When Work Runs on Two Minds

Imagine there are two ways to get to a fitness centre: an escalator and a staircase. Which one would you choose?

Some would say, “I’m anyway going to the gym—let me start the step count now. Stairs.”

Others would say, “I’m going to the gym—let me relax for a few seconds more. Escalator.”

This simple example reflects a familiar human reality: the gap between intent and action.

Now place that in today’s work environment, where humans increasingly operate alongside AI agents. As agentic AI systems begin to take on more autonomous roles, what happens to motivation? To effort?

In our first essay in this series, we outlined three emerging organisational archetypes. Here, we take that forward by asking a more fundamental question.

As agentic AI systems begin to carry knowledge across workflows rather than individuals holding it, coordination shifts from people to the infrastructure that carries that knowledge. This changes how work happens, how people learn, and how behaviour is shaped.

As knowledge begins to move through systems rather than remain within individuals, coordination shifts from people to the infrastructure that carries that knowledge.

Learning and Motivation as Reinforcement Systems

For decades, organisational behaviour rested on a relatively stable loop: people applied knowledge through effort, effort produced outcomes, outcomes generated feedback, and feedback reinforced learning and motivation.

This loop explains how expertise builds, how incentives work, and how performance improves over time. Much of organisational design has focused on strengthening this loop—making the metaphorical staircase feel more rewarding to climb.

Artificial intelligence introduces a second reinforcement system.

Agents do not learn or “feel” motivation in the human sense. Their behaviour is shaped through data, reward structures, and feedback loops. What they do reflects what has been optimised during training and refined during deployment.

Organisations are now operating with both systems simultaneously.

Humans and agents participate in different parts of the same workflow, but they learn differently, respond to different signals, and evolve at different speeds. Each organisational archetype must therefore reassess what learning and motivation actually mean.

High Maturity Organisations: When Learning Sits in the System

In high maturity organisations, knowledge is embedded into systems that execute workflows across functions. Agents draw on organisational memory, past decisions, and defined objectives to produce consistent outputs.

Learning, in this environment, shifts form.

Agents improve through training and feedback loops that refine system-level performance. Humans engage by interpreting outputs, refining objectives, and shaping how the system evolves.

Skill formation changes as a result. Repetition moves from human execution to system optimisation. Human expertise develops around understanding system behaviour, identifying gaps, and improving how learning is encoded.

Ironically, the agents would be extremely insulated from sudden but impactful real-world signals until those signals are fed as significant enough data into their training. For example, a sudden outbreak of the Iran war may demand a change in the agent behaviour for certain organisations, but their AI agents would not know about the war until they are re-trained with the relevant data points and/or instructions. In contrast, training human resources may become easier in such cases as they would already have awareness of such signals and would be mentally prepared for a change in their approach as a result.

Motivation also shifts.

Engagement now depends on how clearly individuals can see their influence on outcomes mediated through systems. When that visibility is strong, motivation aligns with system performance. When it is diffuse, engagement weakens—even if output remains high.

Agent behaviour, by contrast, is precise but bounded. It reflects the reward structures and constraints embedded in training. If those signals are well defined and updated, behaviour remains aligned. If they are narrow or static, performance efficiency stays consistent but the resulting impact begins to degrade in less visible ways.

Mid Maturity Organisations: When Learning Splits Across Systems

Most organisations will operate in a mixed environment.

External agents bring structured intelligence shaped at scale. Internal teams adapt outputs, handle exceptions, and connect systems to real operating conditions.

Learning splits across these layers.

Agents improve within the boundaries of their training. Humans learn through exceptions, corrections, and context-specific adjustments. As a result, people often interact with only parts of a workflow, especially where interpretation or intervention is required.

Learning becomes more situational than cumulative.

Motivation reflects this fragmentation. Where work involves continuous adjustment without full ownership, engagement depends on whether outcomes can still be meaningfully connected back to effort and problem-solving.

At the same time, agents operate on incentive structures designed externally and adapted internally. Their behaviour remains stable only within those definitions and can stagnate if not recalibrated. How much and how easily an externally trained agent can be calibrated based on the organization’s need would emerge as the most important success criteria for agent adoption.

Over time, the organisation reflects the interaction of two reinforcement systems rather than a single, unified one. ‘Credit sharing’ between humans and machines would become an important debate point in this archetype and would need a clear attribution metric to keep humans motivated while non-complacent.

Human-Centred Organisations: When Learning Remains Direct

In human-centred organisations, the traditional reinforcement loop remains intact.

People perform the work, see outcomes, and adjust behaviour directly. Learning builds through repetition and experience. Motivation remains closely tied to ownership and visible impact.

This produces strong engagement and adaptability. It also means that knowledge moves through people rather than systems, and scaling depends on how effectively that knowledge is shared.

In such environments, agents play a more limited role, largely dependent on human prompts and guidance. They do not operate as independent learning systems within the organisation.

Which raises a critical question: do agents need motivation at all?

When Agents Need Motivation

Agents require motivation in the form of incentive design.

Their behaviour is shaped by reward functions, objective structures, and feedback loops. In human systems, motivation has long been studied and designed, balancing performance, collaboration, and long-term growth.

With agents, this layer moves upstream into system design.

Agent providers make foundational choices about what is optimised, how trade-offs are handled, and how success is measured. These decisions shape behaviour before the agent is ever deployed.

This raises new questions. Will providers make their incentive structures explicit? How will organisations evaluate them? And how will differing incentive logics influence adoption and trust?

The answers will shape how multi-agent environments evolve, particularly in mid-maturity systems.

Changing Agents, Changing Behaviour

When organisations change agent providers, they are, in effect, introducing a different learning and incentive system.

When organisations change agent providers, they are, in effect, introducing a different learning and incentive system.

A new agent may prioritise speed differently or weigh accuracy, cost, or risk in another way. These differences can appear subtle in outputs but meaningful in behaviour.

Even today, agents exhibit recognisable tendencies—helpful but overly compliant, capable of generating complex outputs but limited in incorporating iterative feedback, and so on.

Over time, such shifts influence how decisions are made across the organisation.

In human systems, behavioural change emerges gradually. With agents, it can appear much faster, because it is embedded directly in the system.

Multiple Agents, Multiple Logic

Many organisations will operate with multiple agents simultaneously: a customer service agent from one provider, a sales system from another, and an internally developed operations agent.

Each carries its own training logic and incentive structure.

Individually, they may perform well. Collectively, they may reflect different priorities—speed, completeness, or cost efficiency.

Humans work across these systems, aligning outcomes and resolving differences that originate from how each agent is trained.

The Direction of Travel

Across all three maturity levels, a common pattern emerges.

Learning is distributed across humans and systems. Motivation is shaped through both behavioural and engineered reinforcement. Outcomes reflect how well these systems are aligned.

As organisations operate with multiple agents, often from multiple providers, alignment becomes a structural requirement.

Not just alignment of processes, but of how systems interpret objectives, how incentives shape behaviour, and how learning flows across boundaries.

This creates a new coordination layer, shaped not just by process, but by the interaction of different learning and incentive systems. The architectural choices around this coordination layer would become much more important for a successful execution than the efficiency of the agents and humans themselves.

Organisations once aligned people. They must now align systems that learn in fundamentally different ways.

That is where the next layer begins: how to assess, standardise, and align these systems at scale. We take that up in the next essay in this series.