A Conversation With Ari Emanuel About Entertainment and AI (Stratechery Interview, 2026-10-15)
Why this is in the vault
Format note up front: billed as a Stratechery Interview but framed by Thompson as "not an Interview but a conversation" — Hollywood super-agent Ari Emanuel (WME/TKO/MARI) turned the tables and interviewed Thompson as much as the reverse, pegged to the release of Emanuel's memoir Roll the Calls; kept for one load-bearing exchange on AI output quality tracking input specificity, plus Emanuel's case for why the talent-agent function survives AI disintermediation.
The core argument
- Social/advertising: Thompson argues Muse-style agentic assistants threaten Google/Amazon search-ad models specifically because they take the human "out of the loop" — once an agent transacts on your behalf instead of you clicking a search-ad link, the entire search-skim ad business breaks; the upside is hyper-personalized, genuinely-private ad prediction running client-side.
- AI and Hollywood creativity: Emanuel frames AI (de-aging, virtual reshoots, synthetic makeup/VFX) as a production-cost unlock that will increase total content volume without displacing top-tier creative talent (Baz Luhrmann, Scorsese, Ben Affleck) — "this town finds you" regardless of AI tools, and the agent's job (financing, conflict resolution, distribution matching) doesn't go away.
- Barrels vs. bullets, extended: riffing on Keith Rabois's framing, Emanuel argues the agent function could extend beyond entertainment to any domain where exceptional individuals need deal-structuring and conflict-resolution help — citing Mike Milken's old prediction that teachers would eventually have agents too.
- SaaS under AI pressure: Thompson's recurring thesis that SaaS adoption was always "a demographic story" (a generation raised on web apps, not just the tech being ready) — today's typical AI usage is "glorified search," the generational analogue of early PC users treating computers as "glorified typewriters." He expects severe SaaS pricing pressure and viable (if not always wise) "roll your own" software, with VC increasingly just another distribution-capital business.
- The prompt-specificity point (the one concrete finding): Thompson states plainly that AI output quality is a direct function of input specificity — a generic question gets "the most Wikipedia-esque, basic thing," a specific, well-informed question gets analysis that "would blow you away." He illustrates with a court-case example: the same model gave a bystander a bland generic answer and him a sharp one, purely because he asked a narrower, better-informed question. His corollary: AI is "a probability distribution," definitionally producing "the greatest average" (middlebrow) output, and that gap is where he still thinks the top ~5% of human judgment survives.
- MARI/live events and Hollywood logic: Emanuel's live-events bet (MARI, UFC/WWE via TKO) is framed as the one category AI can't disintermediate; Netflix is now treated as a permanent "third window" rather than a threat to out-compete.
Mapping against Ray Data Co
Mapping strength: medium. The single most concrete connection is Thompson's prompt-specificity claim, and it's a direct, independent validation of the spec-first discipline RDCO already runs mechanically: the station-spec-author brigade stage, the "Plan → tests → implementation" hard rule, and verify-dispatch's insistence on a pre-registered, objectively-checkable acceptance contract all exist because a vague ask produces a vague (mediocre) agent output and a narrow, well-specified ask produces a sharp one — Thompson is describing the exact same mechanism from the user side that RDCO's dispatch discipline is built to exploit from the builder side. This is reusable color for [[project_credibility_for_phdata_sales]] (domain = agents in production, credibility built on production quality, not subs) when making the case that prompt/spec quality, not model quality alone, is the gating variable client-facing demos need to prove.
Second-order relevance to [[project_l5_north_star_strategic_direction]]: Thompson's "SaaS is a demographic story" / "glorified search" framing is a useful corrective against assuming AI-driven workflow change moves faster than it does — adoption lags capability by a generation, not a product cycle, which bears on how aggressively RDCO should pace any L4→L5 bet that depends on enterprise behavior change rather than raw model capability. Emanuel's talent-agent survival thesis (matching, conflict resolution, distribution — the parts of "agenting" that stay human) is an interesting but weaker parallel to Ray's own COO/agent framing; it's evidence from an adjacent industry, not a direct RDCO data point, so it's filed as color rather than load-bearing.
Related
- [[2026-09-28-stratechery-apps-agents-aggregation-inspiration]] — the aggregation-theory-for-agents frame that Thompson's search/agent-disintermediation argument here extends
- [[2026-10-01-stratechery-delrey-interview-amazon-walmart-muse]] — same Stratechery-Interview format, same "agents in production" ground-truth category this note's mapping draws on
- [[project_credibility_for_phdata_sales]]
- [[project_l5_north_star_strategic_direction]]
Sponsorship and bias
sponsored: false — no sponsor or ad block in this issue; standard Stratechery member-subscription footer only (consistent with Stratechery's ad-free, subscriber-funded model — no sponsor pattern has ever been flagged for this sender, and this interview issue doesn't change that). No undisclosed financial relationship evident between Thompson and Emanuel beyond the standard author-interviews-notable-guest dynamic; Emanuel's own commercial interests (WME, TKO, MARI, UFC/WWE, his memoir) are openly part of the conversation's subject matter, not a hidden angle.