Ask the CEOs in a new survey who owns AI orchestration, and 42% will tell you it's them. Ask the CTOs and CIOs, and 57% will tell you it's them. About a quarter of respondents say there's no single clear owner at all, which may be the most honest answer in the whole report.
That's from The Orchestration Gap, a study of 801 C-suite executives at US companies with at least $500 million in annual revenue, produced by the agency Code and Theory with WSJ Intelligence.[1] Its argument is that companies have bought plenty of AI but haven't connected it into something that works as a whole. The findings are executives' own perceptions rather than audited performance, and the report says so. Still, the views of the people signing off on AI budgets are worth reading, and a few of these are worth reading twice.
Here are my margin notes.
183 tools, and counting
The average company in the survey runs 183 AI tools and platforms. Thirteen percent run 500 or more. That's more AI tools than most of us have browser tabs open, and some of us have a browser tab problem.
Two-thirds say they routinely discover redundant AI investments or overlapping pilots. Routinely. At that point it isn't a discovery so much as a ritual.
The report's own summary may be the best line in it: "AI seems to be creating more activity, not more progress."
The money is fine, apparently
Asked what's blocking them, executives put culture and siloed mindsets first (61%), followed by fragmented data (58%) and the lack of a central layer connecting their tools (56%). Skills (38%) and budget (29%) came last. Large companies reporting that money isn't the main problem is a sentence worth screenshotting, because it may not come around again.
The ranking matters, though. If the biggest blockers are how people work together and how information moves between them, another tool or another specialist hire won't close the gap on its own.
The best-connected system in the building
Eighty-three percent of CEOs say they're confident their company delivers a consistent brand experience. Meanwhile, 73% of all respondents say their customer-facing channels don't run on a fully unified data model and brand architecture. Different questions, to be fair. But it's hard not to notice that confidence is the one thing that appears to be fully integrated.
The new hire with no manager
Only 18% of organizations say AI agents that make decisions are held to the same accountability and reporting structures as human leaders. Among companies running 100 or more AI tools, it's 17%.
Picture hiring someone, giving them access to customer data and the authority to act, and then declining to tell them who their manager is. Most HR teams would have questions. With AI agents, it seems to be common practice. The report's fix is sensible: every agent gets a defined role, a named human accountable for its decisions, limits on what it can do, and a record of what it did and under whose authority.
What the numbers have in common
Put these findings side by side and a pattern starts to show. Hundreds of tools, and no agreement on who owns them. Pilots that keep turning out to duplicate each other. Confidence in a consistent customer experience, alongside widespread reports of disconnected channels. Agents acting without a named person answerable for them.
I don't read any of these as a shortage of software. Each one looks to me like a missing agreement: about who decides, what a tool is for, or what a good result looks like.
The idea on page 20
Two passages carry the report's most important idea. Page 6 describes work passing through "dozens of AI-assisted decisions, iterations and handoffs until the output is no longer clearly connected to objectives that started it." Page 20 says each step in a connected workflow should "inherit the customer need, the business objective, the decisions already made and the permissions to act."
That second sentence is the whole game, and it's worth noticing how much it asks for. Software can carry a business objective from one step to the next, but someone has to decide what the objective is and write it down first. The same goes for who is allowed to decide what, and what a good result looks like. The report calls the problem orchestration, which is a fair name for the connections. I'd add that a large part of it is intent.
You can't orchestrate your way to a goal nobody wrote down. An agent can't stay inside authority nobody defined. And no integration will tell you whether 183 tools are working if nobody agreed what working means. That's why the ownership numbers at the top are more than a punchline. As the report puts it, "shared responsibility without explicit decision rights quickly becomes unowned responsibility."
What to do on Monday
The report's most practical advice comes near the end: start with one workflow, connect it all the way through, and measure whether it cuts delays, reduces rework or improves the customer outcome.
I'd add one step before connecting anything. Write down what that workflow is for, who can decide what along the way, and what done looks like. If you can't, you've found your gap, and it isn't in the tech stack.
Sources and evidence notes
Code and Theory / WSJ Intelligence, The Orchestration Gap: Why AI Investment Isn't Driving Enterprise Performance. Page numbers refer to the supplied 23-page PDF (printed and PDF page numbers match). Ownership: p.10 and Figure 3. Tool counts: pp.4–5, Figure 1. Redundant investments and "more activity, not more progress": p.3. Barriers: p.7, Figure 2. CEO confidence and 73% channel figure: p.13, Figure 4. Agent governance: p.15, Figure 5; recommended agent controls: p.16. Drift passage: p.6. "Inherit the customer need…" and one-workflow advice: p.20. "Unowned responsibility": p.11. Fieldwork June 2026; reported July 2026; 801 C-suite executives (CEO, CTO/CIO, CMO roughly one-third each) at US companies with at least $500M annual revenue (p.21). The report states its findings "reflect respondents' reported perceptions and organizational practices" and "should not be interpreted as audited measures of business performance or as proof of causation." WSJ Intelligence is a unit of The Wall Street Journal Advertising Department, not the newsroom (pp.1, 21, 23). Ownership percentages come from different executive groups, not paired responses within the same companies.
