The traditional hierarchy is giving way to a smaller human-agent network.
Your company was designed for a world in which intelligence was scarce, expertise had to sit on payroll, and coordination required layers of management.
That world no longer exists.
AI is making intelligence abundant, expertise callable, and coordination programmable.
On March 4, I published an argument through Forbes Technology Council called The Unbundling of the Firm.
My thesis was intentionally provocative:
AI is not here to improve the company. It is here to dismantle it.
The argument began with Ronald Coase. In 1937, he asked why companies exist at all. His answer was transaction costs. When finding talent, negotiating work, sharing information, monitoring quality, and enforcing agreements are expensive, it makes sense to bring people inside one organization and coordinate them through management.
AI changes that equation.
An agent can search, analyze, write, code, compare, monitor, and execute without requiring a new department for every capability. A small team can access expertise and operating capacity that once required a large payroll.
The direction was right. The destination was too simple.
The minimum viable firm is shrinking. The maximum viable firm may be expanding.
AI does more than make small companies more capable. It can also make enormous companies easier to coordinate.
The pressure lands on the organizations in between.
The signal
This is no longer only a provocative prediction.
The authors of an NBER working paper call the broader shift a Coasean singularity. They argue that AI agents could dramatically reduce transaction costs across search, communication, contracting, enforcement, and verification. They also make clear that the net effects remain an empirical question.
McKinsey has applied almost the same logic to strategy. Its 2026 analysis says smaller, specialized firms connected through AI-mediated ecosystems could operate with economics once reserved for large, integrated companies.
Microsoft proposes a related organizational model in its 2025 Work Trend Index. Instead of a static org chart, it describes a dynamic Work Chart where lean teams form around outcomes and use agents as digital colleagues. In its survey, 46 percent of leaders reported that their companies were using agents to fully automate workflows or processes.
Then there is operating evidence. Klarna reported that its AI assistant handled 80 percent of its customer-service chats in 2025. The company also reported that annual revenue per employee rose from about $344,000 in 2022 to $1.24 million in 2025 while headcount fell 49 percent from Q4 2022.
Those company-reported numbers do not prove that AI caused every change, or that every business should copy Klarna. They do show that output and human headcount no longer have to move together in the old way.
The pattern will not be universal. Gusto payroll research covering about 7,700 small businesses found that greater AI exposure was associated with higher revenue and slightly more hiring six months later. The study measured exposure, not confirmed AI adoption, and it does not establish that AI caused those gains. Some firms will use productivity to shrink. Others will use it to expand.
The deeper shift is not a universal move toward fewer or smaller companies.
It is a shift from fixed boundaries to programmable ones.
What most people miss: AI cuts two different costs
AI can reduce the cost of coordinating work outside the company.
That favors unbundling. A small firm can rent specialized capabilities, deploy agents, and connect providers without building every function internally.
AI can also reduce the cost of coordinating work inside the company.
That can favor concentration. In An Economy of AI Agents, Gillian Hadfield and Andrew Koh argue that artificial agents may transmit information quickly, reduce some monitoring costs, and allow data and algorithmic improvements to be duplicated across a firm. Those mechanisms could make scale more powerful, not less. They also identify new coordination, opacity, and alignment problems, so internal coordination does not become frictionless.
This produces a more interesting possibility.
AI may create a barbell economy:
On one side, tiny AI-native firms with very little organizational baggage.
On the other, giant platforms that control data, compute, distribution, trust, or customer relationships.
In the middle, companies carrying layers of coordination that no longer create enough context, judgment, or accountability to justify their cost.
The middle is not vulnerable because it is medium-sized.
It is vulnerable when it pays a high coordination tax without turning that coordination into an advantage.
Every layer that mainly routes information, waits for approval, or translates priorities now has to justify itself.
The AI Boundary Test
The practical question is not, "How many people can AI replace?"
It is, "Where should the boundary of this company sit now?"
For any recurring workflow, ask four questions.
1. Can we specify the outcome?
Can we clearly describe the required result, constraints, inputs, and acceptable failure conditions?
If not, the work still depends heavily on human interpretation.
2. Can we verify the result?
Can the output be checked cheaply, consistently, and independently?
If execution is easy but verification is expensive, you have not removed the work. You have moved it into review.
3. Does context compound?
Does the workflow improve when it has access to proprietary history, customer knowledge, operational data, and lessons from prior decisions?
If that context compounds inside the organization, keeping the capability close may create an advantage.
4. Who owns the consequence?
Does the decision require trust, empathy, professional judgment, legal responsibility, or a human willing to stand behind the outcome?
If yes, accountability remains a human job even when much of the execution becomes digital.
These questions reveal three possible designs:
Unbundle it: The outcome is clear, verification is cheap, and the capability is widely available.
Agent-enable it: Agents perform much of the work, while an accountable human owns the judgment and exceptions.
Keep it core: Proprietary context, trust, and responsibility are the source of the advantage.
The one-person company is possible in more categories than before. It is not automatically the optimal design for every mission.
This week's move: Redraw one workflow, not the whole company
Choose one recurring workflow that crosses at least two people or departments.
Write down:
The outcome the workflow must produce.
The steps an agent can execute.
The information the agent needs.
The checks required before the work moves forward.
The moment where a human must make or own the decision.
The context that becomes more valuable when retained.
Then make one decision: unbundle it, agent-enable it, or keep it core.
Do not begin with layoffs or a reorganization.
Begin with the economics of the work.
Your org chart is a record of coordination decisions made when intelligence and expertise were expensive. AI is changing those costs.
The company of the future is not automatically smaller. It is more deliberate about what belongs inside.
Reply and tell me: Which recurring workflow in your company is paying the highest coordination tax today?
The companies that win will not be the ones with the most agents or the fewest employees.
They will be the ones that know exactly where human judgment, proprietary context, and accountability still create value.
The firm is not disappearing.
Its boundaries are becoming programmable.
Until next week,
Burhan