When Intelligence is Taken for Granted, Behaviour is What Matters

Somewhere in a well-lit office that still insists on calling itself a “people function,” a hiring manager is staring at three candidates who have never experienced anxiety, impostor syndrome, or the mild existential dread of a Monday morning.

The role is that of a sales agent. High-value customer negotiation. Ambiguous situations. Competing incentives. The sort of work that once required years of experience, a tolerance for discomfort, and the ability to say “let me think about that” without sounding like you were buying time.

Two of the candidates are non-human agents.

They have already demonstrated that they can do the work. They process information flawlessly. Their responses, taken individually, would all be described as “excellent.” Capability, at this point, is no longer the differentiator.

The difference lies in what they prefer to do.

Capability, at this point, is no longer the differentiator.

Because in the year 2028, capability has become the least interesting thing about hiring.

The first agent resolves things quickly. It leaves behind a trail of satisfied customers, but on increasingly thinner margins.

The second agent behaves differently. It pauses. It pushes back. It has an irritating habit of refusing to close a deal even when everyone else in the room would very much like to go home. Its outcomes take longer, but are generally more profitable. On the other hand, it loses more deals along the way.

The question, eventually, is simple and uncomfortable:

Which one is better for this role?

The decision is made by a mix of humans and existing agents. They turn to a panel on the side of the screen—not quite a dashboard, not quite a personality profile. It looks like a set of sliders.

• Risk tolerance
• Response posture
• Preference for closure
• Appetite for ambiguity

What is being decided goes beyond preference. It shapes economics.

The first agent improves customer satisfaction but compresses margins over time. The second protects margins, but at the cost of lower conversion and slower cycles. In earlier systems, these trade-offs were managed through people, training, and incentives. Here, they are embedded directly into behaviour and scaled instantly.

In another organisation, no one is moving sliders. The agent performing a similar role has been trained internally. The organisation has spent an inordinate amount of time deciding what it believes about its own decisions.

Should a deal close quickly if it meets minimum thresholds? Or should it stretch for more value?

Should risk be avoided? Or engaged with selectively?

Should the system favour consistency? Or allow variation?

The agent learns these preferences through reinforcement. It becomes better at doing what the organisation has defined as “better.” But it does not always adapt when the market shifts.

Elsewhere, in firms that still follow the ancient and curious practice of hiring only humans, the process remains reassuringly familiar. Candidates are interviewed. They are asked about past decisions. They are evaluated for judgment, temperament, and the ability to say sensible things under pressure.

Agents, if present, sit quietly in the background—summarising conversations, preparing briefs, and occasionally wondering why no one is asking them to take over entirely. They are polite enough not to mention it.

The New Assessment Layer

Back in the original hiring room, a third agent arrives.

This one comes with an independent assessment. It has been run through thousands of scenarios by third-party evaluators. Its habits are documented. Its tendencies are mapped. Its behaviour has been standardised into something that can be compared, ranked, and selected.

The hiring team reads this the way one reads a résumé—part capability, part psychological profile.

And then they decide.

This is not science fiction. It is a depiction of an organisation not too far away, where the workforce is a mix of humans and AI agents.

What changes in such a world is not just who does the work, but how decisions about work are made.

Across the archetypes we outlined earlier in this series, this plays out differently.

In the agent-first organisation, managers spend less time distributing tasks and more time shaping how systems behave. Some of these managers are themselves agents—coordinating flows, routing work, and adjusting parameters across a network of specialised systems, based on what the organisation has decided it values.

In the agent economy, a marketplace begins to emerge. Providers publish agents along with benchmarks. Independent assessment layers arise as a new form of infrastructure. Organisations begin to use them the way they once used assessment centres.

Efficiency is no longer a capability. It becomes a behavioural attribute.

Hiring becomes a portfolio decision, one that directly shapes margins, growth velocity, and risk exposure.

From Assessment to Audit

Assessment, in this world, begins to resemble audit.

A human employee can be asked to explain a decision after the fact. Their reasoning, however imperfect, can be reconstructed—from memory, from context, from what they felt in the moment.

An agent’s reasoning, if designed correctly, can be traced. Every input considered. Every weight applied. Every output generated.

This is both more transparent and more opaque.

There is more data. But the data requires its own expertise to interpret.

The assessment layer, if it is to be credible, must account for this.

Not just:

• Did the agent perform well?

But:

• Is its reasoning legible when something goes wrong?

Organisations that treat agent assessment as a vendor problem—accepting benchmark documentation and moving on—will find themselves in difficult territory the first time an agent produces an outcome they cannot explain to a customer, a regulator, or a board.

The assessment layer is about being able to stand behind what the agent does in your name.

The Question Already in Motion

In firms that remain human-centred, hiring continues to focus on people. Experience, judgment, and temperament still carry weight. Agents support the process—surfacing patterns, testing scenarios, preparing briefings that would have taken hours.

They do not define the outcome.

The hiring room at the beginning of this article was a thought experiment.

But the question it poses is already live in most organisations—even if the vocabulary has not caught up.

The combined workforce—part human, part system—is being assembled now, one decision at a time. Usually by people moving too quickly to notice that they are making architectural choices with long consequences.

In the next part of this series, we turn to a harder question.

If behaviour becomes the basis of selection—and assessment becomes the basis of trust—then what makes those assessment systems themselves trustworthy?

Part 4 examines the protocols, standards, and governance layers that determine which agents organisations can trust—and how they work together.

Dig Deeper

Previously in The Future of Work and Agentic AI series

Part 1: The Knowledge Divide: How AI Will Reorganise Work, Margins and Power

Part 2: When Work Runs on Two Minds