How self-learning AI agents will reshape operational workflows

Google’s recent whitepaper, “Welcome to the Era of Experience,” signals a shift in the way AI agents are trained. Google’s paper hypothesizes that allowing AI agents to learn from the experience of agents rather than solely from human-generated training data will enable autonomous AI to surpass its current capabilities. After all, even when trained on data sourced by humans with deep domain-specific expertise, AI agents based on auto-regressive large language models (LLMs) can only mimic the human patterns, limiting their potential for improvement. 

If AI agents are to achieve this new level of sophistication, they must be given access to data generated from the agent’s interactions with its environment. This gives the agent a far richer and evolving pool of data to learn from to continuously improve and iterate its actions. In an enterprise context, this could bring significant benefits, accelerating manual and repetitive workflows that are ripe for “agentification” such as operations management. 

[ Related: Agentic AI – Ongoing news and insights ]

But, what does this look like in practice?

Why are AI agents well-suited to this new era of learning?

While LLMs are great at collecting and digesting operational information, AI agents that learn from experience can act and react to the outcome of an action, then learn from its aftermath. In an operations management context, this also allows the agent to try alternative remediation options based on the signals it receives.

Experience-based training for AI agents offers strong potential because it allows agents to act autonomously in real-world situations, guided by rewards that emerge from the environment. In the context of operations management, this means agents can learn from past incidents, events, customer tickets, application and infrastructure metrics and logs, as well as any other metrics made available to them.

While modern-day hype cycles demand rapid results, much of the promise of AI agents lies in how they will improve operations management over time. Given enough time and training data, the AI agent will be able to plan actions and predict their consequences in the environment—i.e., predict the reward—much better than a human. This will allow organizations to minimize human intervention in digital operations, freeing them up to focus on higher-value innovation work. 

How can experience-based learning improve operations management?

The objective of continual improvement also applies in the context of digital operations, where organizations work to improve their processes and move from a reactive mode of operations management to a preventative one. Just as AI agents can improve their processes by learning from their experiences, engineering teams can improve operations management by adopting a similar self-reflective approach.

Experience-based learning in this context requires human engineers to conduct post-incident reviews to understand an incident and establish actions to prevent that incident from recurring. However, in many cases, the learnings from a post-incident review are siloed to individual teams and not shared with the wider organization. What’s more, it is common for organizations not to conduct incident reviews for minor or well-known incidents, which further hampers the chances of teams improving their operations management processes.

Given that organizations do not consistently conduct post-incident learning reviews or share their findings across the wider organization, operations management is ripe for “agentification” powered by self-learning agents. Instead of burdening busy human engineers with post-incident reviews, AI agents can conduct these reviews and then apply this valuable experience-based training data. Coupled with an agent’s ongoing interaction with new incidents as they arise, these smart self-learning agents can power far more efficient, agent-driven operations management.

What digital operations improvements will the agentic era bring?

AI agents trained on their own experiences have the potential to revolutionize operations. There are several practical applications for agents already being seen in operations management:

Site reliability engineering (SRE): As demand for SRE skills continues to rise, AI agents can help engineers rapidly resolve issues, including diagnosing problems, surfacing historical context, and recommending or taking actions.

Operations insight: Operations teams sometimes struggle to make sense of their environment, given the number of tools they have to manage. AI agents can analyze signals across ecosystems to uncover trends and suggest improvements to existing processes.

Incident management: When incidents carry a harmful business impact, AI agents can reduce response times and human error by stepping in, even before the incident is declared, to help and proactively identify and resolve anomalies.

Together, these use cases create value through continuous improvement of the IT operations life cycle. Automating common and recurring tasks for engineers will mitigate risk and avoid drops in revenue. Above all, AI agents will increase an organization’s resilience, ensuring it can stay online with minimal human input.

While many organizations have successfully adopted LLMs, these models have limited ability to effectively manage a business’s operations. Experience can fill in the gaps. Past experiences and the information associated with them can enable AI agents to go beyond simply reproducing the work of engineers and generate significant ROI by reducing engineer toil over time. 

Allowing AI agents the freedom to manage operations, while releasing engineers to pursue truly transformational work, will be the key outcome of the era of AI operations and will transform the way many enterprises operate.

—

Generative AI Insights provides a venue for technology leaders—including vendors and other outside contributors—to explore and discuss the challenges and opportunities of generative artificial intelligence. The selection is wide-ranging, from technology deep dives to case studies to expert opinion, but also subjective, based on our judgment of which topics and treatments will best serve InfoWorld’s technically sophisticated audience. InfoWorld does not accept marketing collateral for publication and reserves the right to edit all contributed content. Contact doug_dineley@foundryco.com.