In 1901, two bicycle makers in Dayton, Ohio, had a problem. Their gliders weren’t producing the lift predicted by the published tables.

Wilbur and Orville Wright could have kept rebuilding their wings. Instead, they questioned the tables themselves.

The brothers built a small wind tunnel, tested hundreds of wing shapes, and recorded what actually worked. They created their own tables, based on information they could trust. Each experiment informed the next design. Two years later, they flew.

The wind tunnel didn’t design or fly the plane. Its value was in helping the Wrights turn individual experiments into knowledge they could use again and again. It gave them better information, but they still had to decide what that information meant and what to do with it. Their judgment is what turned their experiments into flight.

AI presents accounting firms with a similar opportunity. It can prepare work faster than ever before. But the real advantage isn’t simply speed. It’s what happens when an experienced accountant reviews that work, corrects it, and explains why.

If that knowledge is captured, the next piece of work can be better. Just like the Wright brothers’ wind tunnel, each new lesson learned from new information improves the next attempt, while the judgment about what is right remains with the people doing the work.

When answers are cheap, judgment is priceless. The opportunity of AI isn’t just to make the work faster. It’s to make the firm smarter by turning its people’s judgment into knowledge the whole firm can use.

The asset no ledger records

Late in busy season, a partner glances at a client file. An insurance payment has been recorded as an expense. “Not that one,” she says. “They pay for a full year’s cover every March. Prepay it, then release it monthly.”

Another partner knows which client wants a phone call before the numbers go out, and which would rather read them alone. A senior knows which unusual balances are perfectly normal for a particular business, and which deserve a closer look.

We call this wet code: the unwritten rules and client-by-client instinct that live in senior heads and nowhere else.

It is a firm’s most valuable asset, and it appears on no balance sheet. It takes decades to grow. It is the reason clients stay. And as a generation of partners approaches retirement, it is leaving one farewell lunch at a time.

It has always belonged to the firm. It has just never lived anywhere better than SOPs nobody opens and Visio diagrams nobody updates. And even when it is recorded, it rarely becomes knowledge the rest of the firm can easily find and use.

Two bad roads

For generations, the profession has passed this knowledge down through apprenticeship.

A junior prepares the work. A senior spots something wrong and explains why. The junior remembers the correction and applies it the next time. Over hundreds of such moments, experience becomes judgment.

It works. But it takes years, demands enormous amounts of senior time, and loses too many talented people along the way.

Now AI threatens to create the opposite problem. If an agent prepares all the work, the junior no longer learns by doing it. They become a reviewer. But how do you review something you’ve never learned to prepare? How do you recognize a wrong answer when it looks perfectly plausible?

A leader at a top-50 firm, who also teaches a graduate course in AI and finance, told us this week that his students can now pass his entire course using AI. What they struggle to do is recognize when an answer is wrong. That is the danger: in automating the work, we risk removing the learning that created our experts.

We shouldn’t have to choose between slow apprenticeship and blind automation. There is another way.

On a Toyota production line, any worker who spots a problem can pull a cord to stop production. But stopping the line is only the beginning. The team investigates what went wrong and changes the process so the same mistake doesn’t keep happening.

The lesson is simple: a correction should improve the process, not just fix the individual mistake.

The same should be true of AI in an accounting firm.

An agent prepares the work. An accountant reviews it. When something is wrong, they correct it and explain why. That correction becomes knowledge the firm can reuse, while the reasoning remains visible to the next person learning the work. A junior can see not just what was changed, but why, and learn to apply that reasoning themselves.

The human isn’t just there to approve the machine. They’re there to make the intelligence of the system more visible, more precise, and more useful.

What we built

This is what we’ve built Artifact to do.

Artifact is an AI platform that puts a firm’s most valuable asset, human judgment, at the center.

Artifact follows a simple loop: Adopt, Learn and Scale.

Adopt

Start with one team, working on the clients and tasks already in front of them.

Nothing needs to stop, and nothing needs to be ripped out. Agents begin preparing work the firm’s way, within existing processes. People remain responsible for approving it, and our team works alongside yours.

The point is to make AI useful from day one, without asking the firm to change everything about how it operates.

Learn

Every review is an opportunity to teach Artifact how your firm works.

When a senior corrects an entry, the important thing isn’t just the correction. It’s the reason behind it, stated in plain words anyone can read, change and keep. Instead of disappearing into an email, a review comment or someone’s memory, the lesson becomes available for future work.

Think of a sherry solera: tiers of barrels where the oldest is never emptied and each year’s new wine is blended into it. The partners’ experience stays in the record, the team’s fresh insight keeps adding to it, and nothing is poured away.

Scale

What your people know then reaches every client, every new hire and every agent. The partner’s note about the March insurance is no longer a hallway lesson for one junior. It is how the work is done, wherever it applies.

This is what makes the loop powerful.

Adoption creates opportunities for review. Reviews create knowledge. Knowledge improves the next piece of work, making it easier to adopt AI across more clients and teams.

The more the firm uses Artifact, the more of its own intelligence goes into future client work.

At the center of it all, a person still decides what good looks like.

What it makes possible

Many firm leaders we work with tell us that saving hours, on its own, isn’t a compelling enough reason to adopt AI. A firm that bills by the hour knows the problem: when five hours become one, the client asks why they paid for five. Hours saved only become anything when the firm decides what to do with them.

The real unlock begins with capacity: the same team, more clients, no more late evenings. Then it goes further. Firms that hired a hundred graduates a year now hire fifty, so keeping those fifty matters more than it ever did. The junior’s lessons return, each one carrying the reasoning of the person who made it, and a junior who learns the firm’s way faster is a junior who stays. The senior’s repeat questions stop arriving, and their hours go to the problems only they can solve. The partner’s method finally belongs to the firm, and what the team notices on each client reaches the people deciding where the firm goes next.

That is where growth lives: more time with each client, advice that reaches past the books, and new services built on what the firm already knows. For a smaller firm, this becomes the depth of judgment larger firms took decades to grow.

Nobody wants a diagnosis from an AI doctor with no doctor in the room, even when it is right. Clients feel the same about their books. The last five percent, the judgment and the relationship, is what they pay for, and it becomes more valuable as the rest gets cheaper.

Accountants shouldn’t have to become prompt engineers or trust a black box. They should be able to do what they already do best: review the work, explain what needs to change, and see that lesson put to use again.

Why we’re here

Every firm we talk to is carrying the same three things at once. First, fewer people are coming into the profession than it needs, and the partners who hold the firm’s know-how in their heads are starting to retire. Second, the pressure to put AI to work has never been higher, from clients, from competitors and from staff who already use it at home. And, third, the underlying AI models are becoming widely available, and getting access to powerful AI is no longer hard or much of a differentiator. The hard part is getting AI to work the way your firm works, and trusting what comes back.

We built Artifact for that hard part.

The way a firm works is its intelligence: how its partners think, how each client is handled, which exceptions are allowed and why. It has always belonged to the firm. It has just never lived anywhere better than a senior’s head, an SOP nobody opens or a diagram nobody updates.

Artifact is where it lives. We capture it from your people’s reviews, keep it in plain words your firm can see, change and take with it, and put it to work in every agent the firm uses. The models will keep improving for everyone. What they learn from your people stays yours, and it compounds. That’s the advantage we want firms to build: not simply access to better AI, but a growing body of institutional knowledge that makes their AI more useful, more reliable and more their own.

That is the triangle, and it answers the three questions every firm asks us.

(1) Is it safe to start? Adopt: one team, from the next close, a person approving every step and one of ours beside them. Agents which work the firm’s way, from day one.

(2) Why Artifact, and will it last? Learn: every review joins a record the firm owns, so the agents work more like your people with every close.

(3) Is it worth it? Scale: what the firm knows reaches every client, every new hire and every agent, and growth comes out of the loop. Capacity first, then deeper client relationships, specialized services and work that drives client revenue.

We are not the only company with agents, and we will not pretend to be. We are the one that started from the firm’s judgment rather than from the model, and built the product and the people to make it stick. It is a slower way to build. We think it is the only one that lasts.

Rafael Casas spent twenty years beside this profession, at the companies that build its software and the bodies that train its people. He joined us because he had reached the same conclusion: the firms that do best in the next decade will own how they work and keep their people at the center of it.

So we make three promises.

A person decides

Agents prepare the work, and nothing is posted to a ledger until someone at your firm has approved it.

What you teach stays yours

Everything Artifact learns is written in plain words that your firm can read, change and keep. It is shared with no other firm, and it is secured like the client data it describes.

We will be beside you

Adoption is the hard part, so our people sit with yours, from the first team to the whole firm.

The Wrights learned from every experiment without surrendering their judgment. Toyota turned every correction into an opportunity to improve the line. We’re building for accounting firms to do both: keep the decisions with their people, and make what those people know improve the work that follows.

Ariel Harmoko, co-founder and CEO · Rafael Casas, Head of Deployment