Use cases to production, yesterday

Three years ago, in one of Europe’s smaller markets, a very large automotive manufacturer asked a reasonable-sounding question with an unreasonable deadline attached.

Show us something magnificent. Something that would prove AI was more than a slide. They were not short of ambition. They were short of use cases that survived contact with production reality. We were among the voices saying the threshold had shifted, that the next twenty-four to thirty-six months would look different, and that it was worth treating the moment seriously.

So we showed something small on purpose.

Take an unexplained line on a corporate card feed. No merchant name you recognise. No obvious category. Paste the fragment into an early model (GPT‑3 era, before the vocabulary of agents existed) and ask what it might be. The system would guess, often plausibly. Petrol forecourt. Motorway services. A vendor string that maps to something a human would have inferred in ten minutes with a search engine and a policy PDF.

It was not magic. It was a shift in default. The machine could offer a first hypothesis fast enough to change how people thought about triage, routing, and exception handling. We spread that story internally, highlighted a handful of patterns where similar inference might compound, and suggested the boring work too. Architecture. Infrastructure readiness. Governance conversations. Risk assessment before scale.

The answer, politely, was not yet. Details later. Readiness could wait. The loop would remember who had been useful when urgency arrived.

The loop closes

The same class of enterprise is back now with a different verb.

Production. As soon as possible. Like yesterday.

Not a pilot. Not a lab. Not a centre of excellence slide with a footnote about phase two. Use cases into projects, into runbooks, into whatever counts as “real” this quarter. APIs on the table. Results expected. Leaders who did not move fast enough on AI described, in corridor language, as not embracing change.

From outside, it is almost comic. The organisation that wanted magnificent proof before platform spend now wants production before governance catch-up. The team that was kept out of the architectural conversation now owns the integration risk. Time did not remove the trade-offs. It compressed them.

I do not think this is unique to automotive. I think we will see much more of it.

Two risk appetites in one building

Inside the same headquarters you often find two populations.

One group has higher risk appetite, or higher urgency, or both. Bonuses, board narratives, and vendor timelines align. Shipping something visible feels rational. Governance is framed as friction. Compliance is something you bolt on after the demo wins.

Another group understands reputational damage, regulatory attention, and the cost of a use case that works in a video but fails under audit. They are ringing bells now. They want governance tracks, compliance clarity, and board-level assessment that can push business and technology back from poorly thought use cases without sounding anti-innovation.

Both groups can be right about different things. The failure mode is when only the first group sets the clock.

What actually helps when the clock is loud

When everything must land yesterday, the temptation is to fund one isolated use case and declare victory. That can work once. It rarely compounds.

What compounds is a small set of use cases that share technological patterns. Agents with explicit skills. Memory and tools behind a platform boundary. The same intake path, the same assurance ladder, the same way to promote a workflow from experiment to production without reinventing security review each time.

Bulk, in this sense, is not a big-bang programme. It is deliberate reuse. Two or three patterns that rhyme, deployed with shared plumbing, so success in one lane raises the odds in the next. That is how you increase the chance of a real operating change instead of a trophy pilot.

You still need people who can argue for that in rooms where the CEO wants APIs this month. Subject matter experts who understand model limits, data boundaries, and what “production” means in a regulated enterprise. Not to stop the work. To steer it toward use cases that survive scrutiny and can be repeated.

Platform beats heroics

The organisations that will weather this wave are not the ones with the flashiest single demo. They are the ones building a platform posture early enough that the fifth use case is cheaper than the first.

That means architectural understanding even when nobody wants to fund it yet. Infrastructure that can host agents with skills, not just chat widgets. Governance that can say yes with conditions, not only no with slides. And honesty about where you are on the maturity curve, so “production yesterday” does not become “production theatre” by Friday.

The market will keep looping. Proof demanded, then production demanded, then damage control demanded. The teams that map patterns, govern in proportion, and ship in small related batches will look boring until they look inevitable.

Field note from the build log. If this loop sounds familiar, you are probably not alone.

Questions: business@vmcorp.cz