After a McKinsey survey found 94% of corporate AI spending produced no measurable return, Microsoft and AWS are pouring billions into teams of engineers embedded inside customer companies.
Nine out of ten companies have deployed AI in at least one business function. Nearly all of them report no significant benefit from the spending. The industry's response, announced within a week of each other in July 2026 by Microsoft and AWS, has been the same: send the engineers.
The pattern marks a structural shift in how AI vendors price their products. Companies that spent the last three years selling software licenses, API tokens, and per-seat model access are now selling something closer to a consulting engagement. Microsoft's recently announced Microsoft Frontier Company, backed by a $2.5 billion investment and a 6,000-person team, exists explicitly to embed engineers inside customer organizations. AWS's parallel Forward Deployed Engineering program, funded with $1 billion, follows the same script.
The hinge that makes the timing legible is a McKinsey "State of AI" study published in late April 2026. As of the end of 2025, the consultancy reported, 89% of companies had deployed AI in at least one function, but 94% said the spending had produced no significant benefit. The vendor response is not to fix the AI. It is to wrap the AI in human implementation labor and price that wrap as the product.
"The currency that the customers are always talking about right now is speed," said Francessca Vasquez, vice president of Frontier AI Engineering and Services at AWS. The remark places the value on throughput. Speed, in this framing, is what an embedded engineering team delivers. The AI model is the substrate. The engineers are the throughput.
OpenAI and Anthropic moved the same direction this spring, dispatching their own engineering teams to client sites in partnership with major investment funds. Palantir pioneered the embedded-forward-deployed-engineer model more than a decade ago, when its commercial business was still small. The current programs from Microsoft and AWS look like a Palantir playbook scaled up by cloud balance sheets.
What changes for enterprise buyers is the procurement question. They are no longer buying access to a model. They are buying access to a team that will sit inside the company and produce measurable outputs from the model. The distinction matters for budgeting, because the engineer component is labor, not software, and labor scales differently than a license. It also reframes vendor accountability. If a deployment still produces no benefit, the explanation can no longer be that the customer did not buy enough model capacity. The humans are now part of the product, and their work is what the contract will be measured against.
The capital-markets pressure raises the stakes. Despite strong growth in its cloud business, Microsoft has disappointed markets, and its shares have shed nearly a quarter of their value since their recent peak. The data-center buildout that funded the AI push needs a visible return. Embedded engineers are the mechanism for converting compute capacity into reported customer outcomes that justify the next spending cycle.
There is a falsifier for this read. If enterprise AI deployments start producing measurable return on investment without a services overlay, the engineer armies will look transitional rather than structural. If the 94% figure moves materially over the next two McKinsey surveys, the mechanism will look like a fix that worked. If the figure does not move, the embedded-engineer programs will harden into a permanent services layer, and procurement teams will be buying labor wrapped in a software contract for the rest of the cycle. The next McKinsey "State of AI" report, expected next spring, is the trigger to watch.