Enterprises are moving from pilots to programs with AI agents, and the money is following. New research indicates organizations will run dramatically more agents in 2026 and allocate a larger slice of IT spending to make them work at scale, even as integration, security, and data quality issues persist.
Adoption Accelerates As Use Cases Mature
The 2026 Connectivity Benchmark Report from Salesforce, MuleSoft, and Deloitte Digital finds AI agents are now used across 89% of business teams, with the average organization operating 12 agents today and planning a 67% increase to roughly 20 within two years. Digital leaders expect to devote about 19% of their IT budgets to agent-focused initiatives as projects shift from experimental pilots to production-grade workflows.
Companies are splitting their approaches: 36% are activating embedded agents inside existing enterprise platforms, 34% are building custom agents, and 30% are adopting pre-built SaaS agents. The mix reflects a pragmatic calculus—buy for speed, build for differentiation, and embed for operational reach.
Integration Is The Decider For Enterprise AI Agents
The roadblock is not enthusiasm; it is plumbing. The report shows the average enterprise manages 957 applications, but only 27% are connected. Among organizations further along in agentic transformation, app estates are even larger—averaging 1,057—yet connectivity improves only modestly to 32%.
That fragmentation creates systemic drag: 71% of IT leaders say their systems are overly interdependent, and 82% cite data integration as a major hurdle to AI adoption. Half of AI agents currently operate in silos instead of being orchestrated as multi-agent systems. While 40% are using agent-to-agent protocols and 39% are testing Model Context Protocol, 86% warn agents can add complexity without proper integration, and 96% say seamless data access is foundational to value.
Real Benefits, Real Constraints For AI Agent Programs
Where integration is strong, impact is tangible. Ninety-four percent of respondents report agents accelerate developer speed and efficiency, allowing the same share of teams to redirect effort to higher-value work. Preferences are shifting as well: 90% of developers favor AI-assisted approaches for building integrations and APIs, even though 72% still encounter friction when applying AI to the integration layer.
The employee experience uplifts are notable. Organizations that have fully adopted agents are nearly five times more likely to report significant improvement in employee experience (59%) than those still planning (12%). Business stakeholders are seeing the knock-on effects in operational visibility and customer engagement as end-to-end journeys become more connected.
Data Quality And Architecture Top The Risk Ledger
Despite the momentum, 97% of IT leaders report challenges with agentic transformation. Data quality is the single biggest obstacle, named by 25% as the top risk, and 96% admit they struggle to use data from across the business for AI. Teams spend 36% of their time building and testing custom integrations, work that rarely scales and often cements technical debt.
Architecture matters. A decisive 94% say agents will require a more API-driven foundation. Security scrutiny is also intensifying after tests on enterprise agent platforms, including those from major vendors such as Microsoft and ServiceNow, exposed exploitable behaviors—reminders that agent autonomy must be matched with strong governance, isolation, and auditability.
Budgets Shift From Models To The Middle Layer
The budget story is changing. Instead of pouring spend solely into model access, leaders are allocating funds to the connective tissue: event-driven integrations, data products, vectorized retrieval layers, and policy enforcement. Procurement teams are asking tougher questions about protocol support (A2A, MCP), observability, and fine-grained access controls. CFOs want line-of-sight to measurable outcomes—cycle-time reduction, case deflection, revenue lift—not just model benchmarks.
Best-practice playbooks are taking shape. High performers standardize on APIs and reusable connectors, implement human-in-the-loop checkpoints for sensitive actions, align to frameworks such as the NIST AI Risk Management Framework and ISO/IEC 42001, and create agent product owners accountable for end-to-end value. Organizations that stage deployments—starting with bounded tasks, then layering orchestration and policy—see faster time to trust and clearer ROI.
The 2026 Outlook For Scaling Enterprise AI Agents
With 98% of organizations committed to agents and 95% agreeing they boost developer speed, the direction of travel is unmistakable. Yet 64% worry they won’t hit their implementation goals. The gap between ambition and execution will be closed less by new models than by better integration, cleaner data, and resilient architectures.
The takeaway is straightforward: 2026 will be the year agents scale, budgets follow, and the bottleneck moves from AI capability to enterprise connectivity. Winners will treat agent deployments as data and integration programs first—and AI projects second.
