Production capability rose and return did not
Domino Data Lab’s 2026 survey finds production capability rose while return stayed flat. Its cross-tabs link governed production to rules that kept pace.
Whatever is holding enterprise model programs back, it is no longer the ability to reach production. Domino Data Lab’s 2026 enterprise survey has 93% of organizations reporting improved production capability, up from 88% in 2025, while return still fails to outpace investment for 57% of them, exactly where that figure sat in 2025. Two waves of one survey, and only one of those numbers moved.
The capability line rose and the payback line did not
Both headline figures come from the same respondents in the same fielding, which is what makes the pair worth reading together rather than quoting one at a time. An organization that could not put models into live systems last year and can today has solved a genuine engineering problem. On this evidence it has not yet changed what it earns.
Regional splits do not rescue the number. 51.1% of North American respondents, 66.9% in the United Kingdom, and 67.0% in continental Europe report return at or below what they put in. Even the healthiest of the three leaves most of its respondents at or below the line they invested.
Nearly as many organizations run agents they cannot account for
43% have agent software running in production with governance in place. Close behind them, 41% are piloting (12%) or scaling (29%) without the governance to manage it. Those two groups are nearly the same size, and only one of them can say what an agent did last Tuesday, on whose authority, and against which rule.
Scaling is the load-bearing word in that second group. Piloting without rules stays a contained problem; scaling without them grows the exposure faster than any record of it.
The organizations whose rules kept pace are the ones in production
The survey’s most useful material is its own cross-tabs, which arrive in two cuts under two different labels. Merging them into a single governance claim would overstate what either one measured, so both are reproduced here as the release states them:
| Group, as the release labels it | What that cut measured | Share |
|---|---|---|
| Governance fully keeps pace | Reached governed production | 67.5% |
| Governance only partially keeps pace | Reached governed production | 17.2% |
| Governance fully integrated | Delivery speed much improved | 75% |
| Governance falling behind | Delivery speed much improved | 23% |
On the first cut, an organization whose rules kept pace is 3.9 times as likely to have reached governed agent production as one whose rules only partly did. The second cut puts a different question to a differently labeled group and comes back pointing the same way. Two questions, two groupings, one direction.
A cross-tab measures association and nothing stronger. Respondents graded their own rules and their own delivery inside one questionnaire, so an organization inclined to call its governance fully integrated is also inclined to call its delivery improved. Read narrowly, the split still says something operators can use: wherever these organizations got agents into production, the rules and the records were already standing.
September 4, 2026 amendment: McKinsey repeats the capability-return split
McKinsey fielded its survey of 1,719 respondents across 97 countries from May 4 through June 8, 2026. It repeats Domino’s split on a larger base.
80 percent report improved individual productivity, while 37 percent report a positive company EBIT impact, almost unchanged from 2025. Capability moved inside the worker’s day. Return still did not move through the company.
McKinsey reports that 20 percent say operating costs, including token costs, limit model use. It also reports that roughly one-third, 32 percent, declined at least one software purchase because coding agents supported an internal build. Cost limits use before profit appears, while coding capability already changes the purchase ledger.
McKinsey reports self-reported answers, not a measured profit result. McKinsey is neither a Muniment customer nor an endorser.
Domino sells what its survey rewards
Name the incentive plainly. Domino Data Lab sells tooling for governed model work, and the sharpest finding in its own survey is that governance is what separates production from pilots. BARC Research conducted the study independently on Domino’s behalf, which disciplines the fieldwork without changing who commissioned it or what answer would suit them.
Its base is 639 senior leaders at Director level and above, in organizations with annual revenue of $100 million or greater, surveyed in April 2026 across North America (397), the United Kingdom (148), and continental Europe (94), in financial services and insurance, life sciences, and the public sector. That is a narrow, senior, large-company sample. It says nothing about the mid-market or about sectors outside those three.
Domino’s own page for the report carries no figures at all and hands over nothing without a form. Every number above is therefore quoted from the public release rather than from the document it summarizes, which is the version a reader can actually check without trading an email address for it.
Domino Data Lab, BARC Research, and the surveyed organizations are neither Muniment customers nor endorsers. Their survey is useful public evidence because it measured the thing most vendors assert and then published the cut that complicates its own headline. The 41% piloting and scaling without governance have the harder year in front of them: the deployment is done, and the record is the part they still owe.