Chapter 11: A Call to Reinvent Yourself in the Age of Multi-Agent Systems

Companies optimize what they already do; pioneers reinvent what they will do.

Optimizing Is Not Reinventing: The Distinction That Changes Everything

The author issues a warning that deserves to be amplified: most companies are focused on optimizing what exists, when the era of multi-agent systems demands reinventing. The distinction is not semantic, and it is worth pinning down before going on, because it determines whether the investment in AI produces durable competitive advantage or simply cuts transient operating costs.

Optimizing is improving a process within its paradigm: doing what is already done faster, cheaper, or with fewer errors. Reinventing is questioning the paradigm itself: asking whether what is done should be done that way at all, or whether the availability of a new capability makes possible (and necessary) a radically different way of producing the same result, or a result that was previously impossible. Using a swarm of agents to process 20% more invoices per hour is optimization. Redesigning the entire financial cycle around agents that close the month in real time, eliminating the very concept of a “monthly close,” is reinvention.

Why Multi-Agent Systems Force Reinvention: The Physics of Phase Change

Multi-agent systems are not just another tool to slot into the current process; they are a capability that alters the physics of the problem. When a swarm can handle in parallel what used to take an entire department and days of work, the org chart that structured that department stops making sense in its original form. When intelligence (the capacity to reason, synthesize, and decide on unstructured information) becomes abundant, cheap, and distributable, the bottlenecks move to unexpected places: from execution to judgment, from generation to validation, from cognitive capacity to trust and accountability.

In materials physics, a phase change is not a quantitative change within the same state; it is a transition to a state with qualitatively different properties. Water doesn’t get “colder” as it freezes: it changes state. Multi-agent systems represent for the organization exactly that kind of transition: not more speed in the current state, but an organizational state with qualitatively different properties. Inserting multi-agent systems into structures designed for a scarcity of intelligence is like installing a jet engine on a cart (the power exists, but the chassis wastes it and often turns it into a risk).

McKinsey’s Three Horizons in an Agentic Key

The Three Horizons framework, developed in The Alchemy of Growth (Baghai, Coley & White, 1999) and refined by McKinsey over two decades, holds that managing organizational growth requires attending to three time horizons at once, each with different investments and logic. Applied to agentic transformation:

Horizon	Timeframe	In agentic transformation	Typical investment	Risk if ignored
H1	0–12 months	Automating repetitive tasks with agents: visible cost savings, quick ROI	60–70% of the AI budget	Loss of operational efficiency relative to competitors
H2	1–3 years	Redesigning processes around agentic capabilities: new flows, new roles	20–30% of the AI budget	Loss of competitive position within 18–24 months
H3	3–7 years	Reinventing the business model: which new markets, products, or relationships agentic AI makes possible	10–20% of the AI budget	Structural irrelevance in the medium term

The most frequent mistake in organizations starting agentic transformation is to concentrate 90% of the budget in H1 (visible, measurable, low-risk) and allocate nothing to H3. The result, on a three-to-five-year horizon, is an organization that more efficiently runs a business model its competitors are reinventing.

Map of Reinvention: An Inventory of Cognitive Assumptions

Reinventing yourself requires, as a non-negotiable first step, identifying which of the business’s assumptions rested on a scarcity of cognition. This exercise (which could be called a “cognitive assumptions audit”) consists of asking, for each relevant process in the organization: does this process exist in this form because it is the best way to produce that result, or because it was the only feasible way when intelligence was a scarce and costly resource?

The answers systematically reveal three categories of processes:

Processes that remain optimal regardless of the availability of artificial intelligence: those where value comes from human relationships, ethical judgment, genuine creativity, or legal responsibility that cannot be delegated. These processes should be protected from thoughtless substitution.

Processes that should be redesigned to incorporate agentic capabilities without changing the end result: their current structure is an artifact of past cognitive scarcity, and redesigning them produces the same result at lower cost, higher speed, or higher quality.

Processes that should be eliminated because their very existence is an artifact of limitations that no longer exist: the weekly report nobody reads but that gets produced because “it’s always been done”; the alignment meeting that exists because information didn’t flow in real time; the low-risk approval committee that exists because manual review was the only control mechanism available.

Empirical Cases of Agentic Reinvention: Healthcare, Finance, and Law

The evidence of agentic reinvention in regulated sectors is already robust enough to draw patterns from:

Healthcare: the LLaVA-Med system (Microsoft Research, 2023) and multi-agent imaging diagnosis systems don’t automate the radiologist’s diagnosis; they reinvent the flow: the agent processes 80% of routine images and presents the specialist only the cases that require expert judgment. The radiologist stops being an image processor and becomes an exceptions reviewer. The result is not “the same thing, faster”; it is a qualitatively different role.

Finance: JP Morgan Chase reported (2024) that its legal contract assistance system, based on a multi-agent LLM, cut the time to review commercial loan contracts from 360,000 hours a year to a matter of hours. The process was not optimized: it was reinvented. Lawyers stopped reviewing standard contracts and moved on to managing exceptions and high-value relationships.

Law: the Harvey system (used by Allen & Overy since 2023) lets junior associates produce senior-quality documents in hours, not weeks. The salary pyramid of law firms (which existed precisely because junior-level work was necessary to sustain senior-level work) is being reinvented, not optimized.

The Courage to Cannibalize Your Own

Reinventing yourself hurts, because it often requires actively cannibalizing what works and produces revenue today. This is not just rhetoric: a company that announces “we’re going to reinvent our operation with agents” is announcing, implicitly, that some of its current lines of business or some of its profitable processes will disappear or shrink. Internal resistance to that decision is proportional to the present success of what is going to be cannibalized.

Christensen’s (1997) research on the innovator’s dilemma (originally published in Harvard Business Review and expanded in his book of the same name) documented this mechanism precisely: the companies that best execute their current business models are the most vulnerable to disruption, because their organizational structure, success metrics, and incentives are aligned with optimizing what exists, not with exploring what is new.

In the agentic era, this dilemma presents itself with a speed unprecedented in this sector: the obsolescence cycle of a process that doesn’t adapt can be 18 to 36 months, not a decade. History is unyielding in its verdict: those who optimized the horse lost to those who reinvented transportation.

An Executive Roadmap for Agentic Transformation

■ EXECUTIVE TRACK

An agentic reinvention roadmap that is implementable (not just aspirational) needs four components that most roadmaps leave out:

1. The inventory of cognitive assumptions (quarter 1): who does it, which tools they use, which categorization criteria they apply, which outputs they produce.

2. The reinvention pilot (quarters 1-2): one process from the “redesign” group run in real production with agents, with outcome metrics (not process metrics) defined before starting. Without a real pilot, reinvention stays in the territory of the declarative.

3. The agentic governance model (in parallel from quarter 1): who decides what autonomy to grant to which agent, under what conditions, with what review mechanism. Without governance, reinvention produces chaos; with excessive governance, it produces slowness.

4. The judgment training program (ongoing): teams working with agents need to develop the hardest and most overlooked competency in AI transformation programs: judging the quality of agentic output. It is not knowing how to use the tool; it is knowing when the output is good and when it isn’t. This competency is not acquired in a two-day course.

Alternative version (higher academic density):

The organizational transformation induced by the availability of multi-agent artificial intelligence systems is a phenomenon that goes beyond incremental technological innovation and enters the territory of what Christensen (1997) called disruptive innovation: a kind of change that does not improve performance on the metrics current customers value, but introduces a different set of properties (greater availability, lower cost, wider distribution) that redefine which attributes are valued. The optimal adaptive response to this type of disruption is to build in parallel an alternative model able to capture the new value, even at the cost of partially cannibalizing the first, not to optimize the existing business model.

Key Takeaways from Chapter 11

• Optimizing is improving within the paradigm; reinventing is questioning the paradigm. Only the second produces durable competitive advantage in the agentic era.
• Multi-agent systems represent a phase change for the organization: not more speed in the current state, but a state with qualitatively different properties.
• The Three Horizons framework applied: H1 (automation, 0–12 months, 60-70% of the budget), H2 (redesign, 1–3 years, 20-30%), and H3 (reinventing the business model, 3–7 years, 10-20%). Ignoring H3 produces structural irrelevance.
• The cognitive assumptions audit sorts processes into three groups: optimal (protect), redesignable (transform), and eliminable (discontinue).
• JP Morgan Chase, Allen & Overy, and multi-agent medical systems show that reinvention is already happening in regulated sectors at real scale.
• Christensen’s dilemma (1997): the best executors of the current model are the most vulnerable to disruption; in the agentic era, the obsolescence cycle is 18–36 months.
• The executive roadmap needs four non-negotiable components: an inventory of assumptions, a pilot in real production, agentic governance, and a judgment training program.