The company brain for businesses becoming AI‑native.


Afanasyevs builds the operating layer: the level between the systems a company runs on and the decisions it has to make — a working model of how the company actually operates, on top of the tools it already owns.



A group operating system for a seven-entity minerals trading group. The case →

The shadow loop

The barrier to AI automation is no longer model capability.

It is the absence of structured, reliable company knowledge.

Knowledge is fragmented across people, documents, channels, tickets, databases and operational systems. Humans compensate with memory and experience; agents cannot. So the written process becomes a fiction — and the real one lives in a shadow loop. Every automation built on a fictional process fails on contact with the real one.

What this is

One operating layer, above the systems you already run.


Context, rules, evidence, permissions, procedures — captured from how the company actually operates, structured, and held in one governed place. It records how decisions were made: the case, the evidence, the reasoning, what followed. From that record, the work that can be automated safely is handed to agents under explicit rules.

The goal is not to replace institutional knowledge — it is to turn it into an executable system.

How it is built

The human stays on the switch.

Interviews start it, work keeps it current

The model is drawn from interviews and shadowing, then maintained by the operations running through it — so it reflects the process in use, not the one described once.

Exceptions become precedent

What the system cannot close is routed to a named human with a deadline. The decision is recorded with its reasoning; recurring ones are engineered back into the automation, so coverage grows where the system was weakest.

Autonomy is earned in stages

The rule for moving work to an agent: its outcome must be verifiable, its actions reversible or contained, and its precedent sufficient. Deterministic where a rule will do; agents only where they earn their place.

Measured, not asserted

Two numbers to hold an engagement to: the share of cases closing without a human, and whether repeat exceptions are falling. Agreed at the start, reported as the build runs.

The engagement

Nothing is built before it is approved on paper.

  1. Diagnose

    The operating model as it actually is: process maps, an entity-relationship diagram, workshopped with your team.

  2. Design

    Scope, logic and gates approved on paper. Fixed fee — no meter running.

  3. Build

    The operating layer, the synchronisation, the interfaces, the migration of history.

  4. AI layer

    Only where a written business case justifies it — and only inside your access-control perimeter.

The system is handed over with logic your team can edit and a backlog your team feeds itself.

Selected work

Seven companies. Seven ledgers. One brain.

“This is our new personalised ERP.” — the group’s owner

An industrial minerals group across seven entities in Southeast Asia, Oceania and the EU. Seven ledgers synchronised in real time, one operating layer above them. Group sales analysis that matched the team’s own hand-built benchmark — then surfaced transactions the benchmark had missed. Six months, fixed fee.

Read the case Writing from the practice →

Are we a fit?

Recognise your business?

Most groups recognise three or four of these, rarely all of them. If several describe the company you run, the conversation is likely to be a useful one.

If some of this is your shape, the diagnostic will show where an operating layer pays for itself first — and where it wouldn’t.

The principal

One person, end to end.

Afanasyevs is the practice of Alexander Afanasyev. Clients get C-level judgement in the room from the first call — without the overhead of a team, and without weeks of discovery after which the senior consultant hands the work to someone else. EU and Southeast Asia.

  • Ex-CEO, entrepreneur. Reads a P&L and an operating model without being taught the business first.
  • Hands-on. Researches, designs, builds and launches the work himself.
  • An agentic system, not a delegated team. The capacity of a team without a junior layer between you and the work.
  • Counts money. Fixed fee; nothing is built without a business case approved on paper.
  • Finishes. The measure is a system in daily production use, not a pilot.
  • Insured. The practice carries professional indemnity and general liability cover, underwritten by Markel Insurance SE. Clients engage a covered counterparty, not an uninsured individual.
exali liability seal — Professional Indemnity for Digital Professions

Cover

Professional indemnity and general liability.

Professional indemnity
€100,000
per claim, triple aggregation per year
General liability
€2,000,000
personal injury and property damage
Territory
Worldwide
with the policy’s US restrictions
Cover period
To 2029
renewed annually

Underwritten by Markel Insurance SE, Munich; brokered by exali AG. Verify the cover at exali →

AFANASYEVS

One 45-minute call: where your group stands, what can safely move to agents, and what a first engagement would look like.

Start with a diagnostic