Skip to content
Book a call

Service · AI operations

AI operations: from pilot to agent-run function

We help you move one business function from the team that runs it today to an agent that does defined work within rules you have written down. Your team stays in place until you decide otherwise.

Last reviewed: .

An AI operations engagement moves one business function from people-run to agent-run in four stages: an assessment, an operating design, a pilot beside the current team, and then either a handover to your team or a managed run.

The usual first function is software delivery. An AI-enabled SDLC already has stages, records and a way to revert a change, so the operating rules have somewhere to sit. The control set those rules have to satisfy is the work of our AI governance consultancy.

If you are the COO or the operations lead, start from the function you run rather than from a coding tool. The assessment names who decides, what must not go wrong, and whether an agent should do any of the work, before any engineering scope is written.

Alex leads each one personally. We are not tied to any model or platform vendor. Each engagement covers one function under a written scope, so you can judge it on its own results before deciding on the next.

Tell us which function you want to move, and what must not go wrong.

Discuss AI operations
  • COOs and function owners
  • CTOs and engineering leaders
  • Risk, security and audit
On this page

What is AI operations?

In a people-run function, people make the decisions, even if they use software to do it. In an agent-run function, an agent carries out defined work within decision rights you have written down. The decisions the design reserves for people stay with them, and there is a tested way to stop the agent and go back to the previous way of working.

We work on one function at a time, because each function has its own decision rights, data, failure modes and regulators. If you change several at once, a problem in one means unwinding all of them.

The shape of the work is in AI operating model for the enterprise. How the fees are put together is in How to price and govern agent-run services. What the shift means for a buyer is in Service as software.

How the work is priced

We do not publish rates; the figures are agreed in the written scope. The fee structure changes with each stage, so an assessment never carries an open-ended share and a managed run never locks you in.

How each stage is priced
StageHow it is priced
Assessment and operating designA fixed fee, agreed before the work starts.
PilotA fixed fee, plus a bonus if an outcome agreed before the pilot is met.
Managed runA monthly base, plus a share of an agreed outcome. The share is capped. You can take the run back.

The outcome is a measure you can check, agreed before the pilot or managed run begins, and we do not suggest a percentage before we understand the function. For a managed run, the cap and the exit are written into the contract. Detail, including where the risk sits, is in How to price and govern agent-run services.

Regulatory mapping

The operating design maps decision rights and controls to the rules that apply to that function. Mapping is not legal advice, and it is not certification. Your counsel decides how the law applies.

What an engagement covers

  • Assessment

    What the function does today, who decides, what data it uses, what a wrong answer costs, and whether an agent should run any of it. Output: a written assessment and a recommendation to proceed, narrow the scope, or stop.

  • Operating design

    Decision rights, the controls, the rollback, and the regulatory mapping for the rules that apply. Output: a design your function owner, risk team and counsel can review before a pilot starts.

  • Pilot

    The agent runs beside the team that does the work now, on real work, inside the decision rights. The team keeps the decisions the design reserved for people. Output: a pilot record of what the agent did, what was accepted, and what was sent back.

  • Handover or managed run

    Your team takes the function, with the runbooks and the way to stop the agent. Or we run it as a managed service on the pricing above, and you can take it back. Output: a handover pack, or a runbook and a contract with a cap and an exit.

What you keep

  • Assessment

    The function as it runs today, and whether an agent should run any of it.

  • Operating design

    Decision rights, controls, rollback and the regulatory mapping.

  • Pilot record

    What the agent did, what a person accepted, and what was sent back.

  • Handover pack

    Runbooks, access and the tested way to stop the agent and fall back.

How an engagement runs

What we do not do

  • We do not give legal advice. Legal sign-off stays with your counsel.
  • We do not certify to ISO/IEC 42001. Certification is carried out by certification bodies. In the UK, UKAS accredits them.
  • We do not resell a model, an agent platform or a coding assistant.
  • We do not start a pilot before the operating design is written, including the way back.

Who it is for

  • Operations and function leaders who want an agent to do defined work in one function, with a named person in their organisation still accountable.
  • CTOs whose first function is software delivery.
  • Risk and audit teams who want decision rights, controls and a rollback written before anything is switched on.
Alex Vakhitov, founder of Comonad Limited

Alex Vakhitov

Founder, Comonad · AI and software architect, London

Alex Vakhitov founded Comonad in 2024. An AI and software architect, he was co-founder and CTO of Cloud KB, UK energy-market software, from August 2015 to March 2022.

Read Alex Vakhitov's profile

Questions about AI operations

What does agent-run mean?

An agent carries out defined work inside decision rights you have written down. It does not mean the function has no owner. A named person in your organisation stays accountable, and there is a tested way to stop the agent and return to the previous way of working.

Why start with software delivery?

Because the work is already staged: specify, change, review, test, release. There is already a record, and a bad change can be reverted. That is the AI-enabled SDLC. Other functions can follow once the pattern works there. If another function matters more to you, we start there instead.

Do you publish prices?

No. Assessment and operating design are a fixed fee. The pilot is a fixed fee plus an outcome bonus. A managed run is a monthly base plus a capped share of an agreed outcome, with a clause that lets you take the function back. The numbers are in the written scope.

Does an outcome fee move accountability to you?

No. A fee says how the work is paid for. It does not decide who is accountable for the function, and it does not decide your role under the EU AI Act. That follows the activity, and your counsel decides it. A named person in your organisation remains accountable.

Are you tied to a vendor?

No. We work with the models and tools your policies allow, and we design the decision rights, evaluations and controls so that they still hold if you change model or vendor.

Sources

Related

Discuss AI operations

Tell us which function you want to move, and what must not go wrong.

Discuss AI operations