ZETA AI Adoption Highlights

The ZETA Q3 2026 AI adoption highlights show a wide gap between spend and results. While 86% of AI investment flows into assistants, only 6% of executives see a return, according to Peter White at Automation Anywhere and Sven Peters at Atlassian.

Stefan Weitz at HumanX reported that 95% of AI pilots never reach production because organizations keep bolting new models onto outdated data structures and org charts. Katrin Lehmann at Mercedes-Benz said the automaker still relies on legacy mainframe systems. Using AI to refactor that code, she noted, compressed an eight-month rebuild into eight days.

What is the reality check behind the ZETA Q3 2026 AI adoption highlights?

Big budgets are going into AI, and the output often looks more like a science fair project than a production line. Weitz reported in April 2026 that global AI spending would reach $2.5 trillion that year, a 44% jump from the year before. Peters noted that only 6% of executives see a return on their AI investment. Weitz added that 84% of companies have not designed a single workflow and that over half of CEOs admit they are getting nothing from their investments.

White said 86% of AI investment today goes into copilots or assistants, and only a tiny fraction of the 88% of companies using AI see a significant revenue impact.

Why do 1,000 blooming flowers become 1,000 dead flowers?

Francis deSouza, COO of Google Cloud and President of Security Products, argued that companies that try to let 1,000 flowers bloom very often end up with 1,000 dead flowers. To get a return, he recommended picking five to seven use cases a year and driving adoption from the top down.

White said 98% of companies report employees using AI tools that IT has not sanctioned. A typical large enterprise runs over 900 applications, and 40% of them now offer agents, so tools keep appearing everywhere. Automation Anywhere built a centralized governance layer for uniform policies, monitoring, and access controls. White admitted this is an emerging space and said he would not claim the problem is solved.

How does AI help with legacy systems?

Lehmann said the cut from eight months to eight days came from testing AI on the company's legacy mainframe systems.

Justin Reock, Deputy CTO at DX, shared several more examples. Morgan Stanley uses agents to create specs from legacy code such as COBOL and Perl, saving over 280,000 hours a year by removing part of the reverse engineering step. Spotify uses an agent that gathers context and remediation steps from runbooks when an incident is detected and pulls the details into Slack for SREs. About 90% of Spotify's incidents now run through that system, which Reock said has significantly reduced mean time to recovery. Fair built an automated code review agent called Fairy that does a first pass on style and semantics, so the human reviewer can see what the agent checked and move to the deeper architectural review. deSouza added that over 50% of the code generated at Google is generated using AI.

Why does agentic AI need connected data?

Some teams focus on writing code, and others are rethinking how they manage the underlying data. Peters reported that Atlassian measured 44% better answer quality and 48% less token usage when data is already connected.

Phil Wiser, CTO at Paramount, pointed to a knowledge graph of the underlying data structure, which can be updated dynamically. He sits through daily pitches from vendors promising to restructure his data for enterprise-wide agents, and he said the real headache is figuring out where to begin.

Peters said Atlassian skips yet another text document nobody reads. An insight theme analyzer agent scans weekly customer interview pages and maps recurring themes onto a Confluence whiteboard.

How do you rebuild workflows around human behavior?

Getting the data right is half the work. The other half is getting leadership to use the tools. Lehmann recalled that Mercedes-Benz had its CEO build his own agent, and that every Friday morning the entire organization drops everything for hands-on expert training on building agents. Over 50% of her workforce already uses AI, and she expects daily usage to pass 70% within the next few months.

Alison Moore, CEO at Chief, shared that a chief product officer uses AI to analyze customer feedback while the decision about which product features to eliminate stays 100% human. She also reported that 87% of women leaders have watched things go wrong when companies prioritize AI without investing equally in their people.

Peters admitted that Atlassian set the wrong incentives at first. AI usage doubled mostly in engineering while legal and marketing stopped using it, so the company now measures super users per function.

The ZETA Q3 2026 AI adoption highlights point the same way. The payoff comes from pointing AI at unglamorous legacy bottlenecks and changing how people work, as in the eight-day rewrite at Mercedes-Benz.

To hear more from the people putting AI into production, see the HumanX 2027 program preview.