Here's the 90 Slide 'AI Eats the World' Talk in 15 Minutes—Plus My Top Takeaways

Benedict Evans’ “AI Eats the World” framing lands on a simple but consequential idea: AI is no longer a speculative breakthrough—it’s becoming inevitable utility, and the real strategic question is where competitive advantage survives as margins, winners, and organizational power shift. Speaking to senior leaders in Singapore, Evans treated AI as a moving label for successive technical waves—databases, search, classical machine learning, and now large language models—arguing that teams keep calling it “AI” only while it’s novel. Once it works, it stops feeling like a revolution and starts behaving like infrastructure. That shift matters because it changes what executives should worry about: not whether AI will arrive, but how value chains and operating models will be reorganized around it.

Evans also used a platform-cycle lens to describe predictable investment waves. Each wave draws massive capital, reshuffles winners and losers, and yet rarely deletes earlier layers. The result is fractal adoption: new AI capabilities stack on top of existing tools rather than replacing them wholesale. Even as big tech pours hundreds of billions—potentially trillions—into data centers and GPUs, the competitive landscape is moving toward models that function more like commodity inputs. That doesn’t mean intelligence disappears; it means the “model” itself is less likely to be a durable moat. The transcript adds nuance by contrasting frontier-led innovation with the quality of open-source distillations, citing a separate deep study suggesting many Chinese open-source models rely heavily on US frontier models and may be less generally flexible.

The most practical leadership warning centers on adoption. Evans’ core point: many organizations run pilots but far fewer use AI daily inside core workflows. Adoption is also described as path dependent—lumpy at first, then compounding based on where teams start. The analogy is spreadsheets: early adopters didn’t just gain speed; they reorganized information flow, enabled self-serve scenario modeling, and changed who “owned the numbers.” With LLMs and agentic systems, the beachhead choice determines which downstream workflows become possible—such as agent-assisted onboarding or engineering support—and which benefits never materialize.

A second-order implication follows from commoditization: organizations should prepare to act like buyers with leverage. Instead of betting on a single “model shop,” the transcript argues for multimodel architectures that route workloads based on cost, latency, data sensitivity, and jurisdiction—reducing lock-in and preserving bargaining power.

Finally, AI is portrayed as an org-chart changer, not just a tech-stack upgrade. Agent-style systems that can read and act across email, Slack, tickets, and dashboards resemble an “informal chief of staff,” shifting bottlenecks from execution to coordination, synthesis, and constraint-setting. That means management layers, span-of-control assumptions, and hiring plans will need to adapt faster than in prior cycles.

The takeaway for leaders is to step back regularly and ask whether the week’s breakthroughs alter strategic operating reality—tech adoption timing, information flow, org structure, and vendor power. In a relentless news cycle, the proposed antidote is disciplined synthesis: distill, reflect, and return with conviction so teams can move with clarity rather than churn.