Mid-2026: what clients are actually asking for

Halfway through 2026, a pattern is clear in our inbound work: the questions have shifted from adoption to governance. Few clients are asking whether to use cloud, Kubernetes, or AI tooling; those debates are settled in most of the organizations we talk to. Nearly all are asking how to control what they already have. The shift is visible in who initiates the conversation too. Two years ago it was usually a CTO with an ambition; this year it is just as often a CFO with a bill or a security lead with a questionnaire.

The three engagements filling our calendar

First, cost and capacity governance for AI workloads that grew faster than their budgets. The experimental AI spending of the past two years produced features customers now depend on, attached to cost curves nobody owns. The asks are concrete: unit economics per feature, daily spend visibility, decision rules for when a smaller model suffices, and an honest accounting of which experiments should graduate or end. We wrote about this pattern in March, and demand has only firmed since.

Second, consolidation of platform sprawl accumulated during the 2023-2024 build-everything years. Multiple clusters where one platform would do, three observability stacks, duplicate CI systems inherited from teams that no longer exist as drawn. The engagements look like archaeology followed by subtraction: inventory what is actually load-bearing, then retire or merge the rest without breaking the products on top. Subtraction is harder than addition, organizationally, because every redundant system has one defender, and part of the work is giving leadership a defensible basis for the call.

Third, audit readiness, as security questionnaires from enterprise customers grow sharper. The generic spreadsheet of five years ago has become pointed: evidence of access reviews, proof of encryption key management, accessibility conformance, and increasingly specific questions about where AI features send data. Clients who built real controls answer in days. Clients who built narratives are finding that reviewers now ask for the configuration behind the narrative, and the gap between the two has become a sales-cycle problem rather than a hypothetical one.

What this says about the market

We read this as a healthy correction rather than a retreat. The experimental phase of the past few years produced real capability and real mess, in roughly equal measure, and organizations are now sorting one from the other. Notably, governance budgets are holding up even where new-initiative budgets have tightened; control over existing systems is being treated as foundation rather than overhead, which is the correct accounting in our view.

For technology leaders, the practical implication is that the skills in demand this year are editorial: deciding what stays, what goes, and what gets a budget and an owner. That is less glamorous than adoption and more valuable at this point in the cycle.

The next few years of our work, we expect, will be about keeping the capability and retiring the mess. If your environment contains both in the usual proportions, the sorting goes faster with people who have now done it across a few dozen organizations, and that is, at mid-2026, most of what we do.