Every company is telling its people the same thing: “Start working with AI. Get ready for an agentic future.”
There is huge potential here, but today almost no one can say how. And unless you can tell your people how, with a clear path forward, expect scepticism.
So you run a pilot. You hand over AI assistants to your team, you install an agent for a few repetitive processes. The results look good: work that took hours now takes minutes. And in the end, nothing fundamentally changes.
Everyone keeps working the way they did before. When you are honest, the company is no faster than it was. Turns out those tasks were never the bottleneck. Speeding them up only made the fast part faster, with an additional bill.
“The real bottleneck is the rest of how you work: deciding what to do, getting aligned, discussing, reviewing, approvals and meetings where work sits and waits.”
All of that coordination is the cost of one thing: there is no shared picture of the work. The decisions, the changes, the reasons behind them sit scattered across your tools, and everyone assembles their own version of the truth from whatever they can find. What is missing is the one thing none of the tools hold: a shared, living context for the work itself.
The problem is not the AI assistant. The solution is something you have owed your colleagues for a long time.
The part of the work your agent can’t see
When work gets delivered, it moves along two dimensions.
The first dimension is how the work flows through the organisation. It is about the goal, the exceptions, the decisions, the dependencies, the next steps and the responsibilities. Call it the process.
The second is the work itself: the case to work on, its history, the guidelines, the templates and the expectations that come with it. It can be a document, a design, code, a presentation, a spreadsheet. Call it the deliverable.
Good colleagues operate both dimensions well. They know how the company operates, and they know how to deal with the work in front of them. A junior who knows the process but not the work, or an expert who knows the work but ignores every rule, brings less than a colleague who delivers both.
Your pilot AI assistants and agents usually focus on one of these dimensions, so they never reach the good-colleague level.
A base AI model knows the world, but knows nothing about you, or how the work needs to be done. Brilliant and helpless. Add a knowledge base and instructions on top of that and it can start generating content, but it can’t move the job forward, because it has no idea what to do next. Wire in your workflow engine and you get the opposite: it knows the steps, but it doesn’t know the work. Only when your AI can do both at once, it stops being a chatbot and starts being a colleague.
Today they can’t, because the process is described in documents while the interactions, decisions and collaboration are spread across email, Teams, Slack, meeting minutes and tickets. The deliverables live in spreadsheets, documents, code, designs and other tools.
For agents to succeed in an organisation, you need to define both the process and the deliverable in a single workflow at all levels. Sounds obvious, right? So why didn’t we do this for our people a long time ago?
Fix it for everybody, not only your agents
You never had to.
People are flexible. They hunt for the right process when they can’t find it. They ask the person next to them how it works. They improvise what the organisation left empty, gather what they need from wherever they can, and pass it on. This is what organisations expect of their people, and they will expect it of their agents too.
But an agent can’t work like that. It can’t ask its neighbour, or pick up the unwritten rules by watching how others get the job done. It only knows what you put in front of it. When an agent is uncertain, it blocks, or starts hallucinating.
So if people and agents are to work side by side and trade work back and forth, both have to answer to the same standard and use the same information about the work and the way it is done. Only then can they rely on each other.
Implementing a separate AI assistant for every task, and defining every interaction point by hand, will not scale. Information does not travel across the handovers between people and agents. And the moment you change the process, all the work you put into those agents has to be redone.
This is why the way we work itself needs to change, in a way so a person and an AI agent can finally collaborate. It is time to rethink how to work in the agentic era.
The work should be the process
What if you redesigned how you work around a single principle?
“The work should be the process.”
The process and the deliverable are two sides of one coin. Define them together, move them together, evolve them together.
Today they live apart. The process goes in one place: the SOP, the flowchart, the company wiki, where you describe the interactions, the decisions, the handovers. The deliverable itself, and what was expected of it, is handled somewhere else. Keep the two apart and you lose something at every handover.
When the deliverable carries its own process, its local context and output travels with it as external context for other depending deliverables. Anyone who touches a deliverable, person or agent, can add to that local context, and it passes down to everything built on top of it. Layer by layer, a hierarchy of context that everyone can work from.
So redesign automation around deliverables, not tasks: a document, a design, a piece of code, a roadmap. Everyone owns its deliverable and can see everything that led to it. It decides how to do the work, and it answers for the result like everyone else, aware of what is expected and how the work will be judged.
Take for example user documentation for a new software feature. It relies on everything that came before it: the feature that shipped, the design that shaped it, the user need it answers, the vision behind it all. When the deliverables carry their own process, the whole external context chain is already there when the page is written: what the feature does, how it works, who it is for, and why it exists at all. With all of this immediately available, a person or agent can write documentation that explains the feature, not just describes it, because the context around it is automatically inferred.
Now the line between what a human and a machine can do blurs. The agent drafts, a person sharpens, the agent revises, a person approves, and each responsibility moves to whoever is best for the next step, human or agent. You stop deciding in advance whether a task is for a person or an agent. You let the best suited for the job take care of it. This is the basis for collaboration. From here you move one step at a time: hand agents the parts they can run alone, keep people in full control of the parts that still need them, and support them with assistants that already hold all the knowledge for the job.
From there the work moves between hands without anyone assigning it in advance. The agent drafts from the inherited context, a person sharpens the framing, the agent revises, a person approves. Each step goes to whoever is best for it, human or agent, because they are all working from the same context the deliverable carries.
Bringing it all together
Put it all together, and the direction is clear: when the work and the process move as one, the context follows the work to whoever picks it up next, person or agent.
None of this needs a clean slate. You keep what you have built: the processes you run, the way your people work today, everything you already have in place. You bring the work and the process around it together, into one place where they finally meet. Context flows through the work instead of being stitched together by hand, unable to evolve.
This is what the agentic era will actually look like: not people on one side and AI on the other, but a new way of working that finally serves both.
This is new territory, and we don’t claim to have all the answers. If you believe this is the direction your organisation should take, let’s talk about how to get started. We help organisations rethink how work gets done in the agentic era, one flow at a time, so people and agents can really collaborate.
—Kristof Martens, Co-founder of Expedait