Skip to content
Views

Technology Blog – How self learning AI agents enhance cloud services

Glowing blue cloud encircled by digital rings above a neon platform, beside a panel labelled “AI Agent”.

Artificial intelligence is changing how software is built, and increasingly how SaaS platforms are run. AI-powered agents now help operate production systems, taking on monitoring, investigation and routine operational work. Building an agent that does useful work, though, is only the first step. The harder question comes later. Once an agent is live, how does it actually get better over time? Most organisations can already tell whether their agents are running. Modern platforms report the familiar operational signals: speed, errors and activity levels. These tell you that an agent executed. They do not tell you whether it made the right decisions. And without that, an agent cannot meaningfully improve.

From running correctly to deciding correctly

There is an important difference between knowing that an agent ran and knowing that it judged a situation well. The first is about reliability. The second is about quality of judgement, and it is the thing that matters most as agents take on more operational responsibility. To close that gap, Mycom extends ordinary operational monitoring into what we call behavioural observability. Alongside recording what an agent did, the platform captures why it did it: the reasoning it followed, the information it had to hand, the decision it reached and how good that decision turned out to be. That richer picture is what turns raw monitoring data into genuine feedback. And feedback is what allows an agent to improve.

A continuous improvement loop

With that evidence in place, each agent decision can be reviewed and scored, automatically at first, so that attention goes to the decisions most likely to be weak rather than to everything at once. The volume of activity is far too high to review by hand, so the platform does the first pass and surfaces the small number of cases a person should look at. Our SaaS experts then review those flagged decisions, trace each issue back to its real cause and make the fix in the right place. Good examples are captured as reference cases, so that future versions of the agent can be measured against a known right answer. Over time this forms a steady loop: observe, evaluate, improve, and protect those improvements from slipping backwards. The result is an agent that does not stand still after it is deployed. It gets better with experience, in a controlled and measurable way.

Why it matters

As AI takes on more of the day-to-day running of software, the quality of its decisions becomes a business concern, not just a technical one. An agent that simply runs is not enough. Organisations need confidence that it is making sound, well-evidenced decisions, and that those decisions keep improving. Behavioural observability provides that confidence. It preserves the knowledge behind each decision, makes quality visible, and turns everyday operation into a source of continuous improvement. Operational expertise stops living only in the heads of a few experienced engineers and becomes a capability the platform itself can build on. Importantly, people remain firmly in control. Engineers decide what good looks like, review the decisions that matter and approve the changes that improve the agent. The platform gathers the evidence and does the heavy lifting; the judgement stays with people. AI augments the team rather than replacing it.

An ever-improving cloud service

For Mycom customers, this is more than an engineering technique. It is how we continuously raise the quality of the cloud service they use. The agents that help run the Assurance Cloud Service are built by Mycom, and the self-improving flywheel is built into them. Every day of operation becomes an opportunity to get better. As our agents learn from real decisions, the service that customers rely on improves steadily beneath them, without them having to lift a finger. This is innovation delivered on the customer’s behalf. We adopt new technology, apply it to the service, and refine it continuously, so customers always benefit from an up-to-date, steadily improving platform rather than one that stands still after it is delivered. The improvement is part of the service itself.

Looking ahead

The next step is for agents to take on more of this improvement loop themselves, moving from people guiding every cycle towards agents that increasingly help refine their own behaviour, always within boundaries that people set. This is the natural next stage of the AI-driven development lifecycle. AI no longer stops adding value once software is deployed. It keeps learning, keeps improving and keeps raising the quality of the systems it helps operate. The future of AI is not only about building better software. It is about building software that can be operated, and improved, more intelligently. Through the Assurance Cloud Service (ACS) platform, Mycom is making that real for Communications Service Providers today.

*This is a summary of a technical article first posted on Medium and written by Dirk Michel, Senior VP – SaaS and Digital Technology at Mycom: “On Behavioural Observability: The Missing Layer for Continuously Improving AI Agents”.*