Under 25 Percent of Companies Have Scaled AI and What That Means for Tech Careers in 2026

Only 22 percent of organisations have successfully scaled AI across multiple business units, according to Gartner research reported by CIO Dive. Spending is going up regardless, with 85 percent of technology leaders planning to increase AI investment next year. That gap between money committed and value delivered is the single most useful thing a technology worker can understand about the current job market, because it explains both why AI hiring is uneven and why some AI roles keep disappearing after a year. At Metaintro, we read this as a question about which work survives the next budget review.

What did the Gartner survey actually find?

Gartner surveyed more than 1,300 leaders at organisations with over 50 million dollars in revenue between January and April of this year, and found that only 22 percent had successfully scaled AI across multiple business units. Scaling across business units is the meaningful test rather than running a pilot, because it is the point at which a technology stops being an experiment and starts being infrastructure. The survey also found that about 11 percent of responding organisations lacked visibility into what they spent on AI during 2025, which is a remarkable admission at this stage of an investment cycle. An organisation that cannot say what it spent cannot possibly say what it earned, and that single gap explains a great deal of the confusion in public commentary about whether AI is delivering. The honest answer for a large minority of companies is that nobody knows.

The diagnosis Gartner offers is about discipline rather than technology. Tina Nunno, a distinguished vice president and fellow at the firm, said that without disciplined measurement tied directly to business outcomes, organisations risk wasted resources and unmet expectations. She also told CIO Dive that the goal is for companies to get more out of AI rather than to risk spending more while either failing to get an appropriate return or creating additional technical risk. Not knowing how to measure a project's success, or when to shut it down, is what keeps organisations from getting there. Both halves of that matter equally. Teams are generally good at starting projects and poor at killing them, and an initiative nobody is willing to cancel absorbs budget and headcount indefinitely while producing nothing that can be pointed at.

Why is spending rising if the results are not there?

Because competitive pressure and budget cycles do not wait for evidence. The same research found 85 percent of technology leaders planning to increase AI investment next year despite the scaling problem, which tells you the spending decision is being made on strategic anxiety rather than on measured returns. This is not unique to Gartner's data. An August report from Infosys found nearly three quarters of senior business executives said they had scaled fewer than 25 percent of their AI pilots successfully, and two thirds said their organisation struggles to measure the return AI generates. Those two findings sit together uncomfortably. A majority of large organisations are simultaneously expanding a programme and unable to say whether the existing version of it worked.

Readiness is thinner still for the next wave. An August Deloitte report, covered by CIO Dive, found only one in five senior managers and C-suite executives said their organisation was prepared to redesign business processes to run autonomously with AI agents. That is worth holding next to the volume of agent announcements, and it is consistent with the pattern in which AI agent projects fail and companies quietly return to hiring graduates. The technology is ahead of the operating model, and the operating model is what employs people. Redesigning a process is slow, political and specific to each company, and no vendor can do it for you. That is why readiness lags capability by years rather than months, and why the roles that bridge the two are the ones that keep getting funded.

Which AI projects actually deliver returns?

Gartner's answer is specific enough to act on, and it contradicts where most effort is going. The most widely pursued use cases, including cybersecurity, threat detection and IT service desk automation, do not always deliver the highest returns. The three use cases Gartner identified as delivering positive returns most reliably were intelligent IT asset and cost optimisation, synthetic data generation, and automated code generation and refactoring. Those are unglamorous, internally facing and measurable, which is precisely why they work. Each one has an existing baseline to compare against, an owner who already had a budget line, and a result that shows up in a system rather than in a survey. Those three properties, not the sophistication of the model, are what separate a project that survives from one that quietly stops.

"Consumers at the end of the day do not care if you are using AI, blockchain technology, whatever buzzy buzzword that you're throwing into your marketing splash page, they don't care. They just care that your product allows them to do X faster and better," Lacey Kaelani, chief executive of Metaintro, told Bizee. That is the same finding from the customer side. Projects tied to a specific outcome someone can measure survive budget reviews. Projects tied to a category name do not, which is the mechanism behind Gartner's separate projection that a large share of AI projects will be scrapped. If you want a quick test of a project's durability, ask what it replaces and what number goes up or down as a result. If nobody can answer in one sentence, the project is exposed.

Why does the ROI gap matter for your job rather than theirs?

Because project cancellations become role cancellations, usually two quarters later. A team hired to deliver an initiative that was never measured properly is a team with no defence when the budget is reviewed, regardless of how well its members performed. That is the underlying reason tech layoffs and AI investment returns have moved together rather than in opposition, and it is why joining a well funded AI initiative is not automatically safer than joining an ordinary one. In many cases it is riskier, because a large budget attached to a vague objective attracts scrutiny the moment conditions tighten, and there is rarely a modest version of the project to fall back to.

The discipline finding gives you a way to assess this before you accept a role. Gartner reported that companies which continuously track return, treat AI as a portfolio of value, regularly assess project performance and discontinue underperforming initiatives saw positive returns on 81 percent of their AI initiatives. That is an enormous difference from the base rate, and it is a property of the organisation rather than of the technology. Working inside a company that measures is materially safer than working inside one that does not, and it is a question you can ask in an interview. There is a good list of things to establish before taking an AI role, and how the company measures value belongs near the top of it. Ask it plainly and listen for specifics. A team that can tell you which metric moved, by how much and over what period is describing a real programme. A team that answers with adoption statistics or a list of tools deployed is describing activity.

What skills does the gap actually reward?

Deployment and measurement rather than experimentation. The bottleneck Gartner describes is not model capability, it is the ability to take something that works in a pilot and make it survive contact with real processes, real data and real users across a business. That favours people who can work at the seam between the technology and the operation, which includes data engineering, integration, evaluation, and the unglamorous work of instrumenting a system so its value can be demonstrated.

It also favours domain knowledge more than the current discourse suggests. The reason so many projects stall is that nobody involved understood the business process well enough to redesign it, which is exactly the readiness gap Deloitte measured. Being the person who understands both the workflow and the tooling is a scarcer combination than being strong at either alone. AI capability is now named in a large share of technology postings, with AI skills appearing in 73 percent of tech job listings, so the differentiator is no longer whether you have used these tools. It is whether you can show a result, and results are what a hiring manager can defend to their own leadership when justifying a headcount request. That is why demonstrating AI skills concretely on a resume and in interviews matters more than listing them.

Is the AI job market shrinking or changing shape?

Changing shape, and the distinction matters if you are deciding whether to specialise. Money is still flowing in, which sustains hiring, but it is increasingly directed at initiatives that can prove value, which changes what those roles look like. The result is fewer speculative research adjacent positions and more roles attached to a specific line of business with a specific number to move. Titles are drifting accordingly, away from research framing and toward delivery framing, and the same underlying work is increasingly advertised under names that mention the business function rather than the technique. That is less exciting and considerably more durable. For anyone early in a technology career this is worth internalising before choosing a specialism, because the roles that look most like the future in a job advert are frequently the ones with the shortest funding horizon behind them.

The broader labour market context argues for durability over excitement right now. Bureau of Labor Statistics data for July 2026 put job openings at 7.3 million against 5.1 million hires, with hiring in professional and business services down 188,000 over the month, which describes a market where a bad move is slow to reverse. Meanwhile the pressure on the people making these decisions is real, with executives openly worried about their own positions if AI adoption fails. That combination, anxious buyers and unproven returns, tends to produce short lived projects, so evaluating the project is as important as evaluating the role. Ask how long the initiative has existed and how many times it has changed direction, because a programme on its third pivot in eighteen months is telling you something about how it will be treated in the next budget round.

What does this mean for your career right now?

Ask better questions before you join something. Find out how the initiative you would be working on is measured, who reviews it, and what has already been discontinued. An organisation that can name a project it shut down is demonstrating exactly the discipline Gartner associates with an 81 percent success rate. An organisation that has never cancelled anything is either extraordinarily good or is not measuring, and the base rate says which is more likely. The question also works in reverse as a signal about the culture. Organisations willing to end things publicly tend to be organisations where you can raise a problem early without it damaging you.

Then position yourself against the durable work rather than the fashionable work. Attach yourself to use cases with an owner and a number, learn enough of the business domain to be the person who can redesign a process rather than only automate a task, and keep evidence of outcomes you personally moved. The relationship between AI skills and salary rewards demonstrated delivery far more than tool familiarity, and in a market where most organisations still cannot scale what they have built, being able to prove you helped one of them do it is a genuinely scarce claim. Write it down while you still have access to the numbers, because the detail that makes the claim credible is the specific before and after, and that is exactly what becomes impossible to retrieve once you have left.

Related Articles

• What Gartner's data says about AI replacing jobs
• Companies cannot track their AI agents and the jobs that creates
• Software engineers becoming AI agent managers
• Free and low cost ways to learn AI skills
• AI skills for non technical roles
• Reskilling and the flight simulator approach to AI skills
• The coming burnout from managing AI agents
• When HR manages AI agents and which skills stay hireable

People Also Asked

Q: How many companies have actually scaled AI?

A: Only 22 percent have successfully scaled AI across multiple business units, according to a Gartner survey of more than 1,300 leaders at organisations with over 50 million dollars in revenue, conducted between January and April this year. Scaling across business units is a stricter test than running a pilot, which many more organisations have done.

Q: If results are poor, why is AI spending still rising?

A: Because the spending decision is being driven by competitive pressure rather than measured returns. The same research found 85 percent of technology leaders plan to increase AI investment next year. Separate work from Infosys found nearly three quarters of senior executives had scaled under 25 percent of their pilots, and two thirds said they struggle to measure the return at all.

Q: Which AI projects are most likely to succeed?

A: Gartner identified intelligent IT asset and cost optimisation, synthetic data generation, and automated code generation and refactoring as the use cases most reliably delivering positive returns. The most widely pursued cases, including cybersecurity, threat detection and IT service desk automation, do not always deliver the highest return. Companies that track returns and cancel weak projects reported positive returns on 81 percent of initiatives.

Future-proof your career by aiming at the work that lasts. Metaintro matches you with roles based on what you have actually delivered, not the tools listed on your profile. Create your free profile and find teams building things that stick.