AI transformation requires judgment systems, not tool deployments.

Close

Productivity Is Not the End of the Analysis

Many executive teams evaluate AI by asking:

Did productivity increase?

It is a reasonable question, but an incomplete one.

Research summarized by Microsoft found that early large-language-model tools increased speed on selected enterprise information-work tasks without a significant decrease in quality. The report also cautioned that the studies covered tasks where these tools were most likely to provide value and called for research across a wider range of roles and work.[1]

Even when production becomes faster, the constraint does not necessarily disappear. It may move.

Drafting can accelerate while review accumulates. Research can become easier while verification becomes harder. Coordination can improve while prioritization becomes the new point of delay.

The Constraint Is Context-Dependent

The bottleneck will not always move to judgment.

In one organization it may shift to data access. In another it may become integration, compliance review, customer acceptance, or managerial attention. Faster output can also fail to improve the overall result when the accelerated task was never the limiting step.

The bounded claim is this: local productivity gains can relocate constraints elsewhere in the workflow. Leaders therefore need to measure the performance of the whole system, not only the speed of an assisted task.

Microsoft's 2026 Work Trend Index offers a broader organizational signal. It reports that AI is placing a premium on judgment, clarity of intent, and work design, while organizational conditions account for more reported impact than individual effort alone.[2] That evidence does not establish where every bottleneck will move. It supports examining the operating model around the tool.

Signal Without Absorption Creates Noise

AI can increase the amount of material ready for consideration: drafts, summaries, analyses, recommendations, and proposed actions.

That abundance can be useful. It can also increase the burden on people responsible for deciding:

• What deserves trust
• What requires verification
• What should be challenged
• What should move forward
• Who owns the final decision

The organization may become faster at producing options without becoming better at choosing among them.

The mechanism is simple. When generation accelerates but review capacity, decision rights, and prioritization rules remain unchanged, work accumulates at the next constrained stage.

Why Management May Become More Important

Managers may become a critical review and coordination layer between AI-generated work and organizational action. They often decide what gets approved, escalated, revised, or treated as official.

But this should not become a universal claim about middle management. Some workflows will distribute review to specialists, automate portions of validation, or move authority closer to frontline teams. The relevant question is not which layer should gain power. It is where informed judgment must occur and whether that role has the capacity and authority to perform it.

A Bottleneck Diagnostic

After introducing AI into a workflow, ask:

1. Where is work now waiting?
2. Which stage has accumulated additional review or exception handling?
3. Which decisions lack a clear owner?
4. Where are people repeatedly correcting, validating, or contextualizing AI output?
5. Which local metric improved while the end-to-end outcome remained unchanged?

Run the diagnostic before and after deployment. The purpose is not to prove that AI failed or succeeded. It is to identify which constraint changed and whether the organization is equipped to manage it.

Signal & Context Perspective

AI changes where work happens.

Sometimes the immediate gain is faster production. The next challenge may be verification, prioritization, governance, or absorption. That sequence is a hypothesis to test, not a fixed maturity model.

The winners will not necessarily be the organizations producing the most with AI. A more defensible advantage may come from recognizing where the constraint moved and redesigning the workflow around that reality.

Open question: If production became dramatically faster tomorrow, behind which person, decision, system, or approval would the additional work accumulate?

Sources

[1] Microsoft Research, "AI and Productivity Report: First Edition," 2023. microsoft.com

[2] Microsoft, "2026 Work Trend Index Annual Report: Agents, human agency, and the opportunity for every organization," May 5, 2026. microsoft.com

Close

The Problem May Not Be the Tool

Organizations often assume that low AI adoption reflects a technical problem. The model is not accurate enough. The output quality is uneven. The integration is incomplete.

Sometimes that diagnosis is right. But technically capable tools can also remain underused when employees are uncertain about acceptable use, accountability, or the consequences of getting something wrong.

Microsoft's 2026 Work Trend Index provides useful evidence for this distinction. Based on a survey of 20,000 workers using AI across 10 countries and analysis of Microsoft 365 activity, the report found that organizational factors such as culture, manager support, and talent practices accounted for twice the reported AI impact of individual effort alone. That does not prove that trust is the cause of every adoption gap. It does suggest that leaders should examine the system around the technology, not only the technology itself.[1]

The Hidden Question Employees Are Asking

Leadership teams frequently ask:

Why are employees not using the tools?

Employees may be asking a different question:

Is it safe for me to use them?

Not only cybersecurity safe. Career safe.

They may be uncertain about whether using AI will be judged, whether an error will be treated as experimentation or negligence, whether the policy is clear enough to follow, or whether greater productivity will simply become a higher baseline expectation.

These concerns should not be treated as universal. The same tool can be received differently across teams because local incentives, manager behavior, review practices, and workflow conditions differ.

Trust Is Operational

Trust is not built through an announcement. It develops through repeated experience.

Employees notice:

• How leaders describe the purpose of AI
• Whether approved and prohibited uses are clear
• How outputs are reviewed
• Who receives credit when the work succeeds
• Who carries responsibility when it fails
• Whether expectations change without discussion

This is the organizational mechanism: people interpret the technology through the consequences attached to using it. If responsible experimentation is encouraged in principle but punished in practice, caution is rational.

What Shadow AI May Be Signaling

Unauthorized AI use can create genuine security, privacy, and compliance risks. It should not be minimized.

But treating every instance only as defiance can conceal another possibility. Employees may be trying to solve legitimate workflow problems that approved tools or processes do not address. In that case, shadow AI is both a control problem and a diagnostic signal.

The appropriate judgment is not to excuse unauthorized use. It is to investigate the demand beneath it while enforcing necessary boundaries.

A Four-Part Trust Diagnostic

Before describing low usage as resistance, leaders should examine four questions:

1. Purpose: Do employees understand what the tools are intended to help them accomplish?
2. Boundaries: Do they know where AI use is permitted, prohibited, or requires approval?
3. Review: Can they see how outputs will be checked and who owns the final decision?
4. Consequences: Do incentives reward responsible experimentation, or punish visible mistakes while quietly rewarding speed?

A weakness in any one area does not prove that trust is the primary constraint. It identifies where further investigation should begin.

Signal & Context Perspective

Technology enters an existing human system. It does not replace one.

Low adoption may reflect poor tools, weak integration, inadequate training, unclear incentives, or distrust. Leaders should distinguish among those causes before prescribing more communication, more licenses, or more pressure to use the technology.

The better judgment is diagnostic: do not assume reluctance is irrational until the organization has made responsible use understandable, workable, and safe.

Open question: What evidence would distinguish employee resistance from rational caution in your organization?

Sources

[1] Microsoft, "2026 Work Trend Index Annual Report: Agents, human agency, and the opportunity for every organization," May 5, 2026. microsoft.com

Close

Capability Is Only the Beginning

Most discussions about agentic AI focus on capability.

Can agents perform complex tasks? Can they coordinate activities? Can they execute workflows?

Those questions matter. But once a system can act, a harder question follows:

Under what conditions should it be allowed to act?

This is not an argument that agentic AI is primarily or exclusively a governance problem. Reliability, security, identity, interoperability, and system design remain technical concerns. The point is narrower: capability alone cannot determine an appropriate level of autonomy.

NIST's AI Agent Standards Initiative reflects this expanded problem. Its work includes standards, secure protocols, agent authentication and identity infrastructure, and security evaluations intended to support agents acting securely on behalf of users.[1]

Autonomy Changes the Control Problem

Traditional generative AI systems usually produced an output for a person to consider. Agentic systems can go further. Depending on their design and permissions, they may initiate actions, call tools, route information, interact with other systems, or complete multistep tasks.

As the distance between recommendation and action narrows, governance has to move closer to the workflow.

Someone still needs to define the permitted boundary. Someone needs to own exceptions. Someone needs to determine what evidence is retained. Someone remains accountable for the operating decision, even when the system performs the action.

The Illusion of End-to-End Efficiency

If a process contains five steps, automating all five can appear more efficient than automating three.

But the steps may not be equivalent. Some are routine execution. Others contain tacit judgment, legal exposure, reputational consequence, or an exception that only becomes visible in context.

The risk is not automation itself. It is granting autonomy based on process sequence rather than consequence.

Human review is also not automatically effective. A person placed at the end of a fast, opaque process can become a ceremonial approver. Oversight has value only when the reviewer has enough information, authority, time, and a clear standard for intervention.

A Practical Autonomy Test

Before an agent is allowed to act without prior review, assess the action across four dimensions:

1. Reversibility: Can the action be undone fully and promptly?
2. Consequence: Could it create material legal, financial, operational, security, or reputational exposure?
3. Observability: Can the organization determine what the agent did, what authority it used, and what information shaped the action?
4. Escalation: Is there a named owner and a usable path for uncertainty, conflict, or exception handling?

The decision rule is straightforward: the less reversible or observable an action is, and the greater its consequence, the stronger the case for mandatory human authorization.

This is a proposed management test, not a universal standard. It should be adapted to the workflow, risk environment, and applicable obligations.

Governance Must Become Operational

A policy can define intent. It cannot, by itself, govern an action at the moment it occurs.

Operational governance appears in permissions, approval thresholds, identity controls, logs, exception routes, testing, and named decision owners. NIST's initiative similarly emphasizes agent identity, authorization, secure interaction, and evaluation, reinforcing that confidence depends on technical and organizational controls working together.[1]

The judgment for leaders is not "human or autonomous." It is where autonomy is justified, where review is mandatory, and what evidence would support changing that boundary over time.

Signal & Context Perspective

Agentic AI expands what systems can execute. Context determines which actions are appropriate. Judgment sets the boundary. Governance makes that boundary real inside the workflow.

Organizations that make those decisions explicitly may be able to expand autonomy with greater confidence. Those that leave them implicit may move quickly without knowing which risks they have accepted.

Open question: Which steps in your workflows are genuinely routine, and which only appear routine because the judgment inside them has never been made explicit?

Sources

[1] National Institute of Standards and Technology, "AI Agent Standards Initiative," updated August 14, 2026. nist.gov