The Next Agent Will Not Wait for a Ticket

For years, most automation stopped at detection. A monitoring system identified a problem, generated an alert, and handed it to a human or another team for investigation. Even when AI entered the workflow, it often acted as an assistant, helping people analyze issues or recommend next steps.

Microsoft is experimenting with a different approach.

This week, the company introduced Project Perception, a security system built around three groups of specialized AI agents. Red-team agents search for vulnerabilities, blue-team agents investigate and assess risk, and green-team agents apply corrective actions to strengthen defenses. Together, they form a continuous operational loop that discovers problems, evaluates them, and responds while keeping humans in control.

Although Microsoft developed the system for cybersecurity, the underlying architecture has implications well beyond security.

Traditional automation usually handles one stage of a workflow. A monitoring system raises an alert, a person investigates the issue, another person decides what to do, and someone else implements the fix. Every handoff introduces delay, context switching, and the possibility that important work never gets completed.

Project Perception removes many of those handoffs by coordinating specialized agents that work together as part of the same operating cycle. One agent identifies a problem, another evaluates its significance, and another carries out an approved response. That pattern could eventually apply to many business functions where work follows a continuous cycle of observation, decision making, and action.

Microsoft's announcement is less about security than about system design. The company is demonstrating how multiple agents can operate as a coordinated service instead of a collection of independent assistants. Each agent has a clearly defined responsibility, but the value comes from how they work together rather than from any single model.

That architecture has broader applications. Software quality, infrastructure operations, fraud detection, compliance, customer support, and revenue operations all rely on recurring operational loops that could eventually be coordinated in a similar way. The opportunity is to shorten the time between identifying a problem and resolving it.

Many organizations still think about AI as a tool that helps employees complete discrete tasks. Closed-loop systems encourage companies to look at entire operational workflows instead. Where does information stall? Which approvals create unnecessary delays? Which recurring decisions follow consistent patterns that can be automated safely?

Organizations that understand how work flows across departments will be in a better position to build systems that improve continuously instead of solving one problem at a time.

Some engineering organizations are already moving in this direction. They are connecting monitoring systems directly to predefined actions, introducing approval checkpoints where necessary, and assigning specialized agents to different stages of the workflow rather than expecting one general-purpose agent to handle everything. Human oversight remains essential, but people increasingly supervise the system instead of performing every operational step themselves.

This issue of Redeployed is brought to you by Tecla: As AI moves from isolated tasks to continuous operational workflows, the challenge is no longer simply deploying agents. Organizations increasingly need engineers who can connect systems, orchestrate AI workflows, and build the governance and infrastructure that allow multiple agents to work together safely in production. The teams moving fastest are combining strong engineering practices with AI-native operations, bringing in talent that understands software architecture, cloud platforms, and AI systems. Tecla helps companies hire senior tech talent in the U.S. and nearshore who already work in these environments, so teams can scale AI adoption without sacrificing reliability.

Continuous automation also creates continuous responsibility. An incorrect finding can trigger an unnecessary action if permission boundaries are not carefully designed. As more specialized agents interact, debugging becomes more difficult and ownership becomes harder to define. Organizations also need to ensure that human oversight remains meaningful instead of becoming a routine approval step that no one questions.

Those operational controls become just as important as the agents themselves.

Technology leaders should start thinking beyond individual AI deployments. The next wave of AI investment will increasingly focus on how systems work together over time. That includes defining responsibilities across agents, designing escalation paths, setting approval policies, and measuring how quickly operational loops identify and resolve problems.

Organizations that develop those capabilities early will be able to automate more complex processes without losing visibility or control.

Project Perception is a security product, but it points toward a broader operating model. As AI systems become more specialized, companies will have new opportunities to connect monitoring, decision making, and execution into continuous workflows. Some of those systems will remain tightly supervised, while others will gradually earn greater autonomy as organizations become more confident in their reliability.

The next generation of AI may be defined by how effectively specialized agents work together.

Connect With Other Technology Leaders

If you want to connect with other technology leaders having real conversations about AI and how it is changing business, check out GILD Curated Circuit.

More to come…

✔️ Microsoft unveils new cyber model, agentic security tools to fight hackers — Axios

✔️ Microsoft launches AI security initiatives to combat AI-enabled threats — Infosecurity Magazine