AI Should Redesign Work,Not Compress Teams Into One Person

Michael D. Penn

SHRM-SCP • SPHR • Business Systems Architect • AI Systems Builder

Founder of Total Freedom Life. Earned all five major HR certifications. Builds AI-enabled operating systems for founder-operators and executives.

About MichaelLinkedInSeptember 12, 2026 • 20 min read

I have been reading a lot of senior AI job descriptions lately.

The titles are different, but a pattern keeps showing up.

One person is expected to own AI strategy, enterprise architecture, hands-on engineering, product direction, governance, program delivery, change management, training, analytics, executive communication, vendor relationships—and sometimes people management and business development too.

My reaction is increasingly the same:

I understand why this is happening. AI really can make one person dramatically more capable. I experience that every day.

As an entrepreneur building AI-native products, I can research faster, compare options faster, prototype faster, inspect code faster, draft faster, analyze more information, and move from an idea to something operational at a speed that would have been difficult to imagine only a few years ago.

That leverage is real.

That distinction may become one of the most important job-design questions of the AI era.

But I now think there is an even larger question behind it.

If AI allows us to create more value with fewer human labor hours, what do we want to do with the capacity it creates — and who should benefit from it?

Businesses should benefit from successful AI investment. Investors should earn returns for capital and risk. Customers should benefit through better products, service, access, and economics. But workers whose jobs, knowledge, and productive leverage are being transformed should have meaningful pathways to participate in the gain too.

That does not require one fixed formula. It can mean compensation, gain-sharing, ownership opportunities, shorter or more flexible working time, reduced repetitive work, greater autonomy, reskilling, internal mobility, stronger job quality, or more time for work that still requires human judgment and relationships.

That is the larger purpose of this argument. Role design is where the transition becomes visible first. The long-term question is whether technological progress produces stronger organizations and better human lives.

AI productivity is real. But task productivity is not job productivity.

There is credible evidence that generative AI can produce meaningful productivity improvements on specific kinds of work.

A 2025 Quarterly Journal of Economics study of 5,172 customer-support agents found that access to an AI assistant increased productivity by about 15% on average, with the largest gains among less experienced and lower-skilled workers.

A randomized experiment published in Science found that professionals using ChatGPT for a set of writing tasks completed them about 40% faster while producing work rated 18% higher in quality.

A preregistered Harvard Business School / Boston Consulting Group experiment with 758 consultants, now published in Organization Science, makes the boundary especially clear. Across 18 realistic tasks inside AI's capability frontier, AI users worked more than 25% faster and completed more than 12% more tasks, with substantially higher performance. But on a deliberately outside-the-frontier task, consultants using AI were 19 percentage points less likely to produce the correct solution than those working without it.

Those findings matter. But notice the unit of analysis.

Noy & Zhang and Dell'Acqua and colleagues measure defined professional tasks. Brynjolfsson, Li & Raymond go further: they measure real job-output productivity—issues resolved per hour—for a relatively bounded customer-support occupation inside one company. That is important evidence that AI can improve meaningful workplace output, not just isolated exercises.

But none of these studies show that a composite senior role spanning architecture, engineering, product management, organizational change, governance, training, analytics, and executive communication becomes proportionally easier because AI accelerates parts of the work.

The International Labour Organization made a related point in a May 2026 research brief on the aggregation paradox of AI: large productivity gains observed at the task or worker level have not yet translated cleanly into comparable gains at the firm, sector, or economy-wide level. Organizational redesign, skills, complementary investment, and diffusion all matter.

The firm-level picture is already becoming more mixed. A 2026 Atlanta/Richmond Fed working paper based on a survey of nearly 750 corporate executives estimates positive AI-attributed labor-productivity growth in 2025 and reports that executives expected it to strengthen in 2026, with stronger expected gains in high-skill services and finance. That does not mean the aggregation problem has disappeared. It suggests that firm-level gains may be emerging unevenly as adoption matures.

Healthy role compression vs. role stacking

The term role compression is already used in different ways, so I am not claiming to have invented it.

The underlying job-design problem is not new either. Organizational research has long examined job enlargement and enrichment, role overload, work intensification, spans of responsibility, and the tradeoffs created when jobs broaden. The AI-era question is whether technology has genuinely redesigned those boundaries—or simply made accumulated scope easier to hide.

What matters here is the distinction between healthy AI-enabled role compression and what I would call role stacking—not to be confused with job stacking, the separate practice of holding multiple jobs at once.

Healthy role compression

Healthy compression happens when AI genuinely changes the work. Some work actually disappears. Some work becomes machine-assisted. The workflow is redesigned. The human moves toward judgment, exceptions, synthesis, relationships, prioritization, and decisions.

Role stacking

Role stacking is different. The old responsibilities remain. The old meetings remain. The old service levels remain. The old accountability remains. And then new disciplines are added because leadership assumes AI will somehow absorb the difference.

There is an important counterexample here: founders and very small AI-native teams may intentionally combine disciplines that would be separate in a larger organization. I do that myself.

AI agents can push this even further. They can absorb real execution capacity across research, implementation, testing, monitoring, documentation, and coordination. In a well-designed system, one human outcome owner supported by agentic systems, shared platforms, vendors, or fractional specialists may genuinely outperform a much larger traditional team.

In some AI-native operating models, the burden of proof may legitimately run the other way: if agents and shared platforms have collapsed enough coordination cost, preserving traditional functional handoffs may be the less efficient design.

The enemy is not breadth. It is breadth without subtracted work, independent challenge, real supporting capacity, or an operating bargain that matches the residual human load and risk.

The lesson is not that compressed roles never work. It is that scope, authority, support, compensation, and upside need to move together—and the bargain should be evaluated against the effort and residual risk the person still carries.

AI does not repeal human limits

AI can scale information processing, drafting, coding, pattern recognition, retrieval, and automation.

It does not eliminate finite human attention. It does not eliminate accountability. It does not remove the cost of context switching. It does not create five independent brains that can simultaneously hold five different organizational problems at full depth.

In experiments published in Organizational Behavior and Human Decision Processes, Sophie Leroy found that when people switch away from unfinished work, attention can remain on the prior task and performance on the next task can suffer—a phenomenon she describes as attention residue.

Microsoft's 2025 Work Trend Index illustrates the attention problem that already existed before companies started redesigning roles around AI. Based on aggregated Microsoft 365 telemetry, Microsoft reported that among the top 20% of users by ping volume, the average time between meetings, emails, or chat pings during an eight-hour workday was roughly two minutes.

Emerging field evidence suggests that this failure mode is worth taking seriously. UC Berkeley Haas researchers Aruna Ranganathan and Xingqi Maggie Ye describe ongoing research in which employees worked at a faster pace, took on broader scope, extended work into more hours of the day, and kept multiple AI-assisted workstreams active at once.

There is important counterevidence too. In a field experiment across 66 firms and 7,137 knowledge workers, Eleanor Dillon, Sonia Jaffe, Nicole Immorlica and Christopher Stanton found that workers randomly given access to an integrated GenAI tool who used it spent about two fewer hours per week on email and reduced work outside regular hours during the latter half of the six-month experiment. They did not detect broader changes in task quantity or composition from individual-level AI provision.

That contrast is exactly why I think the issue is design, not destiny. AI can intensify work, or it can genuinely give time back. What happens next depends on the operating model and the choices leaders make with the capacity created.

The scale asymmetry behind role compression

A company and a human being do not scale AI in the same way.

A company can deploy AI across thousands of employees, millions of transactions, hundreds of workflows, software platforms, proprietary data, and global customer interactions.

An individual can become dramatically more capable with AI, but still has one calendar, one attention system, one finite reserve of judgment, and one set of accountability boundaries.

Agentic systems can absorb real execution and monitoring capacity, which is exactly why the unit of analysis should not be human headcount alone. But whatever residual judgment, exceptions, accountability, and risk remain with the human still have to be designed explicitly.

The expertise pipeline matters too

Much of the work AI can remove first—research, first drafts, routine analysis, pattern recognition, lower-risk decisions, and repeated execution—is also work through which people historically developed expertise.

Automating inefficient developmental work can be progress. But organizations still need a credible answer to that question. The risk is not that every old junior task must be preserved. The risk is removing the old apprenticeship pathway without designing a new one.

The evidence here is still emerging. A 2026 randomized study of developers learning an unfamiliar programming library found that some forms of AI reliance reduced conceptual understanding, code-reading and debugging performance, while more cognitively engaged AI-use patterns preserved learning better.

The constructive implication is important: the choice is not necessarily AI or learning. The better question is how to design AI-supported work so people still practice the reasoning, review, challenge, and exception-handling through which expertise develops.

A practical Role Compression Test

When designing a senior AI-enabled role, ask nine questions.

1. What is the primary business outcome this person owns? A broad role can be coherent when several disciplines serve one outcome.
2. How many distinct functions still require meaningful human ownership? Do not count nouns in a job description. Count the functions whose decisions, exceptions, stakeholder obligations, or outcomes still materially depend on this person.
3. Which responsibilities has AI actually removed, automated, or materially redesigned? Assistance counts when it genuinely eliminates separate human ownership.
4. Where is the team? The answer may include employees, AI agents, shared platforms, vendors, or fractional specialists.
5. Who or what independently challenges the decisions? Independent peers, evaluation harnesses, audit controls, red-team processes, or external review can all provide challenge.
6. What happens when priorities collide? Name two likely near-term collisions and decide in advance which outcome wins and who breaks the tie.
7. Is authority aligned with accountability? Broad accountability without budget, staffing authority, decision rights, or executive sponsorship is not empowerment.
8. Do compensation, working time, opportunity, and upside reflect the changed economics of the role? When productive leverage, accountability, value creation, or residual risk change permanently, leaders should explicitly review the employee's side of the bargain.
9. Would this role still be healthy after two or three years—or is there an explicit transition plan? If the role is intentionally broad for a defined phase, include a credible split, succession, staffing, or scope-transition path.

What better AI-enabled role and transition design looks like

A strong senior AI role can still be broad. The difference is that the boundaries are visible.

Some of that capacity may be human. Some may be agentic, platform-based, shared-service, vendor, or fractional. The critical point is that it exists, is governed, and is not silently pushed back onto the role.

Smaller team and no supporting capacity are not the same concept.

Workers are not irrational to ask where they fit in the value equation

The International Labour Organization estimates that roughly one in four workers globally are in occupations with some degree of generative-AI exposure, while concluding that transformation is more likely than full replacement for most exposed jobs because human input remains necessary.

A Stanford Digital Economy Lab working paper revised in August 2026 likewise reports no evidence of widespread economy-wide AI displacement in its payroll data, but it does find a widening employment gap for workers ages 22–25 in highly AI-exposed occupations, driven primarily by reduced hiring rather than mass separations.

At the same time, worker concern is real. In a 2025 Pew Research Center report based on an October 2024 survey, 52% of U.S. workers were worried about future workplace AI use, while only 6% expected it to create more job opportunities for them in the long run.

Employers are sending mixed signals too. In the World Economic Forum's 2025 employer survey, 77% of employers said they planned to upskill workers in response to AI, while 41% planned workforce reductions where AI could automate tasks.

The more immediate question inside organizations is: What kind of jobs are we creating as AI changes the work?

The Shared AI Productivity Dividend is a transition-design question

Companies should benefit when they invest successfully in AI. They take risk. They buy technology. They build systems. They train people. They fund experimentation. They compete for customers. Better productivity should produce stronger companies and healthy returns.

Customers should benefit too through better products, faster service, lower friction, or lower cost.

But if AI also makes employees substantially more productive, I think leaders should deliberately ask how workers participate in that gain.

That participation does not have to mean one universal formula or a mandated percentage. The principle is not equal division. It is fair representation and meaningful participation in the gains as the economics of work change.

Nor is this a claim that workers have a fixed legal or economic entitlement to a percentage of AI-generated gains. It is a leadership and work-design question: when the economics of a role change, leaders should deliberately consider how the benefits and burdens are allocated rather than letting the answer emerge by default.

It can show up through better compensation, more meaningful work, reduced repetitive work, greater autonomy, more sustainable workloads, reskilling and internal mobility, stronger staffing around high-value human work, and more flexible or reduced working time where productivity supports it.

The OECD's AI Principles, which are addressed to governments, explicitly call for a fair transition for workers and say governments should promote responsible AI at work while aiming to ensure that AI's benefits are broadly and fairly shared.

A June 2026 ILO review adds an important current-state reality check: GenAI productivity gains are real but uneven, and worker-reported time savings have not yet consistently translated into higher measured output, earnings, or employment.

That makes the productivity dividend an operating-model question, not just a compensation question.

It is also why the issue eventually becomes larger than any one employer. Businesses need customers, and customers need purchasing power. If future AI materially reduces the amount of human labor required across large parts of the economy, society will need mechanisms that allow the resulting prosperity to continue circulating broadly enough to support demand, opportunity, investment, and social stability.

The leadership question

AI gives leaders a rare opportunity to redesign work from first principles.

Not just to automate tasks. Not just to reduce headcount. Not just to add more responsibilities to the people who remain.

To actually ask:

• What should humans stop doing?
• What should machines do?
• What judgment must remain human?
• Which roles should broaden?
• Which disciplines still need independent ownership?
• Where do we need fewer handoffs?
• Where do we still need checks and balances?
• How should productivity gains strengthen the company, the customer experience, and the working lives of the people creating the value?

That is a much more ambitious use of AI than squeezing more work into the same organizational chart.

The future of work should not be five jobs in one. It should be a better division of work between people, AI, and teams.

So the next time an AI-enabled job description stretches across strategy, architecture, engineering, product, governance, change, training, analytics, communications, vendor management, and people leadership, ask the question that exposes the operating model underneath it:

Author's note

This article is intentionally pro-AI, pro-business, and pro-worker. The goal is not to preserve old job boundaries simply because they are familiar. It is to distinguish genuine AI-enabled redesign from role stacking—and to build operating models that can sustain the productivity AI makes possible.

Research notes and sources

• Brynjolfsson, Li & Raymond — Generative AI at Work, QJE / Stanford
• Noy & Zhang — Experimental evidence on the productivity effects of generative AI
• Dell'Acqua et al. — Navigating the Jagged Technological Frontier
• ILO — The Aggregation Paradox of AI, 2026
• Baslandze et al. — AI, Productivity, and the Workforce
• Leroy — Attention residue and task switching
• Microsoft Work Trend Index — Breaking Down the Infinite Workday
• UC Berkeley Haas — AI promised to free up workers' time; researchers found the opposite
• Dillon, Jaffe, Immorlica & Stanton — Shifting Work Patterns with Generative AI
• Shen & Tamkin — How AI Impacts Skill Formation
• ILO — Generative AI and Jobs: A 2025 Update
• Stanford Digital Economy Lab — Canaries in the Coal Mine?, revised August 2026
• Pew Research Center — U.S. Workers and Future Workplace AI
• World Economic Forum — Future of Jobs Report 2025
• OECD AI Principles
• ILO — The impact of GenAI on jobs, productivity and work organization
• OECD Employment Outlook 2023 — AI, Job Quality and Inclusiveness

Frequently asked questions