Engineering Firms Are Helping Build the AI Now. Here's What It Means
Aurecon, Arcadis and MAIRE spent September investing in and co-developing AI agents that work from their own project data, a sign the advantage may be shifting from software vendors to whoever owns the project history.
In September three of the world's larger engineering groups, Aurecon, Arcadis and MAIRE, announced plans to put AI agents into how they deliver projects, working from their own documents and data, and in two cases taking a stake in the AI company itself. In the same month Accenture launched a capital-projects business with more than 5,000 practitioners.
What Happened
- 2–3 September: Aurecon and Arcadis back Nomic. Both firms became strategic investors in Nomic, a domain-specific AI platform for the built environment. Aurecon is opening it to its whole 6,700-strong workforce after a four-year working partnership. Arcadis is embedding Nomic's agents into its project-delivery model after a six-month trial with around 150 engineers in 12 countries. Nomic says 86% of them reported that it changed how they approach their work.
- 23 September: Accenture Construct. Accenture's new business pulls together more than 5,000 practitioners, many of them from acquired specialists such as the UK's Soben, to help owners plan and deliver capital projects. One of its lines covers the process industries, including liquefied natural gas. Accenture puts the market for integrated owner-side capital project services at $260 billion.
- 28 September: MAIRE joins Octave CoLabs. The Milan-based engineering group, which according to Construction Digital employs around 10,800 people and builds plants in more than 50 countries, signed up for Octave's 11-week programme to develop a multi-agent system around its own workflows. Fluor and Bechtel joined the programme when it launched in July.
The agents described are document workers. Nomic's, in Aurecon's words, review drawings against firm standards, check code compliance, take a first pass at submittals, research RFIs and answer questions "with traceable, cited results".
Why It Matters
The familiar pattern for enterprise software is simple: a vendor builds it and the firm buys a licence. September's deals bend that pattern. Aurecon and Arcadis took equity in the platform. MAIRE is co-developing agents shaped around its own processes. The firms are no longer just customers. They are helping to build the thing, and they are feeding it the one asset no software company can buy outright: decades of their own project records.
The firms are no longer just customers. They are helping to build the thing.
That changes where the advantage sits. If an agent is only as good as the documents behind it, the firm with the richest, cleanest project history gets the better agent. Octave's chief product officer, Jay Allardyce, framed MAIRE's case in exactly those terms in Construction Digital: "consistency and data integrity" are critical when "failure is not an option".
For energy readers, the shift is closer to home than most AI news. MAIRE builds energy and chemicals plants; Construction Digital notes that its Tecnimont subsidiary took a further US$125 million in LNG awards in August. Accenture's LNG line is aimed at the same owners. When the delivery firms in oil, gas and chemicals start building agents on their own records, the planners in those firms will be among the first to feel it.
What It Means
For project controls teams inside these firms, three things follow.
First, the agents are arriving through document control, engineering and IT, not through planning. None of the September announcements mentions schedules, float or earned value. If project controls wants its data treated as a first-class source, rather than an afterthought, it has to ask.
Second, data hygiene becomes strategy. An agent built on a firm's own records inherits every inconsistent coding structure and every half-finished close-out report. The unglamorous work of standardising activity codes and change logs starts to look like an AI investment.
Third, the human-in-the-loop language is consistent but untested. MAIRE describes its approach as "Human in the Loop", and Aurecon says agents operate under its existing security and governance structures. Those are statements of intent. The only performance figure in the round, Nomic's 86%, measures how engineers felt about the tool, not whether projects finished sooner or cheaper.
What to Watch
- Whether MAIRE, Fluor or Bechtel publish any measured outcome once their 11-week CoLabs programmes conclude.
- Whether firm-wide agent platforms extend from documents to schedules and cost reports, and who in the firm owns that step.
- The ADIPEC leadership roundtable on 4 November in Abu Dhabi, which asks which capital-project decisions can become "machine-led" and where human judgement stays essential.
- More equity-for-access deals, where delivery firms invest in the AI vendors they use.
The software vendors spent the last two years asking engineering firms for their data. In September, some of the firms answered by keeping it, and helping to build the AI themselves.
Method. Sources were checked on 1 November 2026. Workforce sizes, trial figures and market estimates are as stated by the companies or the trade publication cited. They have not been independently verified.