Accenture and Google Cloud have announced the creation of a joint entity dedicated to deploying enterprise agents at scale, with a thousand engineers stationed directly at client sites and sector-specific accelerators. This is not a product announcement: it is a workforce announcement.
Two weeks ago we reported that a survey found almost no productivity gains after three years of AI in the enterprise. Here is the industry's response to that finding, and it is consistent.
The real obstacle was never technical
This is the thread we have been pulling for weeks. The capabilities exist, the models are converging, prices are falling. What is missing lies elsewhere.
We quantified it with the gap between large and small companies: the share of large organisations deploying agents at scale rose from 27% to 40%, while smaller ones stagnated at 22%.
The reason is not that smaller organisations have worse models — they have exactly the same ones. It is that a deployment presupposes data governance, verifiable permissions, a security team, a reorganisation of processes. The model is accessible; everything else is not.
A thousand engineers stationed at client sites is precisely the answer to that problem. You are no longer selling a technology, you are selling the ability to integrate it.
The consulting model, with its virtues and its limits
What works. A company that has neither the skills nor the time to carry out an internal transformation can rent them. This has always been the consulting trade, applied to a new subject.
What sticks. Expertise brought in from outside leaves with those who brought it. This is the classic problem with that model: you get a system that works, without acquiring the ability to make it evolve.
This echoes what we wrote in our decision framework for building versus buying: the decisive question is not "who does it" but "who will maintain it in eighteen months".
This kind of deal is a good indicator of maturity, and it confirms a shift. When a technology is young, the value lies in capability. When it becomes commonplace, the value moves to integration. We saw this with agent harnesses, with the standardisation of protocols, with the acquisition of a distribution platform. A thousand engineers in the field is the human version of the same movement.
What it means for everyone else
Two practical consequences, depending on the size of your organisation.
If you are a large organisation, this kind of offering now exists and considerably shortens the deployment timeline. The question to ask upfront is that of skills transfer: what will you be able to do on your own once they have left?
If you are a modest organisation, this market is not aimed at you, and the gap is likely to widen. The right strategy remains the one we recommend: start small, on a specific use case, with the smallest model that suffices, and build the expertise in-house rather than buying a transformation.
What to take away
This is one of those announcements that people skip and that describe the real state of the sector better than a model release.
It says that the bottleneck has shifted from the technology to the people capable of installing it. And it suggests an answer to the puzzle of missing productivity gains: they are not appearing because the reorganisation that would produce them has not yet happened, and it costs a great deal in human labour.
This is consistent with the precedent of IT, where the gains arrived a decade after the machines. The difference this time is that someone is explicitly selling the acceleration of that delay.