The Friday spreadsheet.

Every Friday afternoon, the same spreadsheet. Export the invoices, check them against the project system, chase the two that don't match, paste the totals into a third tool that doesn't talk to the first two. An hour and a half, most weeks. Done by the most senior person in the room — because the stakes were too high to hand it to anyone else.

That was us. We ran an agency; we built software for a living. Our delivery pipeline was automated end to end — tests, deployments, the lot — because that's what builders do with repetitive work in a codebase. But the operations around the work? Manual. Invoices, document routing, reconciliation, status chasing. The business ran on hours of skilled attention spent on work a machine should have been doing years ago.

That's the busywork trap, and once you see it you see it everywhere: work that's too important to drop but too repetitive to deserve a person. It doesn't show up on any roadmap. It just quietly eats the hours you hired people to think with.

So we did what we knew. We built agents to draft, route, match and reconcile — and we kept a person at every point where a decision mattered. The agent does the typing; someone on the team approves. Not because the technology couldn't go further, but because we'd seen what happens when it does. Automation that removes people from decisions fails the same way every time: quietly, then all at once, and then nobody trusts it again.

That was the thing we couldn't unsee. Everyone kept telling businesses the answer was bigger — transformation programmes, autonomous everything, a revolution per quarter. Meanwhile the actual fix for the Friday spreadsheet was small, specific and boring: an agent that drafts, a person who approves, a number that proves it worked. The industry was selling theatre. The work needed plumbing.

Our own operations went quiet. Not dramatic — quiet. The reconciliation happened before anyone opened the file. The routing routed itself. The senior people got their Fridays back and spent them on the work clients actually paid for. Then clients started asking what we'd done, and whether we could do it for them. We could, and did, and the method held on their systems the way it held on ours — because it was never built to be a demo. It was built to survive a Tuesday.

bild is that method with a name on the door. We design and build agent workflows around the way your team already works — your stack, your protocols, your approval chains. We find the work nobody should be doing by hand and quietly take it off your plate. And we keep judgment in the room, deliberately, at named checkpoints, because that's what makes automation something a team can actually trust.

If you run operations, you know the spreadsheet we mean. Yours has a different name and lives in a different folder, but it's the same work: hours of human attention spent supervising data from one system to another. You were hired to think. Somewhere along the way, the machine work crept in and nobody gave you permission to put it down.

That's the work we take. Here's what we hold ourselves to: we build around what you have instead of selling you what we'd rather build. We agree the number before we start — hours returned, exceptions caught, volume handled — so you'll know whether it worked without taking our word for it. Your team keeps the last word, every time. And we will never dress any of it up as a revolution.

People shouldn't do machine work. People should do the good work — the judgment, the craft, the conversations that need a human in them. We build the systems that make that the normal state of things.

Starts quiet. Ends done.