Three of the past month's analyst reports have used the word "agentic" more than two hundred times. In conversations with regulated industry leaders, the same word produces a complicated response: real interest, real anxiety, and a growing impatience with vendor demonstrations that don't survive a serious procurement review.
What follows is a short field-note version of a longer briefing. The pattern is consistent enough that it warrants writing down.
The pattern this month
Across regulated client conversations in April, three signals converged.
First, agentic AI vendors are quietly retreating from claims about autonomous decision-making in safety-critical contexts. The FAA's latest guidance on AI-augmented MRO scheduling now requires a documented human authorisation chain for any agent that initiates a maintenance task.
Second, the ONR has begun including AI auditability standards in early-stage SMR programme reviews.
Third, manufacturers in the digital twin space are reporting that "agentic" pilots are being repackaged as "supervised assistants" once contracts move from proof-of-concept to production.
The takeaway is not that agentic AI doesn't work. It is that the market is converging on a narrower, more defensible product than the marketing suggests.
Four questions for the boardroom
When an agentic AI initiative arrives at the boardroom, four questions reveal whether it is commissionable in your context.
- What decision is this agent authorised to make alone? If the answer is "none, but it can recommend", you are commissioning a sophisticated dashboard. That is fine, but price it accordingly.
- What is the audit trail? Regulated industries cannot operate AI systems whose outputs cannot be reconstructed and explained on demand. A procurement conversation that does not include an answer is not yet a procurement conversation.
- What happens when the agent is wrong? Rollback procedures, human escalation, and liability allocation must exist before any safety-critical deployment.
- What does scale look like? Most agentic AI today works at one site, with one team, on one task. Scaling involves data integration challenges that the proof-of-concept did not surface. Plan for the data foundations work to absorb 60-70% of the eventual budget.
The implication for your roadmap
If agentic AI is on your 2026 roadmap, the smart move is not to pull the agents forward. It is to pull the data foundation work forward.
Agents that can act in regulated contexts depend on data that is real-time, reliable, and lineage-traceable. Most regulated industries are not yet there. The 2026 budget question to bring to the board is therefore not "how do we deploy agents this year" but "how do we make our data foundation agent-ready, so when the products mature in 2027-2028 we are not eighteen months behind".
Three patterns we are watching for clients in the next quarter:
- Vendors who can articulate the audit trail and rollback path are pulling away from those who cannot. Procurement conversations are starting to bifurcate.
- Boards are quietly asking the chief architect, not the chief data officer, to own AI strategy.
- Insurance markets are starting to price AI-augmented operations differently. That is the canary that tells you the institutional view of risk has changed.
One question worth asking your team this week: if we paused all agentic AI evaluations until our data foundations are fit for purpose, what would change about how we plan the next twelve months?