Databricks recently came out with their 2026 State of AI report. The eBook can be downloaded here https://www.databricks.com/resources/ebook/state-of-ai-agents
The report is based on Data from more than 20,000 organizations worldwide, including over 60% of the Fortune 500. The survey reflects what enterprises are building and running in production.
Here's a summary of what stood out and why it matters.
AI is positively impacting customer service and experience
Of the top 15 use cases in the report, 40% are customer service and experience related, covering support, onboarding, and personalized communications. But the specifics vary quite a bit by industry. Healthcare companies are deploying agents to analyze medical literature. Manufacturing, Automotive and Energy companies are focused on predictive maintenance. Financial services firms are automating regulatory reporting. The pattern across all of them is the same: high-volume, rule-governed work that used to require significant human coordination is now being handled by agents.
Going beyond chatbots to multi-agent orchestration
A lot of enterprise use cases initially started with chatbots, but organizations are continuing to progress into multi-agent solutions. The report refers to compound AI systems, where multiple models, tools, and specialized agents work together to plan and execute complex tasks.
One of the most common architectures emerging is the Supervisor Agent model, now used in roughly 37% of enterprise deployments. A central agent takes a high-level goal, breaks it into sub-tasks, hands them off to specialized agents, and synthesizes the results. It works a lot like how a well-run team operates.
If your organization is still evaluating standalone agent or chatbot use cases, the market has already moved a generation ahead. The time to be designing for multi-agent orchestration is now.
Governance is what separates pilots from production
Companies that actively use AI governance tools deploy 12 times more AI projects into production than those that don't. Companies using structured evaluation frameworks deploy 6 times more.
The difference between an organization running a few AI pilots and one running AI at scale has less to do with model selection or technical sophistication than most people assume. It comes down to governance and evaluation discipline.
What often surprises our clients is that governance doesn't slow things down. Organizations that invest in governance infrastructure earlier in their AI journey tend to remove the friction that causes projects to stall before they ever reach production. Databricks' AI Gateway product grew sevenfold in nine months, which reflects how quickly enterprises are prioritizing this kind of control and accountability infrastructure as their agent portfolios grow.
If you're struggling to move AI initiatives from pilot to production, the first place to look is your governance posture, not your model choices.
Most enterprises are running multiple models
78% of enterprises now use two or more LLM families in the same production environment, mixing models from OpenAI, Anthropic, Meta, Google, and others. Leading engineering teams are routing work intelligently, using smaller and more cost-efficient models for routine tasks while reserving frontier models for more complex reasoning.
Beyond the cost benefits, this approach builds resilience. Organizations running multi-model architecture aren't dependent on any single vendor's roadmap or pricing decisions. Given how quickly the AI landscape is shifting, that flexibility has real strategic value.
Most organizations still haven't crossed the production threshold
Despite the acceleration in adoption, only a small fraction of enterprises have successfully deployed AI agents at meaningful scale. The blockers are rarely technical capacity. They're governance, safety, and quality assurance. As agents take on more autonomy, executing code, interacting with external systems, making consequential decisions, the need for real guardrails becomes non-negotiable.
The organizations making progress share a few things in common. They invest early in evaluation frameworks. They treat governance as a design requirement rather than something to address later. And they build on platforms that provide consistent controls across the full development lifecycle.
Deploying AI agents at scale requires the same engineering rigor as any other mission-critical system. That means building for observability, testing systematically, and establishing clear oversight protocols before agents operate independently.
What we take away from this
The Databricks report reinforces something we try to be direct with clients about. The organizations that will get the most out of AI over the next few years are the ones that execute with discipline across architecture, governance, and evaluation.
Multi-agent systems, governance infrastructure, and model-agnostic platforms are not advanced topics for later. They're the foundation. The strategic question for most organizations right now is sequencing: what to build first, where to invest, and how to move from pilots to production without losing momentum.
If you'd like to work through how these findings apply to your current AI roadmap, we're happy to have that conversation.
.svg)