AI-native operations

Agents run the process.
You keep the gates.

Most consultants map your process and hand you a deck. I rebuild it and run it.

Three processes rebuilt. Three industries. One person.

The problem

Sales quotes a date.
Operations cannot hit it.
Procurement finds out last.

Everyone has a dashboard showing the seam. Almost nobody can change the process itself, and I want to be fair about why: changing it means changing quoting, planning, procurement and customer service at once, across systems bought in different decades by people who have left.

That is not incompetence. That is the actual difficulty. So the seam stays, and the margin leaks through it every quarter.

What changed is not that a model can predict a date. Forecasting was never the hard part. What changed is that an agent can now run the whole path, and stop at the three decisions a human should own: the price, the promise, and the payment.

3
processes rebuilt so agents run them, in 3 unrelated industries
20 yrs
in industrial operations, planning and consulting
9
live production stacks operated, by one person
3,400+
autonomous agent runs, in six months, in one of them

Figures from live commercial systems, measured September 2026. Every one traces to a command you can watch me run — see how.

The loop

Four steps, and the hard part is the last one

Every business process is this loop, whether it is quoting a shipment, scheduling a plant, or approving an invoice. Getting a model to draft the answer is the easy half. The engineering is in what happens before anything is allowed to become real.

01

Intake

Work arrives the way it actually arrives: an email, a PDF, a rate request, a spreadsheet, a phone call, a message in whatever language the person speaks. It gets read, transcribed and structured without anyone retyping it into a system.

02

Triage

Requests get classified, routed and grouped. Anything too ambiguous to act on goes back to the sender as a question rather than becoming a confident wrong answer. Ambiguity is parked, never invented away.

03

Execution

An agent does the work against your real systems: the ERP, the planning tool, the mailbox, the document store. Every action is logged with the reasoning that produced it, so an auditor can follow it later.

04

Gates

Three things never commit without a human: the price, the promise, and the payment. Everything else runs. Rollback is prepared before the change, health is checked after it, and one failure never takes the rest down with it.

This is not a diagram of something I would like to build. It is running in three companies right now, and I would rather show you fifteen minutes of it working than send you a deck.

Ask for the demo →
Where I work

Two ends of one broken loop

Most companies have automated pieces of each of these and connected neither. The gap between what you promise and what you can actually deliver is where the money goes.

Industrial & supply chain operations

Planning, scheduling, procurement, order-to-cash, exception handling. One end of the chain, and the end that inherits every promise made at the other.

Quote to contract

Quoting, pricing, proposal generation, contracting. The other end of the same chain, and usually the one nobody has connected to the first.

The seam between them

Where the promise meets the plant. Sales commits a date, operations inherits it, and nobody owns the handover. This is the expensive part, and it is the part I have spent twenty years on.

And when it fits, the rest of it

  • Custom AI builds
  • Process automation
  • Document processing
  • Team training & workshops
  • Legacy modernisation
  • AI strategy reviews

Not the headline, and not what I would lead with. But if you are ready to pay for it and I can do it well, the answer is yes. If I cannot, I will tell you that instead.

Not sure which end is costing you more?

Most companies are wrong about which side the money is leaking from.

Let's find out together →
Engagements

Four things, with a start and an end

Fixed scope, fixed price, defined exit. No retainers, no staff augmentation, no open-ended transformation programme.

Process Diagnostic

2 weeks

I map one end-to-end process and instrument it. You get an honest answer on where an agent can own a step, where a human gate has to stay, what the loop costs to run, and what the current seam is costing you in margin, rework and late delivery.

You get: One document with numbers in it, and a 90-minute readout where we decide together what is worth doing.

Pilot Loop

6 to 8 weeks

One real process rebuilt so agents run it, in your systems, in production, with gates and a rollback. Not a proof of concept and not a sandbox. A loop that keeps working after I leave.

You get: The running loop, its runbook, and a before and after on the one metric we agreed up front.

Operating-model rollout

3 to 6 months

The full playbook installed across the value chain: intake, triage, agent execution, approval gating, watchdogs, health alerts, and the escalation paths for when an agent gets something wrong. Your team owns it at the end.

You get: A process your people run and understand, not a dependency on me.

Enablement

1 to 3 days

Workshops for operations leadership and their teams. What agent-run process actually requires, what it costs to run, which parts are genuinely hard, and which vendor claims to ignore.

You get: A team that can tell a real plan from a demo.

Most engagements start with the diagnostic. It is deliberately small, and it is also how we both find out whether this is worth continuing.

Start with a diagnostic →
Who you get

Mehmet Kaya

Founder, Zeroth SRL · Brussels

LinkedIn →

I spent ten years at PSI Metals building production planning and due-date quoting systems for steel producers in Germany, Turkey, Poland, Brazil and Indonesia. Due-date quoting is exactly the seam this business is about: what sales is allowed to promise, given what the plant can actually do. Real systems, real factories, real consequences when the schedule was wrong.

Then five years at McKinsey as an Engagement Manager, on operations, advanced planning and digital transformation for global clients. I have written the recommendation deck, and I have been the person who had to make it work at 2am. That combination is unusual, and it is most of why this works.

Since leaving I have built three businesses whose core processes are run by agents in production, across freight forwarding, enterprise sales and software delivery.

The largest is Talent-Ray, where I am co-founder and Head of Technology. Backed by NVIDIA and Google, live with more than 20 companies within three months of launch, running in nine languages on 42 background workers, with integrations into SAP SuccessFactors and national identity verification. I am not describing something I read about. I run it.

Zeroth is how I bring that into other people's operations, alongside a small number of engagements.

MBA, INSEAD. MSc in Mechatronics Engineering, UCLouvain (École Polytechnique de Louvain).