Ten months before a major scientific congress, the real work begins. The deadline is hard. The date won't move. Between now and then, parallel workstreams have to run in sync and stay in sync — strategy, scientific, medical, marketing, regulatory, legal, logistics, educational, external engagement. Each one has gates. Each gate affects the others. A regulatory decision in month three changes what can be launched in month ten. A strategy shift in month six changes what's already been planned. A legal review reopens a question the medical team thought was closed. Trade-offs are constant. None of these workstreams report to each other.
Behind the workstreams sit the stakeholders. Internal: medical, commercial, legal, executive sponsors. External: faculty, congress organisers, HCP audiences who'll attend, partners whose participation has to be negotiated months in advance. Some want different things. Some want the same thing for different reasons. Most don't talk to each other directly. The congress itself is a single moment. Everything before it is sequence. Everything after it is activation — months of follow-on work that determines whether the congress was actually valuable or just expensive.
Almost none of this is what people picture when they hear "congress" or "event." Logistics is one workstream. Event management is one slice of that one workstream. The work I'm describing is the thing holding all the workstreams in relation to each other. It doesn't have a name that fits.
A category of work has been quietly accumulating names for a decade. Chief of Staff moved out of government into the private sector. BizOps came out of Silicon Valley. Integrator from the EOS playbook. Platform PM from product. Orchestrator from the digital and AI corners. The titles aren't synonyms. They overlap at the edges and disagree in the middle. That's the signature of a category the market hasn't resolved.
What they share is shape, not job description. Someone holds a complex system together while other people build the parts. Translation between functions that don't share vocabulary. Sequencing dependencies so the whole thing doesn't deadlock. Holding ambiguity longer than is comfortable, because premature decisions break more than they solve. Knowing when to escalate and when to absorb.
I've been doing versions of this work for fifteen years under different banners — biomedical engineering, materials science, innovation management, scientific consulting, platform building, product ownership, medical affairs education. The unifying thread isn't the domain. It's building things from scratch inside systems that weren't waiting for them. Each domain gave the work a different label. None of them caught the shape.
This would be a vocabulary problem and not much else, except for what AI is doing to the rest of the work.
The legible parts of most jobs are getting compressed. Drafting. Analysis. Synthesis. First-pass research. Structured outputs. Anything that fits inside a prompt. This is the Bundled AI Bet — the gap between a specialist and the tools they have access to is closing fast.
The work between functions doesn't fit inside a prompt. Most of it is context nobody wrote down. Which stakeholder is in a turf war with which other stakeholder. Why the last attempt failed. Which executive says yes in a meeting and reverses it in Slack. The gate that looks optional but actually isn't. The conversation that has to happen before the email gets sent.
You can't extract that into a prompt because nobody extracted it into text in the first place.
So a thing that was always there, and always undervalued because nobody could see it, becomes visible. The work didn't change. The cost of everything around it dropped, and what's left is what matters.
Agentic AI sharpens this further, but not the way the common framing has it.
The current line goes: agents will execute, humans will orchestrate. As if orchestration is a new job created by agents. It isn't. It's an old skill the agents are exposing.
The part that gets missed: you can't shape an agent system if you don't already think in orchestration. The skill comes first. Before the agents. Before the deployment. Before the architecture diagram. Someone has to know what coordination actually looks like — where the seams are, where the handoffs break, which dependencies can run parallel and which can't, what context every agent needs and where that context is stored. That thinking happens in a human head. The agents are downstream of it.
People who can't orchestrate without AI won't orchestrate with it. They'll build agent stacks that look impressive in a demo and fail in the gaps they didn't know existed. The gaps were always there. They're just more expensive now.
The naming question matters because a skill that gets a different label in every domain it appears is a skill being rebuilt from scratch in each place. People doing it can't find each other. The practice can't compound across the boundary.
Somewhere right now, ten months before a congress, the workstreams are pulling against each other and the date won't move. The same shape is being recreated inside every serious attempt to build with agents — same work, same seams, same trade-offs, just with more parallel executors and fewer people who can see the whole. Markets price what they can name. They're about to price what they can't do without.