According to Korn Ferry, more than half of talent leaders are already planning to add autonomous AI agents to their teams in 2026. Not as tools their people use. As participants their people work alongside. The language matters. When agents were tools, the organizational questions were technical: deployment, integration, monitoring. When agents become participants, the questions become organizational: who specified what this agent is allowed to do? Who reviews whether it is doing it? Who is accountable when it drifts? And who owns the interaction points where humans and agents hand work to each other, every day, in workflows that nobody designed for two kinds of worker?
That last question is the one almost nobody is asking. And the organizations that answer it first will build a competitive advantage that compounds for a decade.
A pattern with precedent
This is not the first time a new category of worker entered the enterprise and forced an organizational reckoning. Every major workforce transition in the past thirty years followed the same structural pattern: a new type of worker appeared, the existing organizational toolkit could not accommodate it, and a new function, a new discipline, and eventually a new software category emerged to manage the new workforce shape.
When outsourced labor expanded through the 1990s and 2000s, it produced vendor management offices, SLA frameworks, and an entirely new procurement discipline. The People function did not own it initially. Procurement and operations did. HR absorbed contingent workforce management a decade later, after the organizational shape had already been defined by other functions. The function that owns the new workforce shape is not always the function you expect.
When remote work accelerated in 2020, it forced new performance evaluation frameworks, new onboarding processes, and new ways of measuring presence and engagement. HR’s existing models assumed co-location. The adjustment took two to three years, and most organizations are still recalibrating. Workforce transitions that change the modality of work, not just the location of the worker, require new measurement infrastructure, not just new policies.
When gig and contingent workers scaled through the 2010s, they produced new compliance structures, new classification frameworks, and a legal and organizational apparatus that HR was not initially equipped to handle. Platform companies built the management tooling. Traditional enterprises adopted it slowly. When the new worker does not fit the existing employment model, the existing HR toolkit does not extend. A new toolkit has to be built.
The agentic workforce transition follows the same pattern with a structural difference that makes it harder. The new worker is not a human in a different arrangement. It is a non-human entity that requires specification, measurement, and verification rather than management, motivation, and development. The People function that owns the blended workforce needs a different toolkit entirely.
Gartner identified this in its 2026 CHRO priorities research: “shape work in the human-machine era” is now a top-four priority across 426 CHROs surveyed, and it calls for a “now-next” talent strategy for a blended workforce. The mandate is forming. The toolkit has not arrived.
The undesigned interaction points
Every enterprise deploying AI agents is already building a blended workforce, whether it intended to or not. Humans and agents share decision authority, hand tasks to each other, and operate in overlapping domains. The interaction points between them — the moments where a human delegates to an agent, where an agent escalates to a human, where both contribute to a single workflow — are being designed by accident. Nobody owns them. Nobody is measuring them.
Consider a customer service workflow. An agent handles the initial query, drawing on product knowledge and account history. The issue exceeds the agent’s authority, and it escalates to a human representative. The human resolves the issue with reference to the agent’s prior work.
That single interaction exposes two dimensions most organizations are not managing. The first is a delegation boundary. The escalation threshold was either designed with intention — someone specified when the agent should hand off and what information transfers with the handoff — or it emerged by default, and the human is reconstructing context on the fly. In Decision Surfaces, I described these as the boundary architecture between human judgment and agent execution: the interfaces where delegation happens and where governance becomes observable. Every blended workforce creates these boundaries. The question is whether they were designed or discovered after something went wrong.
The second dimension is brand coherence. The customer experienced both the agent’s voice and the human’s voice in a single interaction. Were they coherent? Did the transition feel intentional, or did it feel like the customer was passed from one system to another with no connective tissue? In Janus Brands, I described the dual-audience economy where both humans and agents are first-class participants in every commercial interaction. That dual-participant reality means both the agent and the human represent the organization. Both need to represent it consistently. The interaction point is where that consistency is either demonstrated or broken.
These two dimensions converge at every interaction point in every blended workforce. The organizations that design their interaction points deliberately, specifying both the delegation boundary and the brand coherence requirement, will deliver experiences that feel intentional and trustworthy. The organizations that leave them undesigned will absorb the variance.
The missing specification
Agents have, or should have, formal specifications that describe their behavioral identity, their decision authority, and their boundaries. The humans who work alongside those agents have none of the equivalent.
No document specifies what a human worker’s role becomes when half the tasks they used to own are now performed by an agent. No framework describes how a human supervisor should evaluate an agent’s work, when to override it, or how to evolve the agent’s scope as trust develops. The agent’s capabilities are expanding quarterly. The human’s role description was last updated annually, if at all. The gap between the two is where confusion, friction, and organizational risk accumulate.
What the blended workforce needs is a Human Guidebook: the human-side mirror of the agent specification. Where the agent specification describes the agent’s behavioral identity, decision authority, and boundaries from the agent’s perspective, the Human Guidebook describes the human’s role, authority, and responsibilities from the human’s perspective, in the context of the agents they work alongside.
A standard job description says what a human does. A Human Guidebook says what a human does in relation to the agents they work with, and how that relationship changes as the agent matures. The distinction matters because the human’s role is not static. It evolves through stages, and each stage demands different skills, different authority, and different measurement.
When an agent operates on internal tasks only, the human’s role is direct supervision. The Guidebook specifies: review every output before it leaves the team, approve every action, correct every drift, document every correction so the agent’s specification can be updated. The skills are editorial. The authority is absolute. The measurement is coverage: did the human review everything the agent produced? A standard job description does not capture this because the role did not exist before the agent arrived.
When an agent interacts externally but cannot transact, the human’s role shifts to quality assurance. The Guidebook specifies: the agent drafts customer communications, the human approves before they send, the feedback loop tightens as trust develops. The skills shift from editorial correction to pattern recognition: is the agent’s judgment improving? Are the exceptions getting rarer or changing shape? The measurement shifts from coverage to exception rate and exception type. A quality assurance role for agent output is not the same as a quality assurance role for human output, and the Guidebook makes the distinction explicit.
When an agent transacts with human confirmation, the human’s role becomes authorization. The Guidebook specifies: evaluate commercial decisions against organizational policy, approve transactions above threshold, flag anomalies in the agent’s decision patterns. The skills are now judgment-intensive: is this transaction appropriate given the context the agent may not fully understand? The measurement is decision quality, not task completion. The human is no longer reviewing every output. The human is evaluating the agent’s judgment at the boundary where organizational risk concentrates.
When an agent transacts autonomously within constraints, the human’s role shifts again: from transaction-level approval to constraint-envelope design and exception handling. The Guidebook specifies: define and maintain the constraint boundaries within which the agent operates, design the escalation triggers, handle the exceptions that fall outside the envelope. The human is no longer in the loop on every decision. The human is designing the loop. The skills are architectural. The measurement is envelope integrity: are the constraints holding? Is the agent operating within them? When exceptions occur, are they the right exceptions?
Each stage requires the human to do different work, exercise different authority, and be measured by different criteria. Without the Guidebook, each transition happens informally. The human figures out their new role through trial and error. The organization discovers what went wrong after the fact. With it, the transition is specified, measurable, and governable. Both sides of every interaction point are documented, and both evolve through a coordinated lifecycle.
This is the bidirectional compilation I described in The Compiled Corporation: the agentic workforce transition requires organizations to translate their operational identity into explicit specifications that both humans and agents can operate against. The compilation runs in both directions. The agent needs a specification it can execute against. The human working alongside that agent needs a specification they can supervise against. Same organizational values, two compiled outputs. The interaction points are where the two specifications meet.
The emerging brand surface
Here is what most workforce-planning conversations miss: the interaction points between humans and agents are not just an operational concern. They are the new brand surface.
In the economy that is forming, every commercial interaction either involves an agent or passes through a system an agent shaped. The quality of the interaction — whether the customer feels understood, whether the decision was appropriate, whether the handoff from agent to human was coherent — depends entirely on how the interaction point was designed. Interaction design is becoming a discipline, not a feature. The organization that specifies both sides of its human-agent workflows, measures performance at the interaction points, and evolves both specifications as capabilities mature is building a brand experience that compounds. The organization that leaves the interaction layer undesigned is hoping its agents and its humans will figure it out together.
The question on the horizon
Every prior workforce transition eventually produced a dominant software category to manage the new workforce shape. ERP and procurement platforms for the outsourced supply chain. Collaboration platforms for remote work. Workforce management platforms for gig workers. Each category emerged because the organizational need was real and the existing toolkit could not serve it.
The blended human-agent workforce will produce its own category. The question is not whether, but what it looks like and who builds it. PwC’s recent research framed it directly: if HR does not lead the design of human-agent work architecture, the decisions will still be made, just without the workforce perspective that a CHRO brings.
A CHRO reading this should recognize the pattern from their own career. The organizational need preceded the tooling in every prior transition. Vendor management existed as a discipline before vendor management software existed as a category. Remote work performance evaluation existed as a problem before collaboration platforms solved part of it. The blended workforce interaction problem exists today, and the humans living inside it are improvising.
A COO reading this should recognize the operational gap. The interaction points between humans and agents are unmonitored, unowned, and accumulating risk. The governance infrastructure described in The Governance Gap Nobody Is Closing addresses the agent side of that equation: specifying, measuring, verifying, and evolving agent behavior. The accountability infrastructure that makes that governance credible produces evidence, not documentation. The interaction layer is where both meet the human side. All three are needed. None is sufficient alone.
The organizations that thrive in the blended workforce will be the ones that designed the human side of the equation with the same rigor they brought to the agent side. The interaction points are where both sides meet. Right now, almost nobody is designing them.
Daniel Davenport is co-founder and Chief Identity Officer of Applied Identities, the governance infrastructure company for the agentic workforce. A business strategist and early-stage technology adopter across four waves — internet, mobile, cloud, AI — he writes about what changes when autonomous agents stop being tools and start being workers, and what enterprises need in place before that transition is safe to make at scale.