Agent Mode AI — the podcast
The audio companion to this site. Two analysts pick one claim from the Holding-up ledger per episode and walk through the evidence: the primary sources, the named speakers, the verifiable numbers, and the current verdict. Holding, Partial, or Not holding. Fifteen to twenty minutes, every Sunday.
- Episode 13 · 12:25 · 2026-08-02
Why IT operations is the highest-exposure agentic-AI workforce population
The IT operations job-family is structurally distinctive in agentic AI exposure terms. AM-012 walks the US Bureau of Labor Statistics Occupational Outlook role-by-role projections through 2033, the World Economic Forum Future of Jobs Report 2025 bimodal displacement-and-creation lists, and Dario Amodei's twenty-eighth of May 2025 Axios projection that AI could eliminate half of entry-level white-collar jobs over one to five years. The procurement-deck choice for the chief information officer is whether the workforce-transition posture is agent-orchestration or agent-replacement.
- Episode 12 · 10:23 · 2026-07-26
What the Anthropic Claude for Chrome disclosure tells procurement
Anthropic shipped Claude for Chrome on the twenty-sixth of August 2025 to one thousand Max-plan subscribers at one hundred to two hundred dollars per month, with a published security disclosure: twenty-three point six percent prompt-injection success pre-mitigation, eleven point two percent post-mitigation, zero percent on URL-injection variants after subsequent patches. AM-009 walks why the disclosure cadence is the procurement signal rather than the rate.
- Episode 11 · 10:50 · 2026-07-19
The seven AI vendor exit clauses that decide whether you can leave
Switching AI vendors in 2026 is a contracts problem before it is a tech problem. AM-145 walks the seven exit-clause families most enterprise master service agreements miss, the three forces that turn the 2026 renewal cycle into a board-level question, and what an EU AI Act Article 16 deployer needs in writing before signature.
- Episode 10 · 10:00 · 2026-07-12
What vendor "successful pilot" references do not tell procurement
McKinsey State of AI 2025 measures twenty-three percent of enterprises scaling agentic AI, thirty-nine percent experimenting, thirty-eight percent with nothing in production or stopped. AM-140 walks the gap between vendor pilot references and procuring-enterprise scaled production, the documented Klarna, Salesforce Agentforce, and GitHub Copilot walk-backs that describe what those references typically obscure, and the six pre-pilot questions a procurement committee can require answered in writing before the contract closes.
- Episode 9 · 12:14 · 2026-07-05
The three questions every CIO should ask about a vendor accuracy claim
Vendor "ready-to-run" agentic AI positioning is not procurement evidence unless three questions are answered: accuracy on which task, against which baseline, measured how. AM-146 walks the 2026 procurement-grade reference shapes — CRMArena-Pro, CMU TheAgentCompany, WebArena, SWE-bench Verified, and the Anthropic Claude for Chrome disclosure pattern.
- Episode 8 · 13:47 · 2026-06-28
Which LLM provider actually stays up?
Twenty-four months of operational record across the five major foundation-model providers. AM-136 documents Anthropic, OpenAI, Google, AWS Bedrock, and Azure OpenAI. Every provider had at least one multi-hour outage that exceeded its own SLA-credit threshold. Single-provider dependency is the 2026 procurement risk; multi-provider routing is the mitigation; LiteLLM, OpenRouter, Portkey are the gateway-abstraction reference implementations.
Claims: AM-136 - Episode 7 · 12:45 · 2026-06-21
Why 88% of agentic AI deployments fail
Correction (10 Jun 2026): the 12/88 statistic central to this episode was retracted; it does not appear in the cited Stanford source. The audio stays up unchanged as the record. Original summary: four datasets converge on the same bimodal shape across enterprise agentic AI. Stanford Digital Economy Lab's twelve-eighty-eight split. McKinsey State of AI 2025's twenty-three percent scaling cohort. MIT NANDA's ninety-five percent pilot-failure finding. The 67-versus-22 build-versus-buy spread.
- Episode 6 · 15:34 · 2026-06-14
What changes for enterprise AI on 2 August 2026
The 2 August 2026 EU AI Act deployer-obligations enforcement window is ninety days from this episode's publication date. Four claims interlock for the procurement reader: AM-127 the deadline-anchored prediction, AM-135 the Article 50 transparency UX, AM-138 the post-enforcement MSA red-team additions, and AM-046 the Article 12 audit substrate that ties the other three together. What changes operationally, what changes contractually, and what the first thirty days of enforcement will probably reveal.
- Episode 5 · 08:57 · 2026-06-07
What the money is doing
Three institutional capital allocators reading the same AI risk surface and behaving differently. AM-118: pension funds and sovereign wealth funds publishing nothing. AM-119: reinsurance market actively repricing AI tail risk. AM-121: SaaS incumbents celebrating with 25% RPO growth and AI-native cohort revenue concentration. The asymmetry is the editorial point.
- Episode 4 · 10:04 · 2026-05-31
What AI doesn't know about your business
Operators edition. Three places where SMB AI tools produce visible work that fails because the tool doesn't know about the regulatory or algorithmic layer it's operating in. OPS-037: AI-drafted invoices fail EU VAT audits on cross-border treatment. OPS-039: AI-drafted contracts miss notarisation requirements in EU jurisdictions. OPS-041: AI-generated marketing content hits Google + LinkedIn + Etsy enforcement. The pattern is the same; the fix takes 30 seconds.
- Episode 3 · 07:42 · 2026-05-24
Whose consent do you need to deploy AI?
The same labour-relations surface, two scales. AM-120: works council consent for AI deployment in EU multinationals under German BetrVG §87, Dutch WOR Article 27, French CSE provisions. OPS-038: collective-agreement triggers at four-employee SMB scale via avv-declaration and Allgemeinverbindlich-declaration. Abby and Avery walk both — the bridge episode between enterprise and operators.
- Episode 2 · 09:43 · 2026-05-17
What you are actually buying when you buy AI
Two procurement surfaces that did not exist on any CIO checklist eighteen months ago. AM-117: AI Bill of Materials becoming a procurement requirement under EU AI Act Article 11 and CycloneDX ML-BOM. AM-121: ServiceNow Now Assist's auditability and lock-in axis — the same Bardoliwalla pitch that justifies platform-incumbent AI also names the five-year commitment risk. Abby and Avery walk both.
- Episode 1 · 12:37 · 2026-05-10
The CIO's new liability
Two insurance markets are repricing AI risk on the same enterprise. D&O insurance is rewriting renewal language to flag AI supervision under the Caremark fiduciary doctrine; the reinsurance market is repricing AI tail risk into 2026 cyber treaties. Abby and Matt walk through both signals and what they say together about how underwriters now see AI risk.