The numbers look impressive on the surface. Gartner forecasts worldwide AI spending will total $2.59 trillion in 2026, a 47% jump from the year before. Agentic AI alone added roughly $70 billion to that revised forecast between January and May. By any measure, business investment in artificial intelligence is accelerating.

But there is a growing gap between what companies are spending and what they are actually getting back.

According to Gartner’s own survey data, fewer than one in three corporate decision-makers could identify specific financial outcomes attributable to their AI investments. A separate analysis found that 18% of companies that deployed AI have since abandoned or rolled back those initiatives due to poor quality and lack of meaningful adoption. And Gartner is projecting that more than 40% of agentic AI projects will be cancelled outright before the end of 2027.

This is not a story about AI being overhyped. It is a story about how companies have been deploying it.

The Efficiency Shift Is Already Underway

The clearest signal that something has changed is how major enterprise buyers are starting to talk about AI costs. Uber’s COO told analysts in May 2026 that the link between AI spend and clear business benefit was “not yet there” and the company capped per-employee AI spend as a result. Lindy, an AI automation startup, switched its entire operation off Anthropic’s models to a cheaper alternative specifically to control token costs. Microsoft reportedly cancelled some AI coding licenses across a division for cost certainty.

These are not fringe cases. CNBC and VentureBeat both reported in late June that a broader shift is happening across the enterprise market, from “tokenmaxxing” (maximising AI usage regardless of cost) to deliberate, outcome-focused deployments. Companies are no longer asking “how much AI are we using?” They are asking “what is this actually doing for us?”

This is what the 2026 inflection looks like in practice: the spending continues to grow, but the accountability standard has changed.

What the Failures Have in Common

When researchers look at why AI deployments fail to produce ROI, the answer is almost never the AI model itself.

The CambrianEdge.ai report that tracked rollback rates found that poor collaboration infrastructure and lack of workflow redesign were the primary drivers. Companies bought AI tools and dropped them into existing processes without changing how work gets done. Staff who were never trained on how to use AI effectively were handed prompts and expected to deliver productivity gains.

Gartner’s research points to a similar pattern. The firms that cannot attribute financial outcomes to their AI investments are typically the ones that approached AI as a software purchase rather than an operational change. They licensed tools without designing the workflows those tools would improve.

The 5% of enterprises that are seeing measurable returns tend to focus on specific, measurable workflows. They define what “success” looks like before deploying. They train their teams. And they measure outputs against the baseline they defined at the start.

What This Means for Business

If your company has spent money on AI tools in the last two years and you cannot point to a specific number that has improved, you are in good company. But “everyone else has the same problem” is not a strategy.

There are three things that separate the companies seeing returns from those that are not.

They pick specific workflows, not categories. Saying “we are going to use AI for customer service” is not a plan. Saying “we are going to reduce the time our team spends answering invoice status questions from 12 hours a week to under two hours, using an AI agent trained on our billing data” is a plan. The specificity is what makes measurement possible.

They invest in training alongside tools. AI agents are only as useful as the humans directing and reviewing them. Companies that have built data literacy across their teams, even basic data literacy, see consistently higher returns from AI investments than those that have not. When your team understands what AI can and cannot do, they stop wasting prompts and start building processes.

They use purpose-built solutions over general-purpose ones. General-purpose AI tools are versatile. Purpose-built AI agents designed for specific tasks in your industry are effective. The difference matters when you are trying to prove a business case.

Where This Leaves the Market

The AI spending number will keep growing. $2.59 trillion this year, more next year. But the mix is shifting. Infrastructure and hyperscaler spending dominates today, accounting for over 45% of the total. Enterprise software spend is where the gap between investment and return is most visible and where pressure is mounting.

What changes in the next 12 to 18 months is not the availability of AI. It is the accountability for results. Companies that have not yet designed AI into their workflows with clear outcome measures are going to face harder internal questions when the next renewal cycle comes around.

The inflection point is not about spending less. It is about spending on things that have a measurable destination.

---

If you’re deciding where to start with agents, start here. The free Working With Claude field guide walks through the ecosystem, Claude Code, and a real rollout plan. Get your copy.