An information services company that injects chemical tracers into petroleum supply chains — so governments can catch fuel dilution and smuggling — is hiring an Operations Program Manager for Bosnia & Herzegovina and Montenegro. The role sits inside their Oil and Gas Division, reports to a VP of Program Delivery, and exists to keep a national fuel-integrity program running across two Balkan markets. Read the posting carefully and you realize what this person actually does: they are the human API between a prime contractor, a government client, customs authorities, fuel terminals, local subcontractors, and a PMO back in the UK and US. That is a logistics-and-reporting job wearing a blazer.
What the role really does
Strip away the corporate values language — "teaming and collaboration," "advancing science and technology for a better world," "competing and winning in the marketplace" — and the essential functions are concrete:
- Stakeholder communication. Regular interface with the prime contractor, the client (a government entity), customs and border authorities, consortium members, and local subcontractors.
- Site oversight. Substantial travel to fuel terminals, storage facilities, border locations, and operational sites across BiH and Montenegro.
- Contractual compliance. Ensuring operational obligations are met — fuel marked at the right volume, at the right terminals, with the right documentation.
- Reporting upward. Regular program updates to PMO teams in the UK and US.
Three of those four are information-routing tasks. The fourth — site oversight — is the only one that requires a human body in a physical location, and even that is thinner than it looks.
How AI takes the role, piece by piece
Stakeholder communication becomes an LLM-managed correspondence graph.
The core of this role is knowing what to tell whom, when, and in what register. A government customs authority gets a formal compliance summary. A subcontractor gets a logistics instruction. The UK PMO gets a status dashboard. The prime contractor gets a risk escalation. Right now, a human reads incoming emails and messages, mentally routes them, drafts responses, and sends them.
An AI system does this with a stakeholder graph and a set of communication policies. Each stakeholder is a node with attributes: role, organization, contract relationship, preferred language (Bosnian, Montenegrin, English), escalation threshold, and communication cadence. Incoming messages are classified by intent — request for information, compliance alert, schedule change, dispute, routine update — and routed to the appropriate response template. An LLM drafts the reply in the stakeholder's preferred language and register, calibrated by prior correspondence. A human reviews escalations; everything else sends automatically.
The system maintains a running context window per stakeholder relationship — every prior exchange, every contractual obligation, every open issue — so no one ever has to ask "where do we stand on the Montenegro terminal issue?" The answer is already generated, updated in real time, and sitting in a shared dashboard.
Reporting upward becomes a live dashboard, not a weekly email.
The posting says the role provides "regular program updates to the Program Operations and Project Management Office teams based in the UK and US." In practice this means someone spending hours each week assembling status decks, writing narrative updates, and joining calls to walk through what a dashboard could show in real time.
An AI system generates this continuously. Fuel-marking volumes flow in from terminal systems. Compliance exceptions trigger automatically. Risk indicators — a terminal running below expected throughput, a subcontractor missing a delivery window, a customs checkpoint flagging an anomaly — surface in a live dashboard. The PMO in London opens a browser and sees the program's status right now, not last Tuesday's summary. When someone needs a narrative, the LLM writes it from the data: "Program operations in BiH maintained 98.6% of contracted marking volume this period. One exception flagged at [terminal] on [date], resolved within [timeframe]. No escalations to government authorities." The weekly status email writes itself. The monthly executive summary writes itself. The quarterly review deck writes itself.
Contractual compliance monitoring becomes automated reconciliation.
The role ensures contractual and operational obligations are met. That means reconciling what was supposed to happen against what actually happened: fuel volumes marked, terminal visits completed, documentation submitted, regulatory milestones hit. Right now a human tracks this in spreadsheets and project management tools, follows up on gaps, and escalates when something slips.
An AI system does this by ingesting the contract's operational requirements as structured obligations, then matching each obligation against incoming operational data. Was the required volume marked at each terminal this period? Did each subcontractor deliver on schedule? Were regulatory reports submitted on time? The system flags exceptions automatically, drafts the follow-up message to the responsible party, and escalates to the VP only when the exception exceeds a defined threshold or remains unresolved past a deadline.
Site oversight becomes sensor data plus exception-based human dispatch.
This is the one that sounds hardest to automate, and it is — but not as hard as the posting implies. The role requires "substantial travel throughout Bosnia & Herzegovina and Montenegro, including travel to fuel terminals, storage facilities, border or customs-related locations, and other operational sites."
Most of that travel exists because someone needs eyes on a site to confirm it's operating correctly. But fuel terminals already have systems: flow meters, tank gauges, inventory management, access controls. A fuel-marking program adds its own layer — marking equipment that records when and where tracer was injected, at what volume, batch numbers, timestamps. That data flows automatically. You do not need a person to drive to a terminal to confirm the marker is being injected if the marking equipment is reporting injection events in real time and reconciling them against terminal throughput.
What you still need a person for is the exception: a terminal where the data says everything is fine but the physical reality is different. Equipment tampering. A meter that's been bypassed. A storage facility where the paper records don't match the fuel in the tanks. For that, you dispatch a human inspector — not a program manager on a scheduled visit, but a local contractor sent because the system flagged an anomaly that requires physical verification.
The scheduled site visit is a ritual. The exception-based dispatch is the actual work. AI kills the ritual and keeps the work.
Where it breaks
Here is the honest part: the thing that does not automate cleanly is the relationship with government authorities in the Balkans.
Fuel-marking programs are government contracts. They exist because a ministry decided to crack down on fuel fraud, hired a prime contractor, and required that contractor to run a program across the country's fuel infrastructure. The relationship with that ministry — with the customs officials at the borders, with the regulators who sign off on compliance — is political. It is built on trust, on prior history, on knowing which official to call when a shipment is held at customs, on understanding that the formal process and the actual process are not always the same thing.
An AI system can draft the email to the customs authority. It cannot take the customs authority's director to lunch. It cannot sit in a meeting room in Sarajevo and read the room when a ministry official says "there is no problem" but means the opposite. It cannot maintain the social capital that makes a government program survive a change in minister.
This is also a region where business relationships carry weight that formal contracts do not fully capture. The subcontractor who has been running fuel deliveries in Montenegro for fifteen years and knows every terminal manager by name is not a node in a stakeholder graph. He is a person, and the trust between him and the program is not something you can encode in a communication policy.
So the human who survives in this role is not the operations manager. It is the relationship manager — someone whose value is physical presence, political instinct, and the social capital to keep a multi-stakeholder government program aligned when the formal systems say one thing and the informal reality says another.
But that person does not need to spend forty hours a week on reporting, compliance tracking, and stakeholder correspondence. The AI handles all of that. What remains is maybe ten hours a week of actual relationship work — calls, meetings, the occasional site visit that matters because of who is there, not what is checked.
The dare
The company posted this role as a full-time contract position. They are hiring someone to do a job that is 80% information routing and 20% relationship management, and they are paying for 100% of a person's time.
If you ran the numbers — the cost of the AI system, the cost of a part-time relationship manager based locally in BiH or Montenegro, and the cost of exception-based site inspection contractors — against the cost of a full-time Operations Program Manager plus travel expenses plus the overhead of a human doing what software does better — which number is smaller?
The AI system plus a local part-timer plus on-demand inspectors is cheaper. The other is a job posting on a board in Bosnia.
What hiring signals reveal about which roles AI is eating is not a theoretical question anymore. It is a line item in a program budget, and the line item that says "human program manager, full-time, with travel" is about to look very expensive next to the one that says "software plus a local part-timer plus on-demand inspectors."
The fuel gets marked. The data flows. The dashboard updates. The reports write themselves. The customs director still needs lunch — but that is one lunch a week, not a full-time salary.

