Research Report:The Hidden Disconnects in Enterprise AI

Enterprise AI is fragmenting into disconnected agent stacks, Jentic research finds

Only 16% report well-defined AI-use processes, despite 89% reporting coding-agent use across enterprise teams

DUBLIN, 20 August 2026: Jentic, the sovereign operating platform for AI agents, today published The Hidden Disconnects in Enterprise AI, a new report examining why organisations struggle to move AI agents from experimentation into safe, scalable production.

The research finds that enterprise AI adoption is progressing along two largely separate tracks. Coding agents are spreading across engineering and other business functions, while business agents (AI agents designed to execute business workflows) are being connected to enterprise APIs, data and systems.

Both types of agent depend on the same foundations, including identity, access control, API quality, observability, governance and testing. However, they are frequently introduced by different teams using separate processes and technology stacks. This creates duplicated controls, inconsistent access to business systems and limited central visibility, making successful experiments harder to govern, reuse and scale.

The report draws on a survey of 114 AI and API practitioners, interviews with 16 enterprise leaders and a review of 183 reports, articles and other industry sources.

Among its findings:

89% of respondents said coding agents are being used across enterprise teams, but only 16% reported well-defined processes for AI use.
51% said their organisation is already using both coding agents and business agents.
67% said their APIs are being used for business agents, yet only 30% had formally assessed those APIs for AI readiness.
Only 28% measure AI's effect on productivity or business value.
Just 23% have a scoring model for prioritising AI use cases.

"For many enterprises, access to models is no longer the main constraint," said Sean Blanchfield, Jentic co-founder and CEO, and report co-author. "The challenge is making enterprise capabilities safely discoverable, usable and reusable by agents."

"Coding agents and business agents may be introduced by different teams, but inside the enterprise they converge on the same APIs, data and business systems," Blanchfield added. "Without shared platform foundations, each successful experiment risks becoming another isolated stack. This risks adding technical debt and complexity that slow down overall deployment. Enterprise AI only scales when capabilities, context and controls get reused across agents, teams and business functions."

Six disconnects fragmenting enterprise AI

The report identifies six recurring gaps:

Coding agents outside the platform model: Adoption is progressing without shared governance and platform visibility.
Rigidity over flexibility: Governance depends on static tool lists rather than controls over access, usage and outputs.
AI-first without API-first: Agents are being connected to APIs before those APIs are ready for machine consumption.
Projects without purpose: Experimentation is progressing without consistent assessment of strategic value and measurable outcomes.
Symbolic security: Policies exist, but agent identity, access controls, monitoring and incident response remain inconsistent.
Limited sandbox leverage: Sandboxes are introduced too late, rather than being used throughout workflow development and validation.

The report also warns against "MCP-washing", where organisations use MCP to expose poor-quality APIs without addressing the underlying API design. This shifts technical debt into another layer rather than creating reliable, reusable agent capabilities.

Shared platform foundations are the route to scale

The report concludes that enterprises do not need to rebuild their API infrastructure from scratch. They need to align it around shared capabilities that can support coding agents, business agents and future agent systems.

It recommends assessing priority APIs for AI readiness, giving agents their own identities and permissions, and making agent execution observable and auditable. It also recommends simulation sandboxes that emulate enterprise APIs and data, allowing agents to rapidly develop and test workflows safely before interacting with live systems.

The objective is to move from disconnected AI projects towards repeatable enterprise capabilities that can be governed, improved and safely reused.

The Hidden Disconnects in Enterprise AI is available at https://jentic.com/report/enterprise-ai-disconnects.