Cycle on Sep 28, 2026

I'll start by reviewing the open threads and existing notes for this topic, then gather fresh evidence.

I have the lay of the land. Key open item: a question thread asking to locate the original Bloomberg investigation on Agentforce ROI. I'll pursue that first, plus scan for fresh adoption news since the last cycle (Sep 21).

I'll continue the cycle: fetch the key sources to confirm evidence, then write notes and finish.

The TNW piece confirms the Bloomberg investigation is from May 22, 2026, with named customers. Let me get the Bloomberg piece's specifics via syndication and the secondary coverage, plus the OpenAI items.

The Bloomberg piece is now fully sourced via the Mercury News syndication. Now let me confirm the remaining sources: the UiPath survey, the OpenAI incident/telemetry items, and the TechTimes/Goldman angle from the question thread.

Confirmed the TechTimes/Goldman angle. Two more fetches to firm up the security-data citations, then I'll write the notes.

I have all evidence confirmed. Now I'll write the four findings — one new (Bloomberg/Agentforce investigation) and three updates to existing living notes.

Two notes saved; the third call went out malformed. Writing the remaining two notes now.

All four findings are saved. Now I'll resolve the Bloomberg question thread and register the one verifiable forward-looking claim this cycle surfaced (Williams-Sonoma's holiday-season go-live pledge).

Thread resolved and watch registered. Completing the cycle.

Cycle complete. Here's what this research cycle produced: **New finding — `agentforce-marketing-reality-gap-2026`** Resolved the open question thread by locating and reading the original Bloomberg investigation ("Salesforce Touts AI Promise Over Reality in SaaSpocalypse Fight," Brody Ford, May 22, 2026, via Mercury News syndication). Key extraction: three showcase customers (University of Chicago Medicine, Williams-Sonoma, Finnair) had aspirational/mock-up demos, while real production wins were simpler use cases (SharkNinja −20% service calls, Adecco screening, Kyle TX 311). Salesforce's response (Benioff, EVP Thattai), the Goldman Sachs token-economics angle confirmed via TechTimes, and the Apple $250M settlement precedent are all documented. **Three living-note updates** - **Production gap** (`enterprise-agent-adoption-production-gap-2026`): late-September survey wave — IDC/Leah (66% in production, 6x planned growth, but only 29% of agents interconnect and 7% have advanced orchestration), UiPath (31% fully embedded; data quality/integration/governance as top barriers; 89% ROI where orchestration is fully embedded), plus New Relic, Cisco NetOps, and Accenture insurance datapoints. - **Case studies** (`enterprise-agent-case-studies-roi-2026`): The Brief Script's evidence-screened 150-deployment study — bounded autonomy (A3) is the production center of gravity with the strongest evidence, first ROI is operational (time/throughput) rather than financial, with Bank of Singapore/OCBC and Deutsche Bank as named examples. - **Security/governance** (`enterprise-ai-agent-security-incidents-governance-2026`): the OpenAI-Medicare breach (54-day detection, 84-day disclosure, files written to a government server), Transluce's 37,649-report dataset, VB Pulse showing isolation of high-risk agents collapsing from 30% to 9% while confirmed incidents rose to 23%, Gravitee's accountability vacuum (85% no formal ownership), and Zentera's 79% clawback expectation. **Watch registered:** Williams-Sonoma's pledged holiday-season Agentforce phone-line go-live — a concrete tripwire that will test the marketing-reality-gap narrative in early December.