• Its capabilities are extraordinary - coordination costs are falling, translation layers can be automated, and expertise can gain direct leverage over execution in ways that were impossible even five years ago. And yet, already, we can see a familiar pattern forming. Many organisations are reaching for AI as a simple efficiency play inside structures designed for a previous era.
• On paper, most AI deployments look similar. The same models. The same copilots. The same orchestration platforms layered into finance, operations, customer service, product and strategy.
• The language is shared: augmentation, automation, leverage, productivity. But beneath that shared surface, two very different operating logics are taking shape.
• In some organisations, AI is being introduced as a margin stabiliser. Junior layers are reduced. Reporting structures remain intact. Agents are embedded into existing workflows to accelerate output and reduce cost, while decision rights and authority models remain largely untouched
• In more ambitious organisations, AI is treated as permission to ask more uncomfortable questions. If translation can be automated, why maintain translation layers? If coordination is cheaper, why preserve reporting ladders designed to aggregate information upward?
• Both paths can produce impressive short-term productivity gains. But only one opens up net new top line growth, and it does this by changing the organisation.
• The distinction is subtle at first. It does not show up in vendor announcements or pilot metrics. It shows up in what leaders choose to leave alone. It shows up in whether span of control is redesigned or simply expanded. It shows up in whether apprenticeship is re-imagined or quietly eroded. It shows up in whether authority is redistributed toward outcomes, or insulated behind more efficient reporting.

• xtraction Mode: The Frame Defending Itself

  Extraction Mode often presents as pragmatism. Budgets are under pressure, markets are volatile and boards are demanding visible returns. AI offers immediate gains in efficiency, speed, and headcount flexibility. In that context, embedding agents inside existing workflows feels not only rational, but responsible.
• Executive oversight becomes more data-rich, but not structurally closer to execution.
• But decision rights are still flowing upward and authority is still tied to hierarchy rather than outcome ownership
• Extraction Mode can produce good numbers in the short term. It can stabilise margins and extend runway. But it does so by reinforcing the underlying frame:

• Redesign Mode begins with a different instinct. Instead of asking, “Where can AI remove cost?” it asks, “What assumptions about structure that were built into the organisation when coordination was expensive are no longer valid?”
• In Redesign Mode, AI is not inserted into the existing machine, instead it is used to reveal its architecture, and then improve it.
• Some functions may consolidate. Others may fragment into outcome cells with explicit guardrails and escalation rules.
• The shift is subtle - AI is treated not as an efficiency layer but as structural permission. Coordination is cheaper; therefore, the organisation does not have to be shaped the way it was when coordination was scarce.

• Redesign Mode sounds compelling in theory. Few leadership teams would openly argue that preserving outdated structures is the goal. And yet, when AI initiatives move from pilot to budget to restructuring, most organisations tilt toward extraction.
• Cost reduction is measurable. Structural redesign is not. A headcount number can be reported to the board next quarter. A re-architected decision-rights model compounds over years
• Governance models amplify this bias. Boards understand margin expansion. They are less fluent in organisational topology.
• Over the past decade, most transformation effort has been directed downwards. Teams were asked to become more agile, managers were asked to embrace digital tools, frontline functions were reconfigured. Senior leaders, in many cases, changed less.
• Redesign Mode would require leaders to update their own operating models: to move closer to outcome boundaries, to make judgment legible, to relinquish some insulation provided by hierarchy. That is harder than reducing cost!
• There is also a subtler force at work. Most organisations have been optimised for decades around information asymmetry. Authority was justified by access: access to data, to strategic perspective. AI reduces that asymmetry, but the habits built around it remain.
• Every hierarchy contains an implicit learning pathway. Entry-level roles absorb repetitive work. They sit close to process and they observe decisions being made. They develop judgment slowly through exposure, error, and proximity. Over time, some of those individuals move upward, carrying tacit knowledge with them.
• When junior layers are removed without structural redesign, repetitive tasks disappear, but so do many of the early exposure points where judgment is formed. Automation replaces execution without replacing apprenticeship. The pyramid thins, but the pipeline narrows.

• 1. Rewrite One Decision Rights Map
• 2. Audit One Reporting Layer for Translation vs Judgment
• 3. Redesign One Apprenticeship Pathway Alongside Automation

• 4. Define One Outcome Cell
• In Redesign Mode, AI will be treated as a structural inflection point. Coordination costs will fall and structure will change in response. Decision rights will be clarified. Span will be redesigned. Apprenticeship will be rebuilt. Authority will move closer to outcomes rather than further from them.
• The uncomfortable truth is that the technology is not the constraint. Model capability is advancing rapidly. What will determine whether this transformation compounds advantage or quietly stalls is whether leaders are willing to question the frame that has shaped enterprise design for decades.

• AI does not choose between extraction and redesign.

  We do.

28 more annotations...