Mark Cuban Says AI's Next Job Wave Belongs to Implementers, Not Developers

Mark Cuban has been making a case about where the jobs go next, and it points somewhere unfashionable. Not model research, not prompt engineering, but the ordinary work of walking into a shoe store or a plumbing company or a regional insurance brokerage and getting AI to do something useful there.

The line that carried the argument around the internet was blunt. "Software is dead because everything's going to be customized to your unique utilization."

Two corrections are worth making before going further. The interview ran on TBPN, the Technology Brothers Podcast Network, which posted the clip in August 2025; it circulated again in February 2026. And in the fuller version of the remark, Cuban was quoting Microsoft's CEO rather than claiming the idea as his own.

What Nadella actually argued

On the BG2 podcast in December 2024, with Brad Gerstner and Bill Gurley, Satya Nadella predicted that business applications as a category would collapse in the agent era. His reasoning was structural. Strip away the interface and most enterprise SaaS amounts to what he called CRUD databases with a bunch of business logic on top. Once that logic moves up into an AI layer capable of reading and writing across multiple back ends, the application has less reason to exist.

This is a narrower claim than the headline version suggests. Nadella was describing where value accrues, not predicting that software companies vanish. IDC revisited the quote a year later and drew the same distinction, concluding that SaaS is being disrupted by evolution rather than decline. Cuban's contribution was the follow-on question. If every company's software gets shaped to its own operations, somebody has to do the shaping.

The number has moved

Cuban puts 33 million US small and mid-sized businesses on the other side of that question. The current figure runs higher. The SBA's Office of Advocacy counted 36.2 million small businesses in its 2026 update, employing 62.3 million people, or 45.9 percent of private sector workers, and producing 43.5 percent of GDP.

What those firms are doing with AI turns out to be more revealing than the headcount.

The Census Bureau's Business Trends and Outlook Survey, which samples roughly 1.2 million businesses every two weeks, found AI use holding between 17 and 20 percent from December 2025 through May 2026. The national rate in early May was 19.8 percent, or about one company in five.

The breakdown by size is where the argument lives. Firms with 250 or more employees reported roughly 37 percent adoption, and those in the 100 to 249 range came in at 32 percent. Between December and May, use climbed among firms with at least 20 employees and showed no significant change among firms below that threshold. The Federal Reserve's own tracking note put firm-level adoption near 18 percent at the end of 2025, which is broadly consistent. Adoption at the small end of the economy has flattened out entirely.

The failures are about integration

The usual explanation for that flatness is that small firms lack budget or interest. The evidence points somewhere more specific, and more useful to Cuban's case.

MIT's Project NANDA report, The GenAI Divide: State of AI in Business 2025, analyzed 300 public AI deployments along with 150 leadership interviews and 350 employee surveys. It found that 95 percent of enterprise generative AI pilots produced no measurable impact on profit and loss. The authors were explicit that neither model quality nor regulation was the culprit. They pointed instead to a learning gap: tools that could not retain context, adapt to a particular workflow, or improve from feedback, dropped into organizations that had changed nothing to accommodate them.

One figure in that report speaks almost directly to Cuban's thesis. Pilots pairing internal staff with outside expertise succeeded about 67 percent of the time. Pilots built by internal IT alone succeeded 22 percent of the time.

Hiring caught up first

Postings for forward deployed engineer, a title Palantir coined more than a decade ago for engineers who embed inside a customer's operation, rose 729 percent year over year on Indeed, from 643 in April 2025 to 5,330 in April 2026. Recruiters tracking the role told TechCrunch that the share of companies planning to hire them jumped from under 10 percent at the start of 2026 to about 70 percent by the end of the second quarter. Compensation for mid to senior roles clusters between $300,000 and $550,000.

Then the largest companies in the industry placed versions of the same bet within a few weeks of each other.

In May 2026, Anthropic launched an AI services venture with Blackstone, Hellman & Friedman and Goldman Sachs, valued around $1.5 billion and aimed at mid-sized businesses. Days later, OpenAI set up the OpenAI Deployment Company with TPG, Advent International, Bain Capital and Brookfield, reported at a $4 billion valuation. On June 30, AWS committed $1 billion to a dedicated Forward Deployed Engineering organization, staffed by thousands of engineers working in pods of five or six inside customer teams on roughly 45-day cycles. Two days after that, Microsoft announced Microsoft Frontier Company, backed by $2.5 billion and 6,000 embedded specialists. Commercial business CEO Judson Althoff rejected the FDE label outright, writing that the effort "goes beyond what has been labeled as Forward-Deployed Engineering."

Four of the most sophisticated capital allocators in technology looked at the same bottleneck and concluded it was not compute and not model capability. It was people who could sit inside somebody else's building and make the thing work.

Checking the analogy

Cuban's framing runs like this: the internet produced web agencies, the cloud produced managed service providers, and AI produces whatever this turns out to be called. The comparison survives scrutiny better than most.

Managed services is a real and substantial market. Estimates for 2026 cluster between roughly $400 billion and $490 billion globally, with the US portion around $107 billion. Penetration is deep, with surveys putting SMB use of at least one MSP somewhere between 51 and 88 percent. It is also an industry of small firms serving small firms. In one directory of 1,626 vetted providers, 87.8 percent had fewer than 50 staff.

That is roughly the shape Cuban is describing: tens of thousands of small operators, each carrying a few dozen local clients, rather than a handful of giant integrators.

Where the argument gets thin

Three objections carry weight.

The money is aimed at enterprises. Every deployment venture listed above targets the Fortune 500, with early clients including Unilever, Novo Nordisk, LSEG and the NFL. A pod of six engineers on a 45-day cycle is not something a 12-person HVAC company can buy. Nobody has solved the economics of serving the long tail, and the flat Census numbers for sub-20-employee firms may be measuring exactly that gap.

The tooling may absorb the role. AWS describes its model as agentic-first, with AI agents meant to keep running in the customer environment after the humans leave, and customers progressing from observers to autonomous operators. If deployment itself becomes automated, this service layer could compress far faster than the MSP layer ever did.

"Software is dead" is a headline rather than a finding. Software ETFs did fall sharply in early 2026, which fed the story. But Nadella's original point concerned where business logic lives, not the disappearance of software companies, and Microsoft now sells both the models and the implementation labor, which gives it obvious reasons to prefer that framing.

The version of this job that doesn't exist yet

Here is the part of Cuban's argument that gets lost in the retelling, and it is the part that actually matters.

The job he described already exists at the top of the market. It has a name, a salary band, four corporate parents and roughly $9 billion behind it. If you want to be a forward deployed engineer at a frontier lab, the path is legible and the money is extraordinary. That is a solved problem, and solved problems do not produce job waves. They produce job postings.

The role Cuban was pointing at is the other one. It is the person who can do this work for a business with eleven employees and a $4,000 annual technology budget. Nobody has built that. There is no delivery model, no pricing, no packaged methodology, no training pipeline, and, judging by the Census data, no customer yet convinced they need it. The gap between what a 45-day AWS engagement costs and what the corner business can pay is the entire opportunity, and it is currently empty.

Web agencies did not emerge because large companies wanted websites. They emerged because somebody worked out how to build one for a dentist at a price a dentist would pay. Managed services did not scale off the Fortune 500; they scaled because a two-person shop in Ohio figured out how to handle IT for forty local businesses at a margin that worked. In both cases the technology arrived years before the delivery model, and the money went to whoever solved delivery rather than whoever understood the technology best.

That is the open position in Cuban's thesis. Not the engineer who can implement AI, because those are being hired by the thousand at half a million dollars a head. The person who can implement AI for $500 a month, profitably, at volume. Whoever cracks that is not taking a job in the next wave. They are building the industry that employs it.