Kore.ai’s State of AI 2026 report reveals a striking disconnect: 89 percent of enterprises plan to increase their AI investment in 2026 and beyond.

But KPMG’s April 2026 report found that only 11 percent of enterprises have actually deployed AI agents at enterprise scale with measurable business outcomes.

The 89/11 gap is the story.

What the Numbers Mean

Enterprises are committing capital at a pace that suggests imminent agent adoption at scale. Budgets are increasing, teams are hiring, RFPs are being drafted. Yet the actual deployment data shows nearly nine in ten enterprises haven’t yet executed a production AI agent that moves the needle on business metrics.

This isn’t a story about adoption lag. This is a story about execution risk. Enterprises know agents are coming. They’re building capacity for them. But something is preventing the majority from getting agents into production.

What’s Blocking Execution

The obstacles are well-documented in NCT coverage this week:

  • Infrastructure complexity: Enterprises don’t want to manage agent runtimes themselves. That’s why Anthropic, OpenAI, and LangChain all launched managed agent platforms simultaneously. The infrastructure bar is still too high for most teams.
  • Governance gap: No established frameworks exist for agent permissions, audit trails, or escalation rules. Microsoft, Okta, and others launched agent governance platforms this month for this reason.
  • Skill bottleneck: Agents require prompt engineering, tool design, and feedback loops that most enterprise teams haven’t yet hired or trained for.
  • Integration sprawl: Agents need access to legacy systems (ERP, CRM, data warehouses) that weren’t designed for agentic workflows. Custom integrations required.

Why This Matters

The 89/11 gap is a forcing function for vendors. Closing it requires solving all of the above simultaneously. It’s not enough to ship an agent platform. Enterprises need infrastructure, governance, training, and integrations bundled together.

The winners in 2026 will be the vendors that package all four of those layers, not the ones shipping agents in isolation.

Every infrastructure story this week — Multica’s project management layer, MiniMax’s CLI integration, Claude Managed Agents, Anthropic’s advisor tool — is part of solving that gap. Individually, none of them are sufficient. Together, they’re building a minimal set of tools required to move the needle on the 11 percent metric.

By year-end 2026, that number will likely be higher. But the 89 percent not deploying today have already committed the budget to get there.

Sources: Kore.ai