Drew Breunig, the independent analyst who publishes long-form dispatches on the AI industry at dbreunig.com, spent part of 2026 at Foo Camp, Open Frontier, and the AI Engineering World’s Fair collecting notes on how the sector actually moves. He organized what he saw using Stewart Brand’s Pace Layers framework, built in The Clock of the Long Now to explain how civilizations survive change without collapsing. Breunig’s application splits the AI ecosystem into fourteen tiers, running from prompts that turn over in days to energy production that shifts across decades.
Brand’s original insight was structural, not metaphorical. A resilient system needs layers that move at different speeds: fast layers test ideas cheaply and fail often, while slow layers absorb only what has proven durable and refuse to move for anyone’s product cycle. Breunig’s fourteen layers run, roughly, from prompts and tool integrations that turn over in days, through agent harnesses and model weights that shift over months. Further up sit chip design, university curricula, data center construction, and energy production, which move across years and decades. Each layer answers to the ones beside it. None of them answer to a news cycle.
The frame’s real value is separating what builders think is the constraint from what is actually holding them back. A product team watching model weights improve every few months assumes the whole ecosystem moves at that tempo. It does not. The chips those weights train on take years to design and fabricate. The data centers that house them take years to permit and build, a timeline Breunig notes has already compressed dangerously from a historical decade-plus norm. Power grids feeding those data centers move slower still, and the university programs producing the next decade’s researchers move slower yet. Builders who mistake the pace of their own layer for the pace of the whole system consistently overestimate how fast the rest of the stack can catch up.
This mismatch explains a specific kind of AI backlash that has nothing to do with model capability. Data center buildout is being pushed to move at the speed of compute demand rather than the speed of the communities, water tables, and electrical grids it depends on. Governance is being asked to legislate at the pace of product launches rather than the pace at which institutions can actually deliberate. Brand’s warning, restated in plain terms, holds that when a fast layer pushes a slow layer past its natural speed, the slow layer eventually breaks or pushes back. Current fights over data center siting and AI energy demand read less like simple local resistance and more like the exact friction his model predicts.
The frame strains on governance specifically. Breunig places governance and safety among the slow, years-to-decades layers, alongside universities and infrastructure. Regulation has not behaved like infrastructure, though. The EU’s AI Act and a wave of state-level rules in the United States moved in response to specific fast-layer incidents, not on a generational clock. Slow layers are supposed to provide stability precisely because they resist sudden change. When governance lurches in reaction to a viral misuse case or a single earnings call, it behaves like a fast layer wearing a slow layer’s costume. That failure mode is more dangerous than the one Breunig foregrounds. It removes the one layer the rest of the system is meant to rely on.
Read seriously, the model works as a diagnostic tool for roadmap risk, not just an explanation for backlash. A team should map every dependency in its plan onto Breunig’s fourteen layers and name the slowest one it actually relies on. That dependency might be a chip supply contract, a regulatory approval, or a hiring pipeline for specialized researchers. That slow layer, not the model’s monthly benchmark gains, sets the real delivery date. Before committing to a launch timeline built on the assumption that models and tools keep compounding weekly, check which of the years-to-decades layers underneath the plan has to move first. Build the schedule around that constraint instead.
Analysis draws on Drew Breunig’s July 3, 2026 essay on dbreunig.com applying Stewart Brand’s Pace Layers framework to the AI ecosystem.
