Survey: Agentic AI Moves from Pilot Phase to Production, Bringing Governance to the Forefront
A new report from Caylent, an AI-focused Amazon Web Services Premier Tier Services Partner, found that enterprises are already moving agentic AI beyond pilots and into production environments. At the same time, organizations are putting strict conditions around autonomy, making governance and control the next major challenge for enterprise AI adoption.
Conducted by Censuswide, the Enterprise Readiness for Agentic Engineering & Autonomous Cloud Operations Survey Report polled 200 senior enterprise leaders at organizations across the United States and Canada. All respondents reported they are are actively exploring agentic AI in engineering or cloud operations.
The research found that 59.5% of enterprise leaders are already running AI agents autonomously in production environments, moving beyond sandboxes and early experimentation.
That marks a significant shift in the enterprise AI conversation.
For the past several years, organizations have focused on whether generative AI could deliver value. Now, the focus is shifting toward a more complex question: How can businesses allow AI systems to act while maintaining security, accountability and oversight?
Enterprises Have Moved Past AI Pilots
Caylent's research suggests that agentic AI adoption has already moved into practical deployment.
Among respondents running autonomous agents, 36% said they are operating agents within defined guardrails in production environments. Another 23.5% reported that agents are broadly deployed across engineering and operations workflows.
The use cases are also expanding beyond experimentation.
Organizations are currently piloting, deploying, or evaluating AI agents for:
- Automated testing, scans, and quality gates: 67.5%.
- Automated incident response or remediation: 60.5%.
- Deploying application code to non-production environments: 48.5%.
- Proposing infrastructure or cloud configuration changes: 48%.
- Agents writing and committing code autonomously: 43%.
The findings show that enterprises are starting with workflows where the impact can be controlled before expanding toward more autonomous operations.
The New AI Bottleneck Is Trust
While adoption is accelerating, enterprises are not giving AI agents unlimited freedom.
The biggest finding from the research is that organizations are willing to embrace autonomy — but only with the right safeguards.
Caylent found that 98% of enterprise leaders would allow AI agents to execute changes in production under specific conditions autonomously. Only 2% said no level of safeguards would make autonomous production execution acceptable.
The research suggests enterprises are not waiting for smarter models. They are waiting for better control systems.
When asked what would accelerate agentic AI adoption, 83% of respondents placed stronger guardrails on equal or higher footing with model intelligence.
Those guardrails include:
- Security scanning and approval gates;
- Automated testing and quality checks;
- Explainability;
- Audit logging and traceability;
- Policy-based limits; and
- Rollback capabilities
The shift highlights a broader change in enterprise AI strategy. Organizations are no longer evaluating AI only by intelligence or accuracy. They are evaluating whether AI systems can operate safely inside complex business environments.
Security and Compliance Become AI Gatekeepers
One of the more surprising findings is where enterprise resistance is coming from.
Engineers are not the biggest obstacle to agentic AI adoption, according to the survey. Instead, respondents identified security and compliance teams as the biggest internal blockers.
Security was identified as the top blocker by 54.5% of respondents, followed by compliance and risk at 48%. Legal and procurement followed at 34.5%, while engineers were identified as a blocker by only 16%.
That changes the traditional AI adoption narrative: The challenge is not convincing developers that AI agents are useful. It is ensuring the rest of the organization has confidence that autonomous systems can operate responsibly.
The Accountability Challenge
As AI agents move closer to production systems, organizations must answer a difficult question: Who is responsible when an AI agent makes a mistake?
Caylent found that 36% of enterprise leaders identified accountability for bugs and security vulnerabilities as their top concern with AI-written code.
The preferred response model is not full autonomy. The research found that 30.5% of respondents prefer that an agent detect and alert, while a human decides whether to roll back or remediate. Another 23.5% prefer an agent to attempt remediation while allowing human override.
Only 8.5% preferred fully autonomous incident closure where the agent handles everything and humans review afterward.
That suggests the future of enterprise AI will likely be supervised autonomy rather than completely independent systems.
The Next Phase of Enterprise AI
The findings point to a broader shift in how organizations think about AI.
The next competitive advantage will not necessarily come from companies that deploy AI agents first. It will come from organizations that build the architecture required for those agents to operate safely and effectively.
Enterprises are moving from asking what AI can do to determining what AI should be allowed to do.
As Randall Hunt, Chief Technology Officer at Caylent, summarized: "The question of whether enterprises will adopt agentic AI is settled. What's left is authority, not accuracy."
For the full report, visit the Caylent site (registration required).