TL;DR
This cycle resolved two open questions and sharpened the picture on both ends of the buying journey. At the top of the funnel, Outsell's study landed (48% of buyers start discovery with AI, 94% use AI before reaching sales), and the frontier moved from AI answers to agent-assembled shortlists. At the governance end, GSA issued the first binding federal AI procurement clause, effective October 19, 2026 — while a wave of reality-check reporting warned that the "autonomous buyer" everyone is building for is still years away.
The public-sector floor just got binding
Federal AI procurement moved from guidance to enforceable contract text, and the scope is narrower than feared — but the obligations are real.
"This is the first binding US federal acquisition rule specifically governing contractor handling of government data inside LLM systems."
— eyeon.ai regulatory brief
GSA issued final clause 552.239-7001 on September 28, 2026, effective October 19 — and the final rule walked back several of the June draft's overreaches: the "material feature" plus direct-government-data two-part scope test replaces the broader "processed by an LLM" formulation, and NIST AI RMF lifecycle tasks replace the rigid four supply-chain roles [gsa-american-ai-clause-gsar-552-239-7001]. The retained operational duties are the ones that bite: no training on government data for other customers, 72-hour incident reporting, closeout deletion of embeddings and fine-tuned weights, and decommissioning liability capped at 25% of the affected task order [gsa-american-ai-clause-gsar-552-239-7001]. This is the contractual endpoint of the private-sector pattern we flagged last cycle, where 90% of legal teams write AI clauses into guidelines — data-handling proof and auditability are now contract text, not RFP preferences.
What to watch: Whether early solicitations apply the material-feature test cleanly, and whether major model providers offer government-ready terms — or cede federal AI deals to vendors with dedicated government offerings.
Discovery is machine-mediated, full stop
The Outsell study resolves last cycle's open question, and the numbers confirm the funnel's top is no longer human-initiated search.
"48% now start vendor discovery with AI, and another 44% use it alongside..."
— Outsell press release via LinkedIn (Rob Keenan)
The study — published August 4, 2026, by Rob Keenan, based on 226 buyers controlling real spend — found 94% of B2B buyers use AI before they ever reach a sales team, and 60% have turned away from search entirely [outsell-2026-b2b-buyers-study]. Caveats apply: the panel is media/events-industry buyers and the full report is paywalled, so treat percentages as directional — though the direction matches IDC's larger-n finding from last cycle that vendor-website evaluation usage fell from 66% to 37% [outsell-2026-b2b-buyers-study]. The practical consequence: absence from AI answers is the new invisibility, and the discovery surface is something you engineer for, not something you buy ads against.
What to watch: Whether Outsell releases crosstabs breaking the paywall, and whether the media-vertical numbers hold in horizontal buyer panels.
Agentic shortlisting makes your documentation a revenue line
The frontier has moved past being citable in AI answers to being parseable by agents that assemble the consideration set themselves.
"When an AI agent builds the shortlist, your brand's visibility is determined by the quality of your technical data architecture, not your creative."
— MarketScale
The MarketScale analysis describes buyers delegating research to agents that "scan indexed repositories, eliminate irrelevant options, and surface a direct comparison" — and warns that specs living behind a form fill or in a non-schema-tagged PDF mean an agent "will either miss it or downrank your brand in favor of a competitor whose documentation is accessible" [aeo-axo-frameworks-2026]. The market is productizing fast: Contentstack launched Canoe on October 1, an AI-visibility tool that runs buyer questions across ChatGPT, Gemini, Perplexity, and Claude — free at $0, Growth at $279/month — with CEO Neha Sampat's framing: "Your next customer meets an agent before they meet you, and most brands have no idea what that agent is saying" [aeo-axo-frameworks-2026]. The AEO playbook from earlier coverage is now necessary but insufficient; ungated specs, pricing, and API docs are the endgame.
What to watch: Whether AI-answer monitoring tools like Canoe become a sales-engineering metric — and whether imitators compress the $279 price point to zero.
The robot buyer isn't here — governability is the scored differentiator
The strongest counterweight yet to the agentic-buying hype landed this cycle, and it reframes what buyers actually evaluate.
"The reality right now is we're simply working with agent-assisted shopping. There's no true agentic autonomy going on."
— Lindsay Walker, Hedera AI Studio — The Register
The Register's October 1 piece documents the gap between Gartner's $15T B2B projection and a present where OpenAI shut down ChatGPT's "Buy It" after fewer than 30 of Shopify's millions of merchants went live, and Amazon blocked Meta's shopping bot [agentic-procurement-commerce-shift]. CMI's parallel piece lands the B2B-specific point: "an agent cannot get a CFO to consensus" — and per McKinsey's 2026 B2B Pulse, the #1 switch driver is inconsistent information across the vendor's teams, a message problem, not a checkout problem [agentic-procurement-commerce-shift]. Meanwhile, the RFP floor is scoring what agents can be trusted on: audit completeness, policy encoding depth, and cost predictability now appear in evaluation criteria that "would not have existed in the same form two years ago," with agentic workflows consuming 5–30x more tokens per task and AI cost underestimation reported at 500–1,000% in some deployments [futurum-agent-control-plane-framework-2026]. The synthesis: build for agent readability now, but don't rebuild GTM around a buyer that is roughly five years from functional.
What to watch: Whether the FTC's August comment period on AI-driven personalized pricing hardens into enforcement that constrains agent-mediated price discovery.
What surprised us
- The government wrote the sanest AI clause. The final rule's "reasonable-efforts" accuracy duty replaced the June draft's prescriptive "Unbiased AI Principles," and the scope test deliberately excludes incidental LLM use — a narrower, more workable rule than nearly anyone expected from the 77-comment draft [gsa-american-ai-clause-gsar-552-239-7001].
- The reseller trap nobody's pricing. GSA's clause overrides conflicting commercial agreements — integrators must promise advance model-change notice and deletion their upstream model providers haven't contractually accepted [gsa-american-ai-clause-gsar-552-239-7001]. Channel partners are exposed before the first solicitation lands.
- The buyer-side trust math is brutal. Only 6% of marketers consistently accept their ad platform's AI recommendations, yet the agentic-commerce vision assumes buyers will let an agent shortlist six-figure software [agentic-procurement-commerce-shift].
- "Agentic" is inflating toward zero information on both sides of the market — The Register's consumer-side agentic-washing and the Entrepreneur RFP analysis describe the same failure mode: labels everywhere, evidence nowhere [futurum-agent-control-plane-framework-2026].
Open threads worth a vote
- [watch] GSA 552.239-7001 implementation: how solicitations apply the two-part scope test — steer which early-solicitation signals we track before the October 19 effective date.
B2B Buyer Criteria Shift for AI — Delta Briefing
Since last time
- Promoted
- GSA's binding federal AI clause — last cycle it was a one-line "private-sector mirror" reference; it's now the lead section, with the final rule text (552.239-7001) analyzed in detail.
- The Outsell 2026 B2B Buyers Study — last cycle an open thread asking someone to locate the primary publication; it landed, and it's now a core section.
- Escalated
- Agent-assembled shortlists — last cycle a single G2 callback about chatbots building the shortlist; now a full section with a productized tool (Contentstack's Canoe) and a documentation-as-revenue-line argument.
- The agentic-buying reality check — last cycle's buy-side agent narrative (Vertice's Ana launch) is now counterweighted by hard evidence that autonomous buying is years out, reframing what buyers actually score.
- Demoted
- IDC's vendor-funnel collapse data (66%→37% website decline, 68% AI comparison) — was a full section; now survives only as a corroborating caveat inside the Outsell section and a context line in the GSA section. Still worth knowing; don't reread.
- The 90% legal-AI-clause stat — was a surprise bullet and procurement-section anchor; now a single "we flagged this last cycle" callback in the GSA section.
- Disappeared
- Carbon's "satisfaction is no longer a moat" section — the 94%-satisfied/59%-reconsidering renewal-auction story, its switch-driver stats, and the related surprise bullets (boring switch drivers, 14% no-AI buyers) are entirely absent from this cycle.
- Vertice's Ana — the buy-side negotiation agent, its 4,000-negotiation track record, and the Ana example-wins surprise bullet are gone. The new reality-check section tempers the agentic-buying narrative but never mentions Ana.
The public-sector floor just got binding (Promoted)
Federal AI procurement moved from guidance to enforceable contract text, and the scope is narrower than feared — but the obligations are real.
"This is the first binding US federal acquisition rule specifically governing contractor handling of government data inside LLM systems."
— eyeon.ai regulatory brief
GSA issued final clause 552.239-7001 on September 28, 2026, effective October 19 — and the final rule walked back several of the June draft's overreaches: the "material feature" plus direct-government-data two-part scope test replaces the broader "processed by an LLM" formulation, and NIST AI RMF lifecycle tasks replace the rigid four supply-chain roles [gsa-american-ai-clause-gsar-552-239-7001]. The retained operational duties are the ones that bite: no training on government data for other customers, 72-hour incident reporting, closeout deletion of embeddings and fine-tuned weights, and decommissioning liability capped at 25% of the affected task order [gsa-american-ai-clause-gsar-552-239-7001]. This is the contractual endpoint of the private-sector pattern we flagged last cycle, where 90% of legal teams write AI clauses into guidelines — data-handling proof and auditability are now contract text, not RFP preferences.
What to watch: Whether early solicitations apply the material-feature test cleanly, and whether major model providers offer government-ready terms — or cede federal AI deals to vendors with dedicated government offerings.
Discovery is machine-mediated, full stop (Promoted)
The Outsell study resolves last cycle's open question, and the numbers confirm the funnel's top is no longer human-initiated search.
"48% now start vendor discovery with AI, and another 44% use it alongside..."
— Outsell press release via LinkedIn (Rob Keenan)
The study — published August 4, 2026, by Rob Keenan, based on 226 buyers controlling real spend — found 94% of B2B buyers use AI before they ever reach a sales team, and 60% have turned away from search entirely [outsell-2026-b2b-buyers-study]. Caveats apply: the panel is media/events-industry buyers and the full report is paywalled, so treat percentages as directional — though the direction matches IDC's larger-n finding from last cycle that vendor-website evaluation usage fell from 66% to 37% [outsell-2026-b2b-buyers-study]. The practical consequence: absence from AI answers is the new invisibility, and the discovery surface is something you engineer for, not something you buy ads against.
What to watch: Whether Outsell releases crosstabs breaking the paywall, and whether the media-vertical numbers hold in horizontal buyer panels.
Agentic shortlisting makes your documentation a revenue line (Escalated)
Last cycle this was a footnote about chatbots assembling shortlists; now there's a named failure mode and a shipped product category.
"When an AI agent builds the shortlist, your brand's visibility is determined by the quality of your technical data architecture, not your creative."
— MarketScale
The MarketScale analysis describes buyers delegating research to agents that "scan indexed repositories, eliminate irrelevant options, and surface a direct comparison" — and warns that specs living behind a form fill or in a non-schema-tagged PDF mean an agent "will either miss it or downrank your brand in favor of a competitor whose documentation is accessible" [aeo-axo-frameworks-2026]. The market is productizing fast: Contentstack launched Canoe on October 1, an AI-visibility tool that runs buyer questions across ChatGPT, Gemini, Perplexity, and Claude — free at $0, Growth at $279/month — with CEO Neha Sampat's framing: "Your next customer meets an agent before they meet you, and most brands have no idea what that agent is saying" [aeo-axo-frameworks-2026]. The AEO playbook from earlier coverage is now necessary but insufficient; ungated specs, pricing, and API docs are the endgame.
What to watch: Whether AI-answer monitoring tools like Canoe become a sales-engineering metric — and whether imitators compress the $279 price point to zero.
The robot buyer isn't here — governability is the scored differentiator (Escalated)
This is the new counterweight to the agentic-buying hype we carried last cycle, and it reframes what buyers actually evaluate.
"The reality right now is we're simply working with agent-assisted shopping. There's no true agentic autonomy going on."
— Lindsay Walker, Hedera AI Studio — The Register
The Register's October 1 piece documents the gap between Gartner's $15T B2B projection and a present where OpenAI shut down ChatGPT's "Buy It" after fewer than 30 of Shopify's millions of merchants went live, and Amazon blocked Meta's shopping bot [agentic-procurement-commerce-shift]. CMI's parallel piece lands the B2B-specific point: "an agent cannot get a CFO to consensus" — and per McKinsey's 2026 B2B Pulse, the #1 switch driver is inconsistent information across the vendor's teams, a message problem, not a checkout problem [agentic-procurement-commerce-shift]. Meanwhile, the RFP floor is scoring what agents can be trusted on: audit completeness, policy encoding depth, and cost predictability now appear in evaluation criteria that "would not have existed in the same form two years ago," with agentic workflows consuming 5–30x more tokens per task and AI cost underestimation reported at 500–1,000% in some deployments [futurum-agent-control-plane-framework-2026]. The synthesis: build for agent readability now, but don't rebuild GTM around a buyer that is roughly five years from functional.
What to watch: Whether the FTC's August comment period on AI-driven personalized pricing hardens into enforcement that constrains agent-mediated price discovery.
What surprised us
- [NEW] The government wrote the sanest AI clause. The final rule's "reasonable-efforts" accuracy duty replaced the June draft's prescriptive "Unbiased AI Principles," and the scope test deliberately excludes incidental LLM use — a narrower, more workable rule than nearly anyone expected from the 77-comment draft [gsa-american-ai-clause-gsar-552-239-7001].
- [NEW] The reseller trap nobody's pricing. GSA's clause overrides conflicting commercial agreements — integrators must promise advance model-change notice and deletion their upstream model providers haven't contractually accepted [gsa-american-ai-clause-gsar-552-239-7001]. Channel partners are exposed before the first solicitation lands.
- [NEW] The buyer-side trust math is brutal. Only 6% of marketers consistently accept their ad platform's AI recommendations, yet the agentic-commerce vision assumes buyers will let an agent shortlist six-figure software [agentic-procurement-commerce-shift].
- [NEW] "Agentic" is inflating toward zero information on both sides of the market — The Register's consumer-side agentic-washing and the Entrepreneur RFP analysis describe the same failure mode: labels everywhere, evidence nowhere [futurum-agent-control-plane-framework-2026].
Open threads
- Closed: The Outsell study location question from last cycle is resolved — the study landed and is now the "Discovery is machine-mediated" section.
- Dropped: The IDC annual data-refresh watch item is not carried forward this cycle; the IDC stats it covered survive only as supporting context. Re-propose it if you want it tracked.
- New this cycle: [watch] GSA 552.239-7001 implementation — how solicitations apply the two-part scope test, before the October 19 effective date.