.
Old-school: [email protected] works too.