AI in 2026: The Shift from Experimentation to Autonomous Execution

Listen to this article

AI in 2026: The Shift from Experimentation to Autonomous Execution

0:00

The three shifts, in ten slides

Save it, share it, or send it to whoever sits on your board.

Swipe to preview · 10 slidesDownload the PDF

The three shifts in numbers

Adoption is accelerating; ROI is not

Zombie AI projects are those neither officially cancelled nor actively delivering value.

The boardroom accountability gap

Board-level AI oversight is measured by public disclosure, not internal practice: the actual gap may be wider still.

The era of AI experimentation is ending.

Here’s probably what’s happening: across companies worldwide, the approach that dominated 2023 and 2024 (pilot projects, proof-of-concept tests, and tech exploration) is shifting. Boards, investors, and operational leaders now want measurable outcomes, clear cost accountability, and AI systems that work autonomously alongside employees.

This shift will shape enterprise AI in 2026. Three major changes are underway, transforming how companies use and manage AI systems.

The Three Shifts Reshaping Enterprise AI

By 2026, Gartner predicts that 40% of enterprise applications will include task-specific AI agents, up from less than 5% today. Traditional AI assistants still require human guidance. AI agents are different. They work autonomously, automating development cycles, handling incident responses, running multi-step business processes, and pulling together and analysing data without needing someone to supervise every move. I’ve mapped eight practical types of these agents now showing up in production systems, and the differences between them matter more than the “agentic AI” label suggests. By 2029, Gartner expects multi-agent systems in which several AI agents work together, learning from live data and adapting to changes without requiring people to touch individual apps. This evolution is projected to generate nearly $450 billion in enterprise software revenue by 2035.

Alongside AI agents, decision intelligence (DI) is evolving from an analytics tool into essential business infrastructure. It combines predictive and prescriptive analytics, shifting the question from “what happened” to “what should we do next” in real time. DI platforms combine data visualisation, on-demand processing, security and compliance controls, and automated decision execution into cohesive systems. When decisions that used to need cross-functional meetings and extensive data gathering can now be made continuously through integrated DI systems, organisations can respond to competition far faster. This shift creates its own governance challenge: as AI systems move from supporting decisions to executing them, the question of accountability sharpens.

These questions have moved from the technology function into the boardroom. The oversight-disclosure and AI-savvy-board figures in the tables above tell a specific story: the boards paying attention are already outperforming their peers on return on equity, yet formal accountability structures remain the exception rather than the rule. The accountability imperative has reinforced demand for AI governance frameworks, even as most companies still lack one.

This accountability requirement rests on a foundation that few organisations have adequately built: data governance at scale. Autonomous agents operating across workflows require clean, auditable, lineage-tracked data. Decision intelligence systems require consistent definitions of key entities and metrics. Regulatory frameworks such as the EU AI Act require organisations to demonstrate data provenance, bias mitigation, and security controls. Organisations that invested early in data governance are moving faster into 2026. Those who treated data governance as a compliance checkbox are now hitting friction at scale, especially when an autonomous agent makes a decision based on stale or biased data.

The Cost Reckoning

Underpinning all of these shifts is a financial reality that boards can no longer ignore: AI implementations have not consistently delivered their promised returns. McKinsey’s own State of AI research finds that most companies deploying AI still fall short of their targeted productivity gains within the first 24 months. Gartner has flagged the same pattern from a cost angle: the average enterprise wastes meaningful sums annually on “zombie AI projects” that are neither officially cancelled nor actively delivering value. This gap between investment and outcome has reshaped board expectations. The central question in 2026 is no longer “Can we do this with AI?” but “Can we afford to do this at scale?”

The central question in 2026 is no longer “Can we do this with AI?” but “Can we afford to do this at scale?”

This cost discipline creates a filter. Organisations will prioritise agentic deployments and DI initiatives where the unit economics are defensible (automating administrative tasks, improving customer service response times, optimising scheduling), evaluated using concrete metrics: hours saved per employee, cycle-time reduction, cost avoidance, and incremental revenue contribution. Exploratory projects without clear business cases will face scrutiny or defunding. That filtering question, which tasks clear the bar for agentic deployment, is worth answering explicitly; I set out the criteria I use to make that call elsewhere.

The window for moving from experimentation to autonomy

Three categories of organisations will emerge. First, the equipped: organisations with mature data governance, board-level AI oversight, and cost discipline will move faster to adopt agentic systems. Their data foundations are solid, governance policies are defined and enforced, and their boards understand AI as a catalyst reshaping competitive dynamics. These organisations will extract disproportionate value from 2026 agentic deployments.

Second, the catching-up: organisations with incomplete data governance, nascent board oversight, and tactical AI strategies will face friction as they attempt to scale. They will encounter data quality issues, governance debt, and compliance challenges in production. Progress will be measurable but slower, constrained by foundational gaps they must fill as they scale.

Third, the stalled: organisations that continue to treat AI as an experimentation function, without board alignment, data governance investment, or cost discipline, will find themselves unable to compete. Their inability to deploy autonomous agents at scale will become a compounding competitive disadvantage.

What Boards Should Do

Organisations should act on six priorities: define AI posture explicitly (pioneer, transformer, functional reinventor, or pragmatic adopter); clarify governance ownership (which AI decisions belong in full-board sessions, which in committees, which are operational); codify AI governance policy with scaling rules, risk thresholds, and escalation triggers; establish concrete outcome metrics reported to the board (ROI by business unit, processes AI-enabled, resilience indicators, reskilling progress); invest in data governance as an operational necessity, not merely a compliance requirement; and build AI fluency in the boardroom through ongoing education and exposure to executives doing the work.

The move from experimentation to AI agents, decision intelligence, and embedded automation will widen the gap between companies. Those with good governance, solid data foundations, board buy-in, and cost discipline will move much faster. Those missing these pieces will face delays that slow everything down and limit the value they can capture. Organisations that haven’t already started are now working with a shrinking window before year-end. Those who act now will be first to deploy AI agents and DI systems. Those who wait will watch the gap get wider.

Those who act now will be first to deploy AI agents and DI systems. Those who wait will watch the gap get wider.