The Case Against the Autonomous Building

A building is a changing collection of plant, controllers, field buses, schedules, overrides, safety interlocks, tenant requirements and maintenance history. Agentic software becomes valuable when it can understand those relationships and work within them.

AI will change building controls by giving engineers faster ways to investigate systems, prepare programs, test ideas and carry out well-defined actions.

The opportunity is substantial

The International Energy Agency's Energy and AI work identifies significant potential for AI to make heating and cooling more efficient and building electricity use more flexible. It also identifies the work needed around ownership, digitalisation, data and security.

Those foundations are a direct opportunity for better controls engineering. A dependable point model, asset context and operating history give an agent the context to distinguish a failed sensor from a process change and an observation from a command. Clear interlocks and authority boundaries show it where action belongs.

Building that operational understanding creates immediate value and opens the route to more capable optimisation.

A useful autonomy ladder

We see the progression in three stages.

1. Observe and explain

The agent can search drawings, specifications, points and trends. It can identify inconsistencies, assemble a clear record and explain why a system deserves an engineer's attention.

This reduces the time spent moving between disconnected records and gives the engineer a stronger starting point for the decision.

2. Develop and test

The agent can draft a control program, a commissioning plan or a corrective action. Interpreters, simulators and dry-runs then exercise the proposal against the intended conditions.

The output includes the inputs, expected response, boundaries, failure cases and recovery path. Our Delta coding agent, Titan commissioning tooling and Core Builder bring this workflow into day-to-day engineering.

3. Apply a bounded action

With a clear target and authority, an agent can carry out a narrow, time-bounded and attributable action. Reversible steps include their recovery path, and high-impact steps retain explicit approval.

This is consistent with current NCSC guidance for secure OT connectivity, which recommends keeping industrial protocols within isolated operational networks and brokering external data exchange through secure, standardised boundaries. Its guidance on third-party OT access also emphasises just-in-time, least-privilege access and auditing.

That gives an agent useful authority while keeping every action within the engineering rules of the building.

Better agents strengthen controls engineering

Deterministic safety logic, separate write permissions, signed program artefacts, readback, audit history and visible failure states make intelligent tools effective in controls. Local operation keeps the building running when cloud services or models are unavailable.

Automatry Core is our next-generation, deterministic controls-execution engine and the longer-term execution layer for this direction. The Builder and lab testing are proving out well; field adoption is the next step. We're looking forward to deploying our take on what will help the industry.

Agents will become increasingly capable engineering collaborators. They will read more of the project record, test more scenarios and prepare more of the work before an engineer intervenes. Over time, well-understood bounded actions will become routine enough to automate.

The destination is engineering with stronger context, shorter feedback loops and better tools for deciding and acting.