Absorb → Automate → Unbundle — Three Phases of Technology Deployment

The three phases

Phase 1: Absorb

Organizations adopt the new technology by plugging it into existing workflows. Nothing structural changes — the org chart, the business model, and the customer experience stay the same. The technology is additive.

In AI (2023-2025), this looks like: coding copilots, marketing copy generation, summarization of meetings and documents, chatbots on customer support pages. The work was already happening; the tool changed.

Phase 2: Automate the obvious

Organizations identify known manual tasks and replace them with the new technology. This requires more organizational change than absorption — roles shift, headcount adjusts, processes redesign — but the categories of work stay the same. You’re doing the same things, faster and cheaper.

In AI (2024-2026), this looks like: automated data entry, automated document processing, automated first-pass code review, automated compliance screening. Systems integrators (Accenture, Deloitte) thrive in this phase because the question is “what should we automate?” and the answer requires domain knowledge, not technology insight.

Phase 3: Unbundle

The phase that actually creates and destroys industries. Organizations discover that things they thought were naturally bundled together were actually bundled because of constraints the new technology removes. When those constraints disappear, the bundle breaks apart, and new products and business models emerge in the gaps.

The historical precedent is instructive: online distribution unbundled physical proximity from retail. Music albums were bundles — unbundled into singles by iTunes. Newspapers were bundles of news, classifieds, comics, and opinion — unbundled by Craigslist, blogs, and social media. Cable TV was a bundle — unbundled by streaming.

In AI, the unbundling questions are: What things are currently bundled together because a human had to be in the loop? What decisions are bundled because the cost of analysis exceeded the value of separating them? What products exist because personalization at scale was impossible?

Evans’ slide deck poses this as the central strategic question: “What can LLM automation unbundle? What things did we not realise were bundles?”

Why unbundling is where the value is

The absorb and automate phases produce efficiency gains — doing the same work cheaper. The unbundling phase produces new markets and new categories. Every major platform shift (PCs, web, smartphones) generated most of its long-term value in the unbundling phase, not the automation phase.

The pattern also explains why early predictions about new technologies are consistently wrong: they focus on automation (“AI will replace X jobs”) when the real transformation comes from unbundling (“AI will dissolve the reason X and Y were bundled together, creating entirely new categories”).

The “infinite interns” frame

Evans characterizes AI’s current capability as “infinite interns” — cheap, fast, error-prone, requiring supervision. The strategic question is not “what can interns do?” (that’s automation) but “what changes when the cost of having an intern do something drops to near-zero?” (that’s unbundling via the Jevons paradox).

The steam engine analogy: by 1900, steam engines gave Britain the equivalent labour of roughly 5x its total population. The value wasn’t in replacing existing workers — it was in making entirely new categories of work economically viable.

See Jevons Paradox vs Cognitive Displacement - The Unresolved Tension for the live debate on whether AI follows the Jevons pattern (demand expands to absorb the cheaper input) or the displacement pattern (the cheaper input replaces humans without expanding demand).

Implications for the “where does AI value accrue” question

The Absorb → Automate → Unbundle framework maps onto the convergence thesis from The Agent Is the Customer - A Convergence Thesis on Where AI Value Accrues:

Absorb = incumbents win (they have the workflows to plug AI into)
Automate = systems integrators and vertical SaaS win (they know what to automate)
Unbundle = native AI companies win (they build for the unbundled world, not the bundled one)

The bifurcation between Category A and Category B incumbents may partly be a question of which phase the incumbent is operating in. Category A incumbents are already in the automate phase and moving toward unbundle. Category B incumbents are still absorbing.

AI + Business Model — parent hub
Jevons Paradox vs Cognitive Displacement - The Unresolved Tension — the “infinite interns” frame directly invokes Jevons
The Agent Is the Customer - A Convergence Thesis on Where AI Value Accrues — the unbundle phase is where native AI companies flip incumbents
AI Is an Industrial Bubble, Not a Financial One — Evans’ capex data ($400B in 2025) supports the industrial bubble framing
Circular Financing as a Bubble Signal in AI — Evans calls out the same Microsoft↔OpenAI circular revenue pattern
Run the Business vs Do the Work - The AI-Era Vertical SaaS Shift — “do the work” products are unbundling what was previously bundled inside “run the business” platforms
System of Action - Evolution Beyond System of Record — the system of action is an unbundling of decision-making from record-keeping
Ben Evans - AI Eats the World — source clipping (90-page slide deck, November 2025)