How we grow with our clients

AI getting stronger does not make human intelligence less important. It makes it more exposed. What we build with you uses AI to take machine work off your people, and then it belongs to you. So the practice can only grow one way: by building things people are glad to own, and by getting better at that each time.

Being an expert in your niche, and seeing the whole system the niche sits in, and having built real things with your own hands, is rare, and it's the combination we hire for. The expert fixes the wound; the systems reader maps the body, and in our people those are the same person. The partners you work with bring or define the client's domain and own the relationship; elegant.work sits behind them with the diagnostic, the AI layer, the memory of what worked and the patterns for fixing things, and makes the work sharper, faster and more reusable.

All of us grew up in chaotic, bleeding-edge impact projects, running things without money alongside people who were idealistic and frequently allergic to following orders, so doing the same work with a budget, inside a conventional organisation, is playing the game on easy mode. We are not consultants. We're the opposite of consultants, which, on a site that sells consulting, is a certain irony.

The code, the content, the accounts and what your company knows live in your accounts, under your name. The logic can be read by any competent person. The AI models underneath are replaceable, and so are we.

The test we build toward: if elegant.work vanished tomorrow, would the thing keep working, would your people understand it, and could another competent person improve it? A manual doesn't answer that. Your people answer it, by running it, handling the exceptions, recovering it when it breaks and changing it safely, and a second person taking over from the first. We have not yet watched a client team pass that test from start to finish; it is the next thing we are proving.

Every engagement teaches us something about how companies get stuck, and we keep it: the methods, the tools, the checks and what the last build taught us, with nothing in it that identifies you. That is learning at our clients' expense, which consultants usually do and hide; we would rather say it out loud, because it is also the reason every next build is faster and cheaper than the last. Anything you need to run your business stays with you; our advantage is how well we build. Whether that scales like a product is still an open question.

The whole game is aligning interests long-term. The usual consulting deal is misaligned from the first day: the longer you need the firm, the better the account. Ours is built the other way round, in three parts.

First, every piece of work has an end and we step out on purpose; when you find the next thing that is stuck, we start a new, finite piece of work from there, so trust adds up over the years while no single piece of work makes you dependent on us.

Second, where what we built generates measurable value, we want to be paid in part out of that value, as a capped share agreed separately and read from your own records. The June version of this, in his words: a success fee on a defined share of year-one savings.

Third, and this is the part that keeps you learning: if we participate in what your agents generate, it is in our own interest to keep those agents current as the models and your business change, so we will do that for free or at cost. We don't know yet which; it depends on the first deals. Any ongoing support beyond that is a separate, cancellable agreement, and cancelling it leaves everything running. Recurring money is allowed. Recurring dependency is not.

Nobody should pay for the deep work up front, so first projects are fixed: scope, price, milestones, what is excluded and how we will both know it is done, agreed before we start. You pay the third parties directly, the media, the model and API use, the infrastructure, so you can see those costs and we are never financing your business.

For the larger builds that come later, the principle is that we want to be paid in full only while the work keeps deserving it. "Keeps working" is not our word against yours. It is a short list of checks agreed in advance that either pass or fail, read from your own systems: the workflow runs, the data is current, it recovers from a failure, it does only what it is permitted to do, nothing in it depends on an elegant.work account, and your people can operate it. Each check is recorded with its source and timestamp before any money moves, you own that record, and it must never become a way of locking you in. Business results, pipeline, acquisition cost, gross profit, are measured too, but they decide only a separately agreed, capped share on top, never whether the base fee is paid, because too much of them sits outside what we control.

Who carries what: if the system doesn't run as specified, if the handover was inadequate or your people were trained and the training didn't take, that is ours to fix at our cost. If your side knowingly switches off something the system needs, changes the process without telling anyone or withholds access, the affected check is paused, and we agree in advance how that is handled. "The humans fucked up the process" cannot become a lazy escape hatch for us. And none of this is live yet: today, no part of any fee waits on later results, because the record you would need to check it from does not exist until we have built it with you.

If the practice only works because Carsten carries all the context in his head, we have not built what we sell. So the way we take in work, decide and ship a change runs on the same kind of system we would build for you, and we log where a human still has to step in. It does not fully pass yet.