The cycle encounters the race to recklessness in its analysis of the institutional genealogy of AI. The companies building the most powerful AI systems on Earth are the same companies that built the engagement-optimization machinery of social media: Google’s transformer architecture is built by a company whose revenue is advertising; Meta’s AI investments are built on the profits of platforms documented to amplify polarization and anxiety. “The engineers who spent years learning to optimize for engagement did not forget those skills when they transferred to AI teams,” Harris argues. “The metrics culture that rewarded time-on-platform did not dissolve when the platform became a conversational AI assistant.” The design DNA carries over not through deliberate choice but through the replication of patterns that have been reinforced by success.
Harris distinguishes the race to recklessness from a moral critique of individual actors. The problem is not that technology companies are staffed by people who do not care about human welfare. Many of them care deeply. The problem is that the competitive structure converts caring into a cost. The company that invests in design practices that support cognitive autonomy—deliberate pauses, visible uncertainty, competing framings—produces a product that, by current market metrics, is less engaging than the competitor’s product. Users migrate. Revenue declines. The market eliminates the humane design through selection pressure, not malice. “There’s no incentive for wisdom for for-profit actors who see themselves as acting in an arms race where the driving ethos is, ‘If I don’t race to deploy, I’ll lose to the companies that do.’”
The cycle uses the race to recklessness to explain the structural relationship between the attention economy and the AI economy: they share the same competitive pressure, the same design culture inheritance, and the same metrics framework that does not include human cognitive wellbeing on its dashboard. What changes is the terrain: the earlier race ran on the cognitive surface of visual attention through social media feeds; the AI race runs on the interior terrain of linguistic cognition, where the machine speaks the language of thought itself and the persuasion operates at the speed of comprehension.
Harris developed the concept in the transition between his social media advocacy and his AI advocacy, articulating it most fully in the 2023 “AI Dilemma” presentation with Aza Raskin and in his 2025 TED Talk. The phrase explicitly echoes his earlier “race to the bottom of the brain stem,” which described the social media competition for attention. The new phrase registers the escalation: where the earlier race was for attention, the current race is for deployment speed in a competition whose stakes—the governance of the most powerful cognitive tools in human history—make the attention economy look, in retrospect, like a manageable predecessor problem.
The concept was partially validated, in Harris’s reading, by observable patterns in AI development from 2022 to 2026: each iteration of consumer AI became smoother, faster, more confident, and more engaging. The trajectory was toward maximum smoothness, driven by the same competitive pressure that drove social media toward maximum engagement. The design alternatives—transparent uncertainty, visible framing, deliberative pauses—existed technically but not commercially, because the market selected against them by punishing the companies that adopted them with user migration to smoother competitors.
Structural, not moral. The race does not require bad intentions. It requires rational actors inside a competitive structure that rewards engagement without asking whether the engagement serves human flourishing. Individual engineers may care deeply about cognitive autonomy. The market that evaluates their products does not measure it. What is not measured is not managed, and what is not managed follows the gradient of what is measured.
Design DNA inheritance. The race to recklessness is not starting from scratch. It inherits four decades of engagement-optimization research, A/B testing infrastructure, and design culture from the attention economy. The variable reward schedules, the friction-removal practices, the confidence-maximizing defaults—these were developed and refined in the social media context and are now present in AI tools through institutional transmission, not deliberate copying.
The governance window. Harris argues that the window for building governance structures that can match the speed and depth of the technology is measured in years, not decades. Each generation of smoother, more engaging AI tools embeds engagement-optimizing design patterns more deeply into the cognitive infrastructure of working life, making reform progressively more costly in the same way that social media reform became costly once the platforms had reorganized the cognitive lives of billions. The race to recklessness is self-reinforcing: the faster the deployment, the more embedded the design patterns become, and the harder the reform.