IBM urges real-time orchestration to manage autonomous AI agents
Companies must move beyond static policies to active, real-time controls to manage autonomous AI systems that make independent decisions across enterprise workflows.
The short version
- Only 18 percent of surveyed organizations maintain a complete and current inventory of their AI assets.
- Just 27 percent of organizations know exactly where AI is being used within their systems.
- 71 percent of organizations with high AI governance maturity have specific guidelines for acceptable AI use.
- 81 percent of executives expect a major AI vendor outage to significantly disrupt their operations.
Why it matters
As businesses shift from simple AI tools to autonomous agents that chain actions together, static rules are insufficient. Readers in leadership or operations need to understand that governance must be embedded into the technical execution to prevent unmonitored risks.
Moving from rules to referees
Philippine businesses are reaching a turning point where simple AI applications are evolving into autonomous agents. These systems can make independent decisions, chain multiple actions together, and adapt their behavior with very little human intervention. Because they operate at speeds no human team can manually monitor, traditional governance policies are no longer enough to ensure safety.
Lope Doromal, a technical community leader at IBM Philippines, argues that organizations must combine governance with orchestration. He describes governance as the rulebook while orchestration acts as the referee who actually enforces those rules during the game. Without active orchestration, control remains separate from execution, leaving companies to hope that teams will interpret and follow policies correctly.
When AI systems are not coordinated under pressure, the benefits of scaling technology can quickly turn into significant risks. Organizations may find themselves fully compliant on paper but completely fragmented in practice. To succeed, companies must place controls inside the way AI systems run instead of keeping them as standalone documents.
The challenge of invisible AI
A major hurdle for enterprise management is the lack of visibility into where AI is actually operating. An IBM Institute for Business Value study found that only 27 percent of surveyed organizations know where AI is being used within their company. This lack of awareness makes it nearly impossible for leaders to perform independent verification of their systems.
To solve this, Doromal recommends establishing an agent registry to transform an opaque environment into a manageable inventory. This registry should be a continuously updated record that identifies every AI agent, its specific purpose, its owner, and its data access permissions. Because new agents enter corporate systems constantly, the registry must be updated through continuous discovery rather than periodic audits.
Beyond just knowing what exists, enterprises need observability systems to monitor what those agents are actually doing. These systems should examine decision paths, identify failures, and maintain audit trails for investigating incidents. For example, IBM Concert Observe can provide task-level visibility into large language model calls and the sequence of actions leading to a specific outcome.
Identity and access management
Every AI agent operating within a business should have a formal identity and access management framework. This framework defines who owns the agent, what activities it is authorized to perform, and which enterprise systems it can access. Doromal suggests that permissions should be limited to the smallest amount required for the task, with the ability to revoke them immediately if behavior becomes unexpected.
Security risks also extend to application programming interfaces, or APIs, which allow agents to connect with external databases and services. If these interfaces are poorly governed, they can expose an organization to unauthorized interactions. Establishing these boundaries is critical because one small setup error can create a risk for the whole company when agents move across many systems.
Accountability must be established before any incidents occur by building a governance architecture in advance. Organizations should identify specific teams responsible for handling incidents and establish pre-authorized shutdown mechanisms. This ensures that there is a clear plan for who has the authority to pause or retire an agent the moment it violates its approved boundaries.
Sovereignty and vendor risk
Businesses must also navigate their dependencies on external AI vendors, models, and infrastructure. A study cited by Doromal found that only 9 percent of global executives fully understand their dependencies across these different layers. Furthermore, 81 percent of executives expect a major AI vendor outage to significantly disrupt their operations, yet 71 percent say switching vendors would be difficult.
To mitigate these risks, IBM suggests pursuing selective AI sovereignty based on business risk and strategic value. This means applying different levels of control over data and models depending on how critical the specific task is to the company. Maintaining full control across the entire AI stack is often neither practical nor cost-effective for most organizations.
IBM offers tools like watsonx Orchestrate to manage agents regardless of where they were developed or what technology they use. For those seeking even more control, IBM Sovereign Core provides a customer-operated control plane built on open standards. These options help companies maintain operational control while still utilizing the power of diverse AI models and infrastructure.
A practical path forward
For businesses with limited technical resources, the best starting point is improving data management. Poor data management is the most commonly cited barrier to effective AI governance because every AI model and agent is built upon the data they use. Organizations should prioritize clean, documented data and incorporate governance into the process from the very beginning of development.
Rather than trying to govern every AI system at once, Doromal suggests selecting one high-impact use case to test the governance process. By walking a single workflow from start to finish, a company can identify weaknesses and delays without needing a massive technical team. This approach allows governance to become a part of daily operations rather than a periodic review on a calendar.
The urgency of this shift is clear in recent studies showing that many Philippine CEOs are comfortable using AI-generated insights for strategy. However, many are investing in these technologies before fully understanding their value because they fear falling behind. By focusing on skills transfer and open standards, local organizations can build governance maturity through daily decisions that mitigate risk and accelerate value.
Try it
Identify one high-impact AI use case in your workflow and map out every decision point, data access requirement, and potential failure point to see where your current policies lack real-time enforcement.
Watch
How many organizations will successfully transition from static policies to real-time orchestration over the next year.
The impact of open standards on reducing vendor lock-in for autonomous AI agents.
Whether CEO-level accountability will become the standard for AI governance across all industries.
What to watch
- How many organizations will successfully transition from static policies to real-time orchestration over the next year.
- The impact of open standards on reducing vendor lock-in for autonomous AI agents.
- Whether CEO-level accountability will become the standard for AI governance across all industries.
Try it
Identify one high-impact AI use case in your workflow and map out every decision point, data access requirement, and potential failure point to see where your current policies lack real-time enforcement.
Sources and links
Read the original at The Manila Times on MSN
First reported by The Manila Times on MSN. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.