# The Next Horizon in Agents — Bret Taylor ## 🗂️ Triage — annotated highlights surfaced > [!note]- Cross Reference > - **Sierra prices Horizon on outcomes, not tokens** — the one falsifiable commitment in the thread; vendor absorbs token-spend risk. Connects to [[Subscription Inference Is Sold Below Cost Creating a Tokenmaxxing Arbitrage]]. #xref #share > - **Customer-interaction memory is framed as a compounding moat** — strategically loaded, least-evidenced claim; flag as asserted, not shown. #xref ## 🧭 Bib capsule Bret Taylor (Sierra), "The Next Horizon in Agents," X thread, 2026 — https://twitter.com/btaylor/status/2077767352431882355 ## 🧠 Author's thesis (Adler Q1+Q2) Taylor announces Horizon, Sierra's platform for agents that pursue long-horizon goals (loan origination, prior authorization, referral scheduling) across days-to-months of customer interaction. The pitch reframes enterprise agents from single-conversation experiences to delivered business outcomes, backed by a context engine plus long-horizon planner, and pairs it with outcome-based pricing where customers pay for outcomes rather than tokens. ## 🧩 Key arguments - Horizon redefines the category from experiences to outcomes — agents that drive multi-step revenue results, not just service conversations. - The missing platform pieces are a context engine (to stitch interactions) and long-horizon planning; accumulated interaction memory is claimed as a durable moat. - Outcome-based pricing shifts token-spend risk onto the vendor — the thread's clearest business-model innovation and its most falsifiable claim. ## 🗒️ Notes — Literature atoms ### Horizon shifts agents from conversations to long-horizon outcomes > [!quote] > Where agents to date have delivered customer experiences, Horizon delivers outcomes. - **Where**: "Introducing Horizon" section - **Why**: The thread's thesis in one antithesis; the cleanest handle for how the agent category is being redefined. ### Context engine plus long-horizon planning fills the platform gap > [!quote] > What existing AI platforms are missing is the context engine to stitch together those interactions, and long-horizon planning - **Where**: "Why Horizon" section, healthcare referral example - **Why**: Names two concrete architectural components — usable vocabulary for evaluating any agent-platform claim. ### Sierra prices Horizon on outcomes, not tokens > [!quote] > you don't pay for tokens, you pay for business outcomes delivered - **Where**: Final section - **Why**: The one falsifiable commitment; the pricing shift may outlast the technical framing. ## 🔖 Permanent candidates - Outcome-based pricing transfers AI cost risk from buyer to vendor — Paying for outcomes instead of tokens moves token-spend risk onto the vendor, reshaping both the buyer's procurement math and the vendor's agent behavior. (No existing Permanent covers the risk-transfer mechanics — worth promoting; attribution disputes are the open condition.) ## Links - [[AI & the Future of Knowledge Work MOC]] — home MOC; long-horizon outcome agents reshape what enterprise knowledge work gets delegated and how vendors capture value - [[AI Cost & Tokenomics MOC]] — the outcome-vs-token pricing shift is a direct tokenomics move: who bears variable inference cost - [[Subscription Inference Is Sold Below Cost Creating a Tokenmaxxing Arbitrage]] — contrast on cost-bearing: outcome pricing has the vendor absorb token risk that flat subscriptions currently subsidize - [[3. Resources/Permanent Notes/Long-Horizon AI Shifts the Bottleneck From Model Capability to Memory and Effort Scaffolding]] — evidence for the architecture claim: Horizon's context engine and planner are the memory scaffolding that becomes binding once an agent must act across days or months. ## ✅ Processing checklist - [[AI & the Future of Knowledge Work MOC]] — locates this long-horizon agent architecture among the broader shifts in AI-mediated knowledge work. - [x] Claim atoms carried from deep-read analysis - [ ] Permanent candidates reviewed for promotion - [ ] Version History Diff check