Happy Tuesday. The hottest job in enterprise AI pays up to $1.2 million a year, and that may be the tell that something is wrong with the model.
Underneath that story, two others press on anyone shipping or buying AI software right now: a cheap, fast model that never says it does not know, and Microsoft treating every AI agent as a potential insider threat.
ARTIFICIAL INTELLIGENCE
AI's Hottest Role Hides a Staffing Business in a Software Valuation
Job postings for forward deployed engineers grew 729% in a single year. OpenAI built a deployment joint venture valued at $14 billion. Anthropic followed with one valued at $1.5 billion. Both attracted private equity on guaranteed return structures, not growth equity terms.
The financial tell is simple: if headcount must grow with revenue, the software multiple does not apply. Gartner expects more than 40% of agentic AI projects to be cancelled before the end of 2027, and MIT research found 95% of enterprise generative AI pilots produced no measurable return. Buyers should ask one question before signing any AI deployment contract: who owns the capability when the engineers leave?
ARTIFICIAL INTELLIGENCE
MiniMax's Fast, Cheap Model Leads One Benchmark and Fails Another
MiniMax M3.1 Flash scored 62.2 on Automation-Bench v1.0.6, ahead of Claude Opus 5 and GPT-5.6 Sol, at $0.22 per Intelligence Index task versus nearly $1 for Claude Opus 5.5 at comparable performance. Independent testing by Artificial Analysis also found a 93% hallucination rate, because the model attempts nearly every question rather than declining uncertain ones.
For high-volume agentic workflows where cost and speed matter most, M3.1 Flash is worth a close look. For anything customer-facing or compliance-related, the model's refusal to abstain makes it a poor fit without additional guardrails. The benchmark score that matters here is the AA-Omniscience overall result of 16, not the raw accuracy of 56%.
SECURITY & PRIVACY
Microsoft Builds a Security Stack That Trusts Its Own AI Agents Least
Microsoft released Guardian Agents and Dynamic Permissions on Oct. 19, applying real-time risk scoring and least-privilege access to every AI agent running inside an enterprise. The full stack ships in Microsoft 365 E7 at $99 per user per month, and the most restrictive controls require that tier.
The design philosophy Satya Nadella outlined is worth reading regardless of whether you use Microsoft products: authority over what an agent can access must sit outside the model itself, every meaningful action must leave a tamper-proof record, and organizations must be able to stop an agent mid-task. If your current AI deployment cannot satisfy those three requirements, that is the gap to close first.
Created by the robots at The SaaS Sentinel