In the first ten days of July, three enterprise software vendors shipped products solving variations of the same problem: nobody can see what their AI agents are actually doing.
On July 8, Quiq launched Verified Intelligence, a three-part control layer of guardrails, pre-launch simulations and step-by-step decision visibility for customer-service agents. The same day, HTEC introduced OneLoopAi, a platform pitched at giving enterprises real-time visibility into AI usage, cost and delivery impact. Also on July 8, Abrigo announced APX, an agentic lending platform for banks due for general availability in the third quarter, with human oversight and institution-specific guardrails in the launch messaging.
Three companies, three industries, one shared premise: agents are already in production, and the accountability layer is being retrofitted around them.
The adoption curve nobody disputes
The demand signal is not in question. Gartner projects that 40% of enterprise applications will feature integrated task-specific AI agents by the end of 2026, up from less than 5% in 2025 — an eightfold jump the firm compares to the speed of public cloud adoption.
"AI agents will evolve rapidly, progressing from task and application specific agents to agentic ecosystems," said Anushree Verma, a senior director analyst at Gartner. "This shift will transform enterprise applications from tools supporting individual productivity into platforms enabling seamless autonomous collaboration and dynamic workflow orchestration."
That is the optimistic reading, and a fair one. Task-specific agents are narrow, auditable in principle, and attached to workflows that already have owners. Abrigo estimates agentic AI can cut manual labor in lending operations by more than 40%, the kind of number that ends procurement debates.
The control layer that didn't ship with it
The gap is in what surrounds the agent. Okta's AI Agents at Work 2026 survey of 292 executives and 492 knowledge workers found 90% of executives confident in their organization's visibility into AI tools — while 52% of knowledge workers admitted using unsanctioned AI tools at work. In the same study, 58% of executives named governance and oversight as their top AI-agent security concern, yet only 32% said they secure AI agents with the same rigor they apply to human employees.
Other 2026 surveys point the same way, with figures varying enough to be read as directional. The finding is structural, not statistical: agents are being granted production credentials faster than anyone is logging what they do with them.
Thoughtworks framed the underlying tension in its Technology Radar as the problem of permission-hungry agents — the agents worth building are precisely the ones that need access to private data, external communications and real systems. Volume 34, published in April 2026, treats zero-trust architectures, sandboxed execution and defense in depth as table stakes rather than best practice.
The gap between capability and accountability
The vendors shipping control layers describe the problem in almost identical language, which is itself evidence the gap is real rather than manufactured.
"The brands that get AI right are the ones that never had to choose between innovation and control," said Mike Myer, CEO of Quiq. "Verified Intelligence is how we make sure our customers never have to make that tradeoff. You get agentic AI that acts, and a control layer that makes sure it acts correctly."
HTEC's pitch is the same argument from the delivery side. "Organizations are investing heavily in AI, but many still struggle to understand the value it delivers as they go; this includes adoption and costs," said Darko Todorovic, CTO at HTEC.
Both statements are commercial. Both also describe a real asymmetry. Agent capability is sold and demoed at the level of the individual task; accountability only matters at the level of the system: who authorized this action, against which policy, with what audit trail, and who answers when it goes wrong. Capability scales with the vendor's roadmap. Accountability scales with the customer's governance maturity, which is slower and has no launch date.
The move-fast position deserves its due. Governance-first programs have a real failure mode: long policy exercises that produce committees and no deployed systems while competitors compound operational learning. Gartner's own guidance pushes CIOs to define agent strategy in months, not years. And a narrow agent inside a single workflow, with a human approving the consequential step, is a defensible posture — arguably more so than the shadow tools half of knowledge workers report using anyway.
The counterargument is about compounding. Ungoverned agents do not stay narrow: they accumulate integrations, inherit credentials, and get chained to other agents. Retrofitting an audit trail onto a system already in production is far harder than building one in.
Regulation is not filling the gap on schedule
Anyone hoping regulators would force the issue this year got a mixed answer. On August 2, 2026, the EU AI Act's Article 50 transparency obligations become enforceable, covering chatbot disclosure and synthetic content marking, and the European AI Office gains full penalty powers over general-purpose AI model providers.
But the Digital Omnibus on AI, agreed in provisional form on May 7 and since in force, defers the high-risk obligations that would bite hardest on agentic systems: standalone Annex III systems move to December 2, 2027, and AI embedded in regulated products under Annex I to August 2, 2028. Member states have been slow to name competent authorities and harmonized standards are unfinished. For the next eighteen months, enterprise agent governance stays a voluntary discipline enforced by internal risk functions and vendor tooling, not by law.
What to watch
Three things next quarter. First, whether Abrigo's APX ships to general availability in Q3 with its guardrails intact; agentic lending in a regulated industry is the sharpest test case available. Second, whether control-layer products get bought as standalone line items or absorbed into platform pricing, which reveals whether enterprises value governance enough to pay for it. Third, the first well-documented agent incident with a named accountable party. The surveys suggest most organizations do not have one.
“The brands that get AI right are the ones that never had to choose between innovation and control. You get agentic AI that acts, and a control layer that makes sure it acts correctly.”— Mike Myer, CEO, Quiq