Which tasks get Delegated to agents, which roles evolve, and what does human work look like when AI handles the routine?

The delegation question

Every technology that automates work forces the same question: what’s left for humans? The agentic web makes this question concrete and immediate. When AI agents can browse, research, compare, schedule, book, purchase, write, and coordinate—what do knowledge workers actually do?

The answer isn’t “nothing,” and it isn’t “everything stays the same.” It’s more nuanced: a Redistribution of tasks, an evolution of roles, and a shift in what makes human contribution valuable.

Task decomposition

Understanding the impact requires decomposing jobs into tasks. Most knowledge work isn’t one thing—it’s a bundle of activities. A marketing manager might spend time researching competitors, writing copy, analyzing data, coordinating with agencies, presenting to leadership, and responding to emails. Some of these tasks are prime candidates for agent delegation; others aren’t.

Tasks that agents handle well: Information gathering and synthesis, routine communication, scheduling and coordination, data processing and analysis, first-draft content creation, repetitive workflows with clear rules.

Tasks that remain human: strategic judgment and prioritization, stakeholder relationship management, creative direction and taste, navigating ambiguity and organizational politics, accountability and decision ownership.

The same job looks very different depending on how the task mix shifts. A marketing manager whose time freed from research and routine writing can focus on strategy, creativity, and relationships might find their role elevated. One whose value came primarily from executing those routine tasks may find less demand for their contribution.

Historical parallels (and their limits)

We’ve been here before—sort of. Every wave of automation, from mechanical looms to spreadsheets to search engines, shifted the task landscape. Work changed, but work persisted.

The spreadsheet didn’t eliminate accountants; it eliminated a certain kind of accounting work and created demand for different skills. Search engines didn’t eliminate researchers; they changed what research meant and what made a researcher valuable.

The pattern offers comfort: automation typically shifts work rather than ending it. But the pattern has limits. Previous automation technologies were tools that enhanced human capability. AI agents are different—they’re autonomous actors that can pursue objectives without continuous human direction.

This doesn’t guarantee mass displacement, but it does mean historical parallels may not fully apply. The right mental model might be less “spreadsheets changed accounting” and more “what happened to horses when cars arrived.” Horses still exist, but their economic role transformed utterly.

The skills that matter

As agents handle more routine knowledge work, the skills that differentiate human contributors shift.

Judgment under uncertainty. Agents excel at optimization when objectives are clear and data is available. Humans remain essential for decisions where the objectives themselves are contested, where key information is missing, and where multiple stakeholders have legitimate but conflicting interests.

Relationship depth. Agents can manage transactions, but human relationships involve trust, empathy, and shared history that agents can’t replicate. Roles built around deep stakeholder relationships—client management, team leadership, partnership development—retain human centrality.

Creative direction. Agents can generate content, but deciding what’s worth creating, what approach fits the brand, what will resonate with audiences—this remains human judgment. The creative director reviewing AI-generated options is different from the copywriter producing first drafts.

System design. As work becomes more agent-mediated, someone needs to Design the systems: what tasks to delegate, how to specify objectives, what guardrails to implement, how to evaluate agent performance. This meta-work of orchestrating human-agent collaboration becomes a skill set in itself.

Accountability. When things go wrong, someone needs to be responsible. Agents don’t take accountability—humans do. Roles that involve ownership of outcomes, especially in high-stakes domains, remain human by necessity.

Organizational adaptation

How organizations adapt matters as much as how individuals do.

Some organizations will use agents to do more with the same team—increasing output per person without reducing headcount. Others will use agents to reduce headcount while maintaining output. The choice depends on growth opportunities, competitive dynamics, and leadership values.

The organizations that navigate this best will likely rethink work design at a fundamental level. Instead of fitting agents into existing roles, they’ll Redesign how work gets done—identifying which tasks benefit from agent handling, which require human judgment, and how the handoffs between them should work.

This redesign requires experimentation. We don’t yet know the optimal human-agent task allocation for most knowledge work. Organizations that prototype different approaches, measure outcomes, and iterate will learn faster than those that either resist change or implement it dogmatically.

The transition period

We’re in an awkward middle period where agents are capable enough to change work but not capable enough to replace it. This creates friction.

Workers are expected to learn to work with agents while also doing their existing jobs. Organizations are adopting agent tools without fully rethinking workflows. The benefits are partial, the disruption is real, and the end state isn’t clear.

This transition will take years, not months. The organizations and individuals who approach it with patience—Investing in learning, accepting temporary productivity dips as new workflows are established, resisting both hype and denial—will emerge better positioned.

Individual strategy

For individuals navigating this landscape:

Audit your task mix. What do you actually spend time doing? Which tasks are candidates for agent delegation? Which represent your differentiated value? Honest self-assessment is the starting point.

Develop agent fluency. Learn to work with AI tools effectively—not just as a user but as an orchestrator. Understanding how to specify objectives, evaluate outputs, and iterate with agents becomes a core competency.

Invest in judgment-intensive skills. Strategic thinking, stakeholder management, creative direction, system design—these become more valuable as routine tasks get delegated. Develop them intentionally.

Build relationship capital. Your network of human relationships—trust you’ve built, context you share, reputation you’ve established—becomes more valuable in a world where transactional tasks are automated. Invest in relationships now.

Stay adaptable. The specifics will shift as agent capabilities evolve. The skill that matters most is the ability to adapt—observing how work is changing, learning new approaches, and adjusting your contribution accordingly.

A realistic view

The agentic web won’t eliminate knowledge work. It will change what knowledge work means and what makes a knowledge worker valuable. Some roles will shrink; others will expand; new ones will emerge that we can’t yet name.

The transition will be uneven—faster in some industries and functions, slower in others. It will create winners and losers, and the distribution won’t always be fair. Managing this transition well—at the individual, organizational, and societal level—is one of the defining challenges of the next decade.

But the fundamental human capacities—judgment, creativity, relationship, accountability—remain valuable precisely because they’re hard to automate. The agentic web handles more of the mechanical. What’s left is more essentially human than what came before.