Every engineer on a capital project has lived this exact five minutes. Someone asks whether the P&ID reflects last week's change order, and what should be a ten second answer turns into opening four different systems, digging up a PDF that was emailed on a Tuesday, and calling a colleague who is somewhere over the Atlantic and not picking up. By the time anyone finds the actual current revision, two people have already built their afternoon around the wrong one. The information was never missing. It just lived in five places that had never agreed to talk to each other.

That exact fragmentation, one project's data scattered across engineering tools that don't share a common brain, is the specific problem behind a partnership announced today, and it comes from a company built almost entirely out of a much bigger industrial software vendor's old stack.

Octave Intelligence plc, the Nasdaq listed software company spun off from Hexagon AB earlier this year, announced on August 31 a technology collaboration with MAIRE, the Milan headquartered engineering group, focused on evaluating and deploying AI inside real engineering and construction workflows. Octave began trading as an independent company in late May, its Swedish depository receipts on Nasdaq Stockholm on May 25 and its shares on the Nasdaq Global Select Market in New York on May 28, carrying forward Hexagon's engineering information management, asset lifecycle, and infrastructure software into a standalone company built for the AI era. MAIRE is not a new customer being courted, it is a relationship going back more than two decades, running across Octave's Design-Build suite, including Octave Forte, Octave OnSite, Octave Loop, and the InConcert engineering information platform. MAIRE itself is a sizable operation, an engineering group delivering technology solutions and project execution across the energy, chemical, and industrial sectors, operating in roughly 50 countries with about 10,800 employees and more than 1,500 delivered projects.

What actually changes here is the layer being added on top of that existing relationship. The collaboration sits inside Octave CoLabs, a customer-led innovation program the company describes as applying agentic AI directly into a customer's own operational workflows rather than shipping a generic assistant and hoping it fits. Jay Allardyce, Octave's Chief Product Officer, framed the MAIRE deal specifically: the partnership focuses on applying AI where it delivers measurable value inside real engineering and construction workflows, not AI as a demo bolted onto existing software. MAIRE, for its part, is treating this as a humans-in-the-loop rollout, with AI meant to give engineers and project teams better information and extend what they can do rather than replace the engineering judgment that a capital project actually depends on.

The MAIRE deal is not an isolated experiment either. On Octave's August 13 earnings call, CEO Stenberg described CoLabs as building agentic workflows directly on a customer's real data, each one ending in what the company calls a validated economic benefit rather than a proof of concept that quietly dies after the pilot. Five marquee accounts are signed into the program, including two of the world's largest EPC firms, Bechtel and Fluor, with three of the five already live. That is a meaningfully different posture than a startup pitching a new estimating tool to a general contractor. Octave is starting from thirty years of engineering context already sitting inside customers' systems of record, and betting that the more defensible AI product is the one built on top of data a firm already trusts, not a fresh model asked to read a project from scratch.

None of this is a robot on a jobsite or a nine figure funding round, and an EPC software partnership will never generate the same headline as an autonomous excavator. But it is a useful data point on where agentic AI in construction is actually proving itself first: not in flashy standalone products, but inside the unglamorous, decades old software that megaprojects already run on, where the win is simply getting five disconnected systems to agree on which drawing is current. That is the same discipline BidForgeAI is chasing on the bid side, building agents against a firm's own historical proposals and pricing data rather than a generic model with no memory of how that firm actually wins work. The pitch keeps repeating across the industry because it keeps being the one that survives contact with a real project.

Sources