Our quarterly briefing for the people who run operations, not research labs. Three shifts worth your attention this quarter, and a concrete move for each.
Adoption: the pilot era is ending
The pattern across our client conversations this quarter is consolidation. Organizations that ran five or ten scattered AI pilots over the past two years are shutting most of them down and funding the one or two that produced measurable results. Boards have stopped asking "what are we doing with AI" and started asking "what did the AI spending return" — a much better question, and a harder one to answer without baselines.
The second-order effect: budget is moving from experimentation lines to operating lines. AI systems are being treated like systems — with owners, SLAs, and renewal decisions — rather than like initiatives.
What to do: audit your pilot portfolio against one question per pilot — what number moved, and can you prove it? Pilots that can't answer after two quarters should be closed deliberately, not left to drift. The ones that can answer deserve an operations plan and a real owner.
Procurement: contracts are catching up to reality
Buyers got burned by the first wave of AI contracts — feature lists, demo-based selection, no performance language — and their procurement teams have adjusted. We're now regularly seeing RFPs that ask for written accuracy thresholds, measurement methodologies, model-change notification clauses, and exit provisions covering data and prompt portability.
Vendors are splitting into two camps: those who will sign numeric acceptance criteria and those who redirect the conversation back to the demo. The gap between the camps is the single most useful signal in an evaluation. We wrote up the mechanics in Buy acceptance criteria, not demos.
What to do: before the next renewal or purchase, write the acceptance table first and send it with the RFP. You'll lose some bidders. That's the feature.
Governance: documentation expectations are hardening
Regulators and enterprise customers are converging on the same baseline expectation, even where the specific rules differ by jurisdiction: know what AI systems you run, what data they touch, who is accountable for each, and how you'd detect and correct a harmful output. Inventory, accountability, monitoring, correction — the framework is stable even as the statutes move.
The practical shift this quarter is that these questions are arriving through customer security questionnaires and insurance renewals, not just from regulators. If your organization sells to enterprises, your buyers' diligence teams are already asking.
What to do: build the AI system inventory now, while it's short. One page per system: what it does, what data it sees, who owns it, how it's monitored, how a human overrides it. Every governance regime we've seen proposed starts from those five facts, and assembling them under deadline is miserable.
The through-line
All three shifts point the same direction: enterprise AI is being absorbed into normal operational discipline — budgets, contracts, audits. The organizations handling this well aren't the ones with the most pilots. They're the ones treating AI systems the way they treat any other system that touches revenue or risk: measured, owned, and documented.
That's the discipline our Scope → Build → Operate model is built around, across every solution we deliver.
We publish this briefing quarterly. For the version specific to federal contractors, see Federal AI guidance: what changed for contractors. Questions about what this quarter means for your operation — book a consult.