Why we build it this way — the beliefs underneath the product.
I spent fifteen years inside finance functions — at Accenture transforming them for other people, then at Starship Technologies, Dott, and Impress running and building them. Every one of those years taught me some version of the same lesson: the tools were never the problem, and the tools were never the solution. The problem was always where human attention went. The best controllers I worked with spent their days re-keying data between systems that didn't talk to each other, and their judgment — the thing they were actually hired for — got the leftovers.
Artifi is my answer to that. Not "AI for finance" as a feature — a different set of beliefs about what a finance function is, applied with the technology that finally makes them practical. Here are those beliefs. If you disagree with them, Artifi is probably not for you, and I'd rather you find that out on this page than three weeks into an onboarding.
1. Trust is the product. Everything else is interface.
A finance function has exactly one non-negotiable output: numbers that can be trusted — by you, your board, your bank, your auditor, your tax authority. Double-entry bookkeeping has survived five centuries not because it's elegant but because it's self-verifying: every entry has a counterpart, every balance has a proof.
AI raises the stakes on this, it doesn't lower them. The moment software can act — post entries, pay vendors, file declarations — the audit trail stops being a compliance checkbox and becomes the product itself. That's why Artifi is, at its core, a real double-entry, multi-entity ledger where every event is recorded with actor, timestamp, and payload — including the agents' events. The conversation is how you work. The ledger is what you're buying.
2. Agents propose. Humans sign.
We believe automation in finance should mean removing keystrokes, never removing accountability. Every entry in Artifi is proposed by an agent and posted on a human's sign-off, through approval lanes that scale with risk — routine entries flow, large or unusual ones wait for the right person.
This isn't a temporary concession to make people comfortable until "the AI gets good enough." It's the design. A signature is information: it records that a specific person looked at a specific thing and accepted responsibility for it. No model, at any capability level, can carry your responsibility. Pretending otherwise is how AI finance products will eventually hurt someone.
3. You should be able to read the rules that run your business.
Somewhere inside your current accounting software is the logic that calculates your payroll taxes, times your revenue recognition, and assembles your VAT return. You cannot read it. Your accountant cannot audit it. When it's wrong — and it is sometimes wrong — you find out from a penalty letter.
We think that's an absurd thing for a business to accept, and it's why Artifi's accounting logic ships as open Plugins, written in language an accountant can read the way they'd review a junior's working papers. Estonian payroll: income tax 22%, social tax 33%, unemployment insurance — it's right there, in text, versioned, with history. If your firm's policy differs, you change the text. You own the logic. You were always supposed to.
4. The interface was never the work.
Forms, menus, status dropdowns, the 47-field invoice screen — none of that is accounting. It's the cost we paid to talk to databases when the only language software understood was clicking. The screen was a workaround, and the workaround became the industry.
That cost has gone to zero. You can now describe what you need — "post the October invoice, allocate by headcount" — and a system can do it, show its work, and ask for your signature. So Artifi has no forms. Not as a gimmick: as the honest conclusion of the premise. The work happens where you already are — Claude, Excel, your IDE — and the books follow you, not the other way around. (How that works in practice is its own essay: Claude Has Four Surfaces.)
5. The accountant gets promoted, not replaced.
Run the previous four beliefs together and a job description changes. When agents do the volume — the importing, matching, classifying, calculating — what remains is exactly the part that justified the profession in the first place: judgment, exceptions, the close that someone signs, the conversation where a founder is told a hard truth about their runway.
We build for that accountant. Validator, not operator. It's also why we work with accounting firms rather than around them: agents do volume, their people own judgment, their margin per client goes up. The disintermediation story other AI products whisper to investors — "we remove the accountant" — is, we think, both wrong and a bad way to build trust in a trust business.
6. Software you can leave is software you can trust.
Lock-in is a confession. If a product needs to hold your data hostage to keep you, it's telling you what it thinks of itself. Artifi's plugins are MIT-licensed and forkable. The ledger is yours, exportable, in formats anything can read. We connect to your existing systems instead of demanding you abandon them — keep QuickBooks, keep NetSuite, route work through Artifi one workflow at a time, and consolidate only if the results earn it.
We'd rather be kept by results than by switching costs. That constraint makes us build better, and it's the only honest posture for a company asking to hold your books.
7. Proof over promises.
The AI era has an evidence problem: every product is "revolutionary," every demo is staged, every percentage is unsourced. In finance — a domain whose entire purpose is verifiable numbers — that marketing tradition is disqualifying.
So we post the receipts. The walkthroughs on this site are real runs against a real ledger — actual reconciliations, actual payroll, actual close, screenshots and all. Every number on our homepage is one we can show you the source for. When something on this site is a plan rather than a fact, we date it. This essay is part of the same practice: these are the actual beliefs, written down, so you can hold us to them.
The vision, plainly
Where this goes: every business gets a finance function that runs itself and is readable by the people it serves. Not a cheaper back office — a different kind: books that are always current because agents never sleep, logic that is always auditable because it's written down, decisions that are always owned because a human signed them. The CFO of a 50-person company gets the operational depth that used to require a department. The accountant gets work worthy of their training. The auditor gets a trail instead of a shoebox.
The strategic case for why this happens now — why the durable layer is the ledger and the open logic rather than the screens — is in the thesis. The practical map of where the work actually runs is in Four Surfaces. This piece is the part underneath both: what we believe, so you can decide whether you believe it too.
Andrew Rudchuk is the founder of Artifi. Disagreement welcome — hello@ar-ti-fi.com.