Perspectives on technology, talent, and the AI shift.

2026: the year enterprise AI stopped being optional

The global AI consulting market is projected to grow from roughly $30 billion in 2026 to nearly $350 billion by 2034 — a compound annual growth rate above 35%. That kind of curve doesn't happen because companies are curious. It happens because the cost of waiting became visible.

For the past two years, the dominant posture toward generative AI inside most enterprises was caution: pilot a use case, measure it carefully, wait for the technology — and the regulatory picture — to settle. That posture has flipped. The share of leaders who describe themselves as cautious about AI deployment fell from roughly half in 2025 to under a quarter in 2026, while the share who have deployed generative AI tools or are actively working toward it has grown to over 40%.

What changed isn't the technology alone — it's the competitive math. 2026 is increasingly seen as the pivot year in which enterprises begin scaling AI agents across customer service, operations, and back-office processes, which is precisely why demand for strategic AI consulting has shifted from "should we" conversations to governance, safety, and operating-model design.

For leadership teams, the practical implication is that AI strategy has stopped being a side project owned by an innovation team. It now sits alongside cloud, data, and talent as a core input to the operating plan — and it needs the same rigor applied to it that any other capital allocation decision would get.

The skills-based hiring shift is changing who gets hired

Credential-based screening — the degree, the previous employer's logo, the years-of-experience cutoff — is losing ground to direct competency assessment. Companies that made the switch report access to talent pools up to 19 times larger, and first-year retention improvements of roughly a third.

The shift is being driven by necessity as much as philosophy. Job postings tied to AI-related skills jumped well over 100% between 2024 and 2025, and specialized roles in AI, cloud architecture, and data engineering are outpacing the supply of candidates who hold the traditional credentials employers used to filter for. Firms that keep screening on pedigree alone are competing for a shrinking slice of a fast-growing demand curve.

• Skills-based assessment widens the qualified candidate pool without lowering the bar — it changes what the bar measures.
• Modern screening tools also carry built-in bias-detection, helping surface exclusionary language in job descriptions before it costs you candidates.
• Roughly seven in ten job seekers now treat workplace flexibility as a dealbreaker, which changes how competitive a role needs to look on paper before a strong candidate will even engage.

For hiring teams, the takeaway isn't to abandon rigor — it's to point that rigor at what a candidate can actually do, which turns out to be a much better predictor of performance than where they went to school.

What "human-in-the-loop" really means for delivery teams

As more of our clients scale agentic workflows, one design question keeps coming up: not whether to keep a human in the loop, but where. That answer determines whether automation actually speeds you up.

The instinct in a lot of early AI rollouts was to treat human review as a blanket safety net — check everything, everywhere. That approach protects against mistakes, but it also erases most of the speed gain the automation was supposed to deliver. The more durable pattern we see working is placing review checkpoints specifically at the points where a decision is ambiguous, high-stakes, or hard to reverse, and letting well-defined, low-risk steps run without a human gate.

Done well, this looks less like a safety net and more like an escalation path: clear audit trails, defined boundaries for what a system is allowed to decide on its own, and a fast route to a qualified person when something falls outside those boundaries. Done poorly, it looks like a rubber stamp — review that exists on paper but doesn't actually catch the failure modes it was designed for.

Getting this right is a design problem before it's a technology problem. It requires actually mapping where judgment matters in a given workflow, not defaulting to "more oversight is always safer." For leadership teams evaluating vendors or internal proposals, that's the question worth asking directly: where, specifically, does a person see this before it ships — and why there?

Cloud service models, explained

IaaS, PaaS, and SaaS get thrown around interchangeably in vendor conversations, but the differences determine how much of the stack your team is actually responsible for — and that's the decision that matters.

Infrastructure as a Service (IaaS) hands you the raw compute, storage, and networking, and leaves everything above that — operating system, runtime, applications — to your team. It's the most flexible model and the one that demands the most in-house capability to run well.

Platform as a Service (PaaS) takes the infrastructure and runtime off your plate, so your team can focus on the application itself rather than the environment it runs in. It's a common middle ground for teams that want speed without giving up control over how the application is built.

Software as a Service (SaaS) hands over the entire stack — you're a user of the finished application, not an operator of any layer beneath it. It's the fastest path to value and the least flexible.

The right model isn't about which is "best" — it's about matching the level of control your team actually needs against the operational burden you're willing to carry. Most mature technology estates end up running a mix of all three, which is exactly why the migration and modernization decisions deserve real strategy, not a default answer.