The Agent Is the Easy Part: Why Agentic AI Needs a Different Kind of Team to Change How Work Actually Gets Done
Everyone is asking some version of the same question: what does it take to be successful in the AI age?
The answer can sound very modern: data, platforms, models, agents, automation. But in practice, much of what makes AI transformation work is rooted in the same capabilities that have made people successful throughout history: the ability to collaborate, learn, adapt, make judgments, and move together toward a shared outcome.
That is why the people side of agentic AI matters so much. The question is not only what the technology can do. It is what roles, mindsets, ways of working, and operating models are needed for AI to change how work actually gets done.
It has become remarkably easy to build an AI agent that does something impressive. Give it a set of documents and it can extract information, compare sources, draft an analysis, monitor for changes, or recommend a next step. With the right tools and a reasonably well-defined task, teams can move from an idea to a working prototype faster than most of us would have imagined a few years ago.
But the challenge is increasingly not whether an agent can perform a task. It is whether the organization can redesign the work around it and align the people needed to make that redesign successful.
Many companies already have much of the talent they need for AI transformation. The harder work is often less about acquiring entirely new skills and more about changing mindsets, building trust across functions, and enabling domain, technology, and change teams to work better together.
A real business workflow rarely consists of one clean task. It crosses systems, data sources, approvals and teams. It contains exceptions that are not written down, decisions that depend on experience, controls that exist for a reason, and handoffs that may have accumulated over years. It can also expose tension between business goals and technology goals if the two are not aligned from the start.
Put an agent into one part of that process and you may make a task faster. That does not necessarily make the workflow better.
This is why we believe the next phase of agentic AI requires a different delivery model, one built around people, process, and technology together.
AI Projects Tend to Encounter Three Gaps
Consider a credit analyst reviewing a company.
AI can extract financial information, summarize filings, monitor news, identify changes, and help draft an assessment. On paper, there is plenty to automate.
But an experienced analyst sees much more than a sequence of tasks. Which sources are trustworthy? What happens when two numbers do not reconcile? Which changes are material? When should an exception be escalated? Which parts of the analysis require judgment rather than retrieval? How does the output fit into the next approval or decision?
This is the domain gap.
You can describe a workflow in a requirements document, but it is difficult to capture every judgment, edge case and practical constraint that someone who has performed the work for years understands instinctively. If domain expertise sits outside the build, those details tend to emerge late, when they are more expensive to address.
Then there is the engineering gap.
A prototype usually has the luxury of a controlled environment. Production does not. Enterprise agents need to operate across existing data, systems, permissions, APIs and security requirements. They need context. They need controls. They need to handle exceptions. And they need to work reliably enough that someone is prepared to use their output in a real decision.
That work cannot simply begin after the business has finished designing the solution. The technical possibilities and constraints influence the workflow itself.
Finally, there is the adoption gap.
Imagine that the technology works exactly as intended. The analyst now receives an AI-generated first draft, with the underlying evidence and potential risks already surfaced.
What happens next?
Does the analyst check everything anyway? Are existing review steps still necessary? Who is accountable for an error? Has the time saved simply created another approval bottleneck somewhere else? Does the user understand where the agent is reliable and where their own judgment becomes more important?
A technically successful deployment can still produce surprisingly little value if the surrounding process remains unchanged.
These are not three separate problems. They are different sides of the same workflow.
The Traditional Handoff Model Starts to Break Down
Many transformation programs are still structured sequentially.
The business defines the requirements. A technology team builds against them. Change management comes in as implementation gets closer. Eventually, the solution is handed over to users.
That structure made sense when technology was largely automating a process that had already been designed.
Agentic AI changes the equation because the technology can alter the process itself.
If an agent can perform work that previously took several hours, the question is not simply how to automate those hours. You need to ask what the person should now be doing instead. If AI can gather information continuously, perhaps a scheduled research process no longer makes sense. If an agent can move between several systems, some handoffs may disappear entirely. If it can make a recommendation, you need to decide where human judgment belongs and what evidence that person needs to make the decision well.
The business design, technical design and operating model begin to converge.
That is difficult to manage through a series of handoffs.
Our own approach to enterprise AI has increasingly reflected this. We start by finding the value in the workflow before building, redesign the process rather than simply adding technology to it, keep people in the lead where judgment matters, and carry measurement, governance and adoption through into operation.
Put the Team Around the Workflow
This thinking led us to the Evalueserve Agentic Squad, a team model focused on answering the people questions behind AI transformation: what talent is needed, how should the workforce adapt, how should business and technology teams collaborate, and what must change in the operating model for AI to create value.
An Agentic Squad brings together three perspectives around the same business workflow from the beginning: domain advisors, forward-deployed engineers, and change and adoption consultants.
They are not three workstreams waiting to hand a project from one to the next. They are different forms of engineering toward the same outcome.
The domain advisor acts as the strategist. They carry the business goalpost, define what success should look like, and orchestrate the team toward that future state, while staying close enough to engineers to understand what is technically possible. Their work is a form of reengineering: helping the business move from the current way of working to the future one.
The forward-deployed engineer provides the muscle behind the transformation. They build the technology, connect the systems, expose what is feasible, and adapt the agentic process as the team learns what works in practice.
The change and adoption consultant operates like the chief of staff for the transformation. They keep the work moving, translate decisions into execution, measure whether the new workflow is producing results, and engineer the human adoption, behaviors, and operating model required to make the change stick.
The value comes from the interaction between them. The domain advisor can tell the engineer that a rule that looks straightforward on paper breaks down in certain real-world situations. The engineer can show the domain advisor what the technology can do and identify possibilities that would never appear in a static requirements document. The change and adoption consultant can spot when a technically elegant process is likely to create a new bottleneck, remove an important control, or ask users to work in a way they are unlikely to adopt.
That changes the questions the team asks.
Instead of asking, Where can we put an agent?, the Squad can ask:
- Which work should disappear altogether?
- Which information could be assembled before a person ever enters the process?
- Where does expert judgment create the most value?
- Which handoffs exist because of technological limitations that no longer apply?
- What needs to connect for the workflow to operate end to end?
- What will the people doing the work actually do differently?
- And, importantly, how will we know whether any of this created value?
This is also why we see forward-deployed engineering as particularly important. The engineer is not simply handed a specification and sent away to build. They stay close to the workflow and the people using it, adapting the technology as the team learns what works in practice. But the same principle applies to the whole Squad: each role stays close to the outcome, engineering the future state from a different angle.
A Better Workflow May Not Mean More AI
There is another reason we think this model matters.
When an AI team begins with the mandate to build agents, the natural measure of progress can become how much has been automated.
We do not think that should be the objective.
There will be workflows where agents can carry out a large portion of the process. In others, AI may be most valuable in gathering evidence, monitoring change, preparing an analysis, or helping a person make a more informed decision.
Some decisions should remain human because the judgment itself is valuable.
The job of the Squad is not to maximize autonomy. It is to work out what combination of people and technology produces a better way of working.
That requires domain knowledge inside the process, engineering close enough to respond to what the team discovers, and change expertise involved early enough to influence the design.
As Agents Become Easier to Build, the Surrounding Expertise Matters More
The technology will continue to improve. Building agents will become faster, models will become more capable, and the underlying platforms will do more of the engineering work for us.
That does not make the transformation challenge disappear.
In many ways, it makes the choices around the technology more important.
The organizations that create the most value from agentic AI may not be the ones that build the greatest number of agents. They will be the ones that know where agents belong, how to connect them to real work, when human expertise should remain in the loop, and how to change the workflow around what the technology can now do.
That is a much broader problem than building an agent.
And it is why we think it needs a different kind of team.
Have a Workflow You Think AI Could Change?
Tell us what you want to work differently. We'll bring the right mix of domain, engineering, and adoption expertise.
