AI agents set to reshape observability as businesses rethink their digital future
The next phase of enterprise technology management is unlikely to be defined by bigger dashboards or more complex monitoring tools. Rather, it will be shaped by artificial intelligence agents that change the way engineers understand and respond to the increasingly complicated systems underpinning modern businesses.
For years, observability has relied on a familiar process: engineers open dashboards, search through logs and gradually piece together what has gone wrong. But as AI models become more capable, that workflow is beginning to shift.
The emerging model is a human directing an AI agent, which then searches and analyses the underlying data. Rather than spending hours collecting information, engineers will increasingly receive investigations that have already been prepared.
This shift promises major productivity gains, but it also presents a challenge for the technology industry. AI systems need to be built that understand not just data, but the unique operational context of every organisation.
From dashboards to intelligent investigations
AI agents are already capable of querying telemetry and identifying potential causes of system failures. As these capabilities improve, observability platforms will increasingly need to support investigations performed on behalf of engineers rather than simply provide tools for engineers to use manually.
This change will significantly alter the demands placed on technology infrastructure: a human investigator is naturally limited by time and attention. They may only compare a small number of time periods, review selected traces and explore a handful of likely explanations. An AI agent, however, can examine far more possibilities simultaneously, comparing large volumes of historical information and testing multiple hypotheses before presenting a conclusion.
This creates a new set of expectations. Systems will need to deliver faster query responses, retain more historical information and provide higher-quality data. If important telemetry has been discarded or heavily sampled, an AI agent may not have the evidence required to make an accurate assessment.
Unlike experienced engineers, AI systems cannot rely on years of institutional knowledge or intuition when information is missing. Their reasoning is limited by the quality and completeness of the data available to them.
The challenge of building a universal SRE agent
Many technology companies are pursuing the idea of a universal site reliability engineering (SRE) agent. This is a single intelligent assistant capable of handling most operational investigations. The concept is attractive: instead of engineers learning complex dashboards or writing specialised queries, they could simply ask questions in natural language and receive useful answers.
However, debugging complex production systems involves more than analysing technical signals as the hardest part of incident investigation is often context: it’s important to understand how a specific business operates and how teams make decisions under pressure.
That knowledge is rarely stored in one place. It exists across internal documentation, runbooks, incident reports, deployment systems, team conversations and the experience of engineers who have operated those systems for years.
The rise of specialised AI agents
The future of observability may look less like one universal AI assistant and more like an ecosystem of specialised agents designed around specific organisations and industries. Different agents could focus on different areas of business operations, from infrastructure and databases to security, payments and customer experience. These systems could be trained and configured around an organisation’s own processes and operational history.
This approach would recognise an important reality: two companies running similar technology stacks may respond to the same technical problem in completely different ways because their priorities and experiences are different.
The most successful observability platforms may not be those that force organisations into a single model of working. Instead, they will provide flexible foundations that allow businesses to create and manage their own specialised agents.
An open foundation for AI agents
Supporting this ecosystem will require a more open approach to observability. Openness in this context means more than using open-source software: organisations will need the freedom to choose the models, tools, interfaces and agent frameworks that best suit their own systems and working practices.
They should also be able to decide where their data is stored, how agents access production environments and how their behaviour is governed and monitored. An agent built for a payments team may require a very different combination of data, permissions and workflows from one designed for infrastructure or security.
This flexibility will become increasingly important as AI technology continues to evolve. Businesses are unlikely to rely permanently on a single model, agent framework or vendor interface — they will need the ability to replace individual components without rebuilding the entire observability environment around them.
The platforms that succeed will therefore be those that act as an open foundation for many different agents, rather than attempting to control every layer of the experience.
Building a shared memory
As organisations create more specialised agents, another challenge will emerge: collaboration. Investigations may begin in many different places — one engineer might use an integrated development environment, another a notebook, another an internal chat system or a custom incident workflow. Without a shared record, valuable knowledge risks becoming trapped in temporary conversations that cannot easily be reviewed or reused.
The future of AI-powered observability will require durable investigation records. These will be places where teams can see what questions were asked, what evidence was gathered, which possibilities were considered, and why a particular conclusion was reached.
Humans remain responsible for decisions
Despite the rapid progress of AI, the role of humans remains central. Agents can investigate systems faster, analyse more information and identify patterns that may be difficult for individuals to detect, but they still lack the business judgment required to balance risks and decide the appropriate response.
The most practical model today is collaboration rather than full automation. AI agents gather evidence and accelerate investigations, while humans provide direction and make final decisions.
Over time, more autonomous systems may emerge, but the immediate future of observability is likely to be defined by a partnership between people and intelligent tools.
Stop treating AI like a senior exec: it's just a brilliant intern
Organisations need to remember that AI is just a software tool, and not meant to solve every...
From stone age to open source, community is key
The tools of our time may be digital, but the forces that move us forward remain human.
Visibility and governance will define the next phase of AI investment
Return on AI investment is rapidly becoming one of the most significant pressure points facing...
