Brands build agents faster than they can govern them

42% of the companies surveyed in our Publicis Sapient Guide to Next 2026 abandoned most of their AI initiatives last year (up from 17%) and only 22% of them deploy AI at scale.

The Guide cites the 42% and 17% figures from a 2025 S&P Global Market Intelligence survey of over 1,000 enterprises, reported by CIO Dive. In that survey, the average organisation scrapped 46% of its AI proofs of concept before production.

Every major tech shift follows the same adoption path — from innovators to early adopters to early and late majorities to laggards. But there’s always a gap in between: the chasm.

Geoffrey Moore named that gap in Crossing the Chasm in 1991. Early adopters buy a vision and tolerate rough edges. The early majority are pragmatists. They want proof that companies like theirs made it work safely.

It’s the moment when early excitement fades, pilots pile up, and the early majority still isn’t convinced it’s safe to scale.

That’s exactly where we are with AI agents.

§Agent debt

We’re witnessing agent debt already forming — as brands rush to build agents faster than they can govern them.

We’re witnessing agent debt already forming — as brands rush to build agents faster than they can govern them.

The term borrows from technical debt, which Ward Cunningham coined in 1992. Quick code is a loan, and the interest grows while it stays unpaid. The Guide defines agent debt as the liability created when agents proliferate and are abandoned faster than enterprises can govern, trace or align them.

The market adds pressure. In June 2025, Gartner predicted that over 40% of agentic AI projects will be cancelled by the end of 2027. It also warned about “agent washing”, vendors rebranding chatbots and old automation as agents. By its estimate, only about 130 of thousands of vendors offer real agentic features.

Four fundamentals for avoiding agent debt:

• Standardize your data or standardize your mistakes.
• Orchestrate first, automate second.
• One registry. One reality.
• If you can’t explain it, you can’t trust it.

In practice, a registry is the inventory companies already keep for employees, servers and suppliers. For each agent, it records what it does, which data it touches, who owns it and when it was last reviewed. The data rule is just as concrete. Ten agents reading ten versions of one customer record give ten confident, inconsistent answers.

Act now: Establish a central agent registry, Appoint orchestration owners, Enforce governance reviews before agent debt compounds beyond repair.

Our Guide to Next 2026 explores how leading organizations are crossing this chasm — and what comes after.