ARTIFI · Thesis
Finance is Sisyphus work.
Put the stone down.
The way a company runs its finance should belong to the company, and stay when people leave. What a finance person knows should belong to the person, and travel when they move. Today both are trapped inside whatever software the company happens to run. The vision behind Artifi is that nothing is lost and nobody re-learns.
01 / The problem underneath
Software built by many hands,
that cannot keep pace.
The stone is not the accounting. It is the software underneath it, and nobody chose it on purpose.
A finance function runs on a dozen systems written by a dozen vendors, each with its own idea of what a vendor is, how an approval works and where a procedure lives. None of them was built for this business, and the business changes faster than any of them can be changed.
Everything people complain about is downstream of that. The re-learning, the adjusting, the fixing and re-fixing, the procedures written three times in three tools. It is Sisyphus work: the same stone up the same hill every month, and every new system, every new vendor release, every new procedure, every new policy, every new entity, every new job rolls it back to the bottom.
Why should a controller spend a career learning software? The time belongs to the work that needs a mind: judgement, exceptions, the decisions no system can make.
Learning, again
Every new system, every new company: the same work re-learned through a different set of screens, fields and menus. Not the accounting. The software.
Fixing, forever
Transformation teams exist to keep the stack standing: integrations, workarounds, vendor updates that move the furniture. The business changes faster than any of it can follow.
Procedures that don't travel
How a vendor is validated, which approvals a payment needs, what the close checklist looks like. Each company rebuilds it from scratch in whatever tool it has, and it stays there.
Expertise that leaks
Forty years of knowing how finance works leaves with the person at every job change, and vanishes at retirement. Almost none of it is passed on in a form anyone can run.
02 / The rule
Work that repeats
should move, not be re-learned.
Most of what a finance function does happens on a cadence: every bill, every statement, every month. Anything that repeats should not live in a person's head and hands. It should move into a skill (how the work is done, written once in plain language), a plugin (the logic that must hold: a tax rule, a calculation, a policy) or an agent (the work that runs on its own when something arrives or a date comes).
That is what the intelligence is for. Not to answer questions about finance, but to take work you describe once and carry it from then on, with the procedure, the validations and the approvals intact. Nobody has to learn it again.
Two things then own it. The company owns its way of running: the procedures, the thresholds, the approvals, the agents on the cadence. Those stay when a person leaves. The person owns their craft: the specific, judgement-heavy skills that make them good at the work. Those travel with them, and when a company adopts one, and someone else improves it, expertise compounds instead of leaking.
03 / Three layers, and the one that was missing
For work to move into a skill, a plugin or an agent, and stay there, three things have to exist. In 2019 only the idea did. Today two of the three are here, and the third is what we build: the foundation that makes the move easy.
01 · EXISTS
Intelligence
A model that understands what you mean, reads the context and can act. In 2019 this had to be built by hand as dialogue trees. Today it is Claude.
02 · EXISTS
The skill
A unit of work a person can write down and hand to the intelligence: how to create a vendor, chase a bill, run a close. Today these are Claude skills, plugins and agents, in plain language, owned by whoever wrote them.
03 · BEING BUILT
The place it runs
A skill only carries if what it runs on is the same everywhere: one record model, one set of governed actions, the procedures and approvals held as definitions rather than as screens, and a simple way to move any piece of work into a skill, a plugin or an agent. This is the layer Artifi is.
The first time round, the interface had to be built because the intelligence did not exist. Now the intelligence exists and the interface is a conversation. What is still missing is the ground under it: a place where finance work is done and recorded the same way in every company, so that the skill for it can be the same too.
04 / The same idea, nine years
Written down before it was possible,
built when it became so.
This is not a thesis assembled for a pitch. It was published, in this order, and each step was tried. The essays are linked; read them against what exists now.
- 2017
Hala.ai
The first attempt at the vision, built before the intelligence existed. A conversational assistant for enterprise software that connected to the systems companies already ran, so that one way of doing a thing, creating a vendor, checking a credit limit, resetting an access, worked across all of them. Each of those became a digital skill, and Hala launched a skills store to publish and share them. The interface and the dialogue logic had to be built by hand; that was the layer missing then.
- 2019
Digital Skills as a Service
The idea written down: knowledge is a dead asset; split expertise into the smallest units, make each one a digital skill, publish it, and let it earn for its author long after they have moved on.
- 2019
Conversational analytics
One conversation as the interface to reports, orders and requests alike. The screens were never the product; the answer was.
- 2020
One space, one planet, one accounting software
Ten tools around one core system, each with its own cost, its own learning curve, its own way of doing the same thing. A convergence was inevitable; the question was where.
- 2023
Personal generative AI
The intelligence had arrived and the tools to use it had not. Skills that travel with the person, a marketplace for them, a common way to reach every business system.
- 2026
Artifi
The missing layer: a governed place to run finance that any company can adopt, so that a skill written once means the same thing everywhere. Intelligence by Claude, skills by the people who know the work.
05 / Where this goes
Skills outlive software.
Expertise compounds.
A finance person arrives at a new company with their skills, plugins and agents, plugs them into the company's, and does the work on the first day instead of learning it. A company stops rebuilding the same procedures in every tool it buys and stops paying a team to keep the stack standing. The time goes to judgement.
Over a career, what someone knows accumulates in a form that can be run, shared and improved by others, and does not vanish on the last day. That is the long version of the vision, and it is why Artifi is built as infrastructure first: the part every company can stand on, so the skills can be everyone's.
REPEATING WORK MOVES TO SKILLS, PLUGINS AND AGENTS. WE BUILD THE GROUND THEY RUN ON.
06 / Who's behind this
Built by the person who had
to close the books.
Artifi is founded and built by Andrew Rudchuk, who has worked on every side of the problem this page describes. As an accountant, an auditor and a business analyst, doing the work. As an SAP consultant for eight years, implementing and supporting the systems the work ran on. At Accenture, transforming finance functions for other companies and building products on machine learning and RPA. As founder of Hala.ai and Solution Director at Teampay, building the software itself. And at Starship Technologies, Dott and Impress, running and building finance across entities and countries. The same lesson from every seat: the tools were never the problem, and the tools were never the solution. Judgement got the leftovers, and every automation attempt either forked into unmaintainable custom code or died as a demo.
The beliefs underneath are written down: trust is the product; agents propose, humans sign; you should be able to read the rules that run your business; software you can leave is software you can trust. The platform's internals are published, not pitched — the block engine, per-org patches, dry-run mode and the effect-guard design are documented in the Building the Agent Platform series, with real confidence thresholds and the incidents that shaped them.
Disagree?
Disagree with the thesis?
Good.
Write to us — or watch it run on your own books first.
hello@ar-ti-fi.com