Where we really are

Before changing anything, an honest snapshot of the starting point. Partly because that is just good method: without a baseline, every future claim of improvement is a nice story. But mostly because of what the snapshot actually showed.

At the level of individuals, AI is already everywhere at Findev, running on corporate subscriptions the company is glad to pay for. It drafts documents, reads incoming contracts and flags inconsistencies, handles the small calculations we used to open a spreadsheet for. And people use it at work exactly the way they use it in life: ask a question, get a thing done, find a thing. It answers when asked and stays silent otherwise. It will not notice a problem by itself, and it will never come to you first. Nobody ordered any of this from above. It spread on its own, quietly, I guess the way email once did.

But the picture is not uniform, and this is the part worth being honest about. Adoption follows people, not org charts: where a team has a pioneer, usage is real and growing; where the doubters sit, it is close to zero. Some colleagues run half their working day through AI and cannot imagine going back. Others tried it, shrugged, and told us plainly: it does not work for me. We take that answer seriously, because if two decades of automation projects taught us anything, it is that forcing a tool on a person is the fastest way to kill the tool. So today, AI adoption at Findev is exactly what voluntary adoption looks like: enthusiastic in some teams, absent in others.

On client work, the story goes much deeper. Our engineers, within each client's policies, have moved well past chat windows: agent pipelines, in-house MCP servers connecting models to real systems, whole toolchains built around this. Though even there, not everywhere and not evenly. That is its own large story, and not the one this series is about.

Because here is the third layer, and this is where the snapshot gets uncomfortable. So far, the company's own part in its AI story has been that of a sponsor: it pays for the subscriptions. The machine that runs Findev itself is essentially untouched. Invoices are still reconciled with clients and counterparties in Excel, with formulas. User accounts and access policies are still provisioned manually, one request at a time. Books close, payments go out, people get hired, all roughly the way it worked three years ago. Not because anyone forbade AI from touching these processes. Simply because making it touch them was never anyone's job.

So this is where we really are, and we suspect many companies would recognize themselves in it: every person has an AI assistant, and the company itself has none.

Which makes us wonder: is this what people mean when they say they are already an AI-first company? Judging by the details of some conversations we have had, for quite a few of them, yes. But for us, this does not count yet.

That gap is the story. A personal assistant works because the context lives in your head, you feed it yourself one prompt at a time, and you are always the one doing the asking. A company has no such head. Its context is scattered across mailboxes, spreadsheets, contracts, and the memory of the person who has been reconciling those invoices for years. Buying more licenses does not close that gap. Something else does, and at this point of the story we did not yet know what.

So we decided to learn from others who started the journey ahead of us.