Rebuilding the Construction Software Stack

I’m seeing a lot of people in the AEC/EPC space building with AI and it’s really exciting to see. Disruption is happening in real time, and fast. But as an AI-native capital project director, I’ve watched this rush with a mix of excitement and skepticism. I am a fully converted believer in the potential for AI to fundamentally reshape how we build. However, we are at a critical crossroads where we risk banking on our reindustrialization aspirations on tools that make it look too easy.

US construction is a $2.2-trillion-a-year industry, with the global market rapidly accelerating toward $20 trillion this decade. Unsurprisingly, it is one of the most coveted targets in tech. Y Combinator has funded over 40 construction startups; venture capital giant Andreessen Horowitz openly laments that “every building you’ve ever been in was designed by software built in 1997” while begging founders to pitch them; and climate-tech funds are raising hundreds of millions to rebuild the physical economy.

But despite the capital begging to be spent, capital projects have remained remarkably resilient against disruption. The tools that do exist are expensive, fragmented, and completely unintuitive. In an industry that still runs on spreadsheets, paper checklists, and phone calls, any new technology needs to meet people where they actually work: in the field, in meetings, parsing sticky notes, and managing never-ending scope changes.

This problem couldn’t be more relevant. We find ourselves at a critical inflection point across multiple industries: the advent and rapid proliferation of frontier pro-sumer AI tools and a point where we have collectively looked at our country and the hollowing out of our industrial base, our decaying infrastructure, the over-financialization of once-mighty, once-innovative conglomerates. We yearn to build! Not everyone is built to learn to code! Turns out, code can be built to learn to code, and people can keep building. But the change is catching us off guard: we also, inadvertently, hollowed out our construction base, and we are now capacity-constrained in even being able to build the thing to build the thing.

What’s worse, reindustrialization doesn't look like building familiar, repeatable projects. We are trying to construct advanced semiconductor fabs, critical minerals processing plants, novel materials facilities, and grid-scale infrastructure using technologies that are either brand-new or were abandoned in this country decades ago. The domestic expertise base is dangerously thin. Where knowledge does exist, it is deeply fragmented between project owners and EPC firms. Both parties hope the other can fill in the gaps, leading to disappointment and frustration.

For decades, the EPC model was simple: an owner hands a project to a third party to buy expertise, execution muscle, and a lump-sum price that moves the financial risk onto someone else's balance sheet. But this only works when projects are entirely predictable. Combine first-of-a-kind facilities with tight timelines, escalating material costs, exploding demand for electrical gear, and inflation, and predicting project trajectory becomes an impossible task.

Owners still enter ambitious projects thinking they can simply hire the experts, only to be shocked by the massive administrative overhead and oversight required. The big players in the EPC space also agree this is no longer working. Larger firms report that their revenue is now 50% to 70% cost-reimbursable rather than lump-sum, because when a cutting-edge scope isn't fully defined, you cannot accurately price the work. But this isn’t innovation; it is just risk mitigation. It’s no secret that projects fail more often than not. IPA, which has benchmarked large projects for decades, puts the megaproject failure rate around 65%. Brent Flyvbjerg has spent a career documenting it: nine out of ten megaprojects run over budget, over and over, across decades and continents.

The energy to fix this is real, and for the first time ever, I’m seeing everyone in the construction industry get excited and even build tools for themselves. This is such a breath of fresh air. But a lot of the AI tools being built for the industry right now (not by the industry) feel like someone discovered capital projects last week and decided the hardest part must be the Process Flow Diagram (PFD). A lot of folks have had the idea to auto-generate process models, automate mass balances, and synthesize techno-economic analyses. They wire these into a clean UI with some dashboards and call it an "AI EPC Killer." These tools are useful, but they are not serious capital project tech. These are underwriting tools—advanced blueprint generators that treat a capital project as a static, financial math problem.

They are fundamentally conflating design with planning.

Design is the "What." It answers the question: In a perfect world with frictionless physics and infinite labor, what are we building? Process engineering is extremely well-documented; automating it with an LLM is just a faster version of what engineering software has done for decades.

Planning, specifically execution planning, is the "How." It is the invisible, highly complex operational and legal architecture required to survive reality. It has almost nothing to do with physics and everything to do with logistics and human behavior. True execution planning answers the boring but extremely important questions:

How many field engineers do we actually need on-site to verify that the work matches the invoices?

Did we structure our contracts so subcontractors are legally obligated to provide frequent, auditable progress metrics—or can they hide their bad news for months?

How do we monitor cash burn in real time rather than waiting for accounting to tell us we went over budget six weeks ago?

If we start to go over budget or behind schedule (or, usually, both) mid-project, what recovery options will we have? Will we even have the data necessary to do an effective analysis? Does a night shift actually add enough productivity to justify the cost?

Historical execution statistics are available for anyone to research, but execution intelligence lives inside proprietary corporate databases at best, and inside people’s heads at worst. Turning a perfect process design into a physical facility without letting the facility eat the business case is the exact thing people routinely underestimate.

A plant is not a PFD with a building or some pipe racks wrapped around it. The buildings, the layouts, the utility strategies, the structural concepts, the procurement packages, and the construction sequences are the project.

I’ve been watching these tools get built and struggling to put into words what I felt, but something was just off. Yes, some of the better ones generate CapEx estimates. But you can look at them and tell they’re built for a snapshot in time, not reality. But megaprojects fail when they start becoming real, and the scope starts getting divided up amongst more teams. They fail at the seams: whenever information has to cross a boundary, some of it is lost.

The capital projects industry currently has a disparate suite of tools, none of which do it all. Improving one of them, or a few, is admirable and necessary. But to really disrupt the industry and deliver step-change execution abilities, we need a true, end-to-end Mission Control platform.

A design tool only tells you how the rocket should fly. But the second you break ground, the project is live, in the air, and experiencing atmospheric friction. Planning isn't a static phase you complete before construction; planning is continuous. It is a dynamic, real-time optimization problem. Every time a critical piece of equipment is delayed in transit, or a foundation pour fails inspection, the project must be replanned. In the current software ecosystem—where field updates live in one app, scheduling lives in another, and financial burn lives in a spreadsheet— data latency is fatal. It takes days or weeks for an issue to ripple through to the forecast, and more to try to solve it.

A capital project is a single, continuous, hyper-complex organism. Design, procurement, scheduling, and contract governance are all different lenses on top of the exact same data stream. A single source of truth eliminates data latency. It ingests the live telemetry of the job site—messy human updates, daily checklists, and logistics realities—and recalculates the destiny of the project. If a procurement date shifts, the 3D model, the contract compliance ledger, and the cash flow model must update simultaneously.

If we manage to build this unified Mission Control infrastructure, we will kick our full industrial might into gear by supercharging engineers and project managers.

In software, there’s a concept of the 10x or even 100x engineer. In capital projects, a unified intelligence engine makes the 100x Operator a reality. This is not about replacing the licensed human engineer; it is about providing them with the cognitive scaffolding to hold scope, cost, schedule, design, and contract compliance in their head at once. If we can give teams a unified knowledge engine, we can kick our full industrial might back into gear by supercharging engineers and project managers.

Critically, this would also allow a lean team to run a highly disciplined project, and actually value-engineer the entire project in real time rather than waiting until the final estimates come in over budget. The limited talent pool for true megaprojects could execute faster, allowing more capacity for more projects. Engineering and construction teams would work with less siloed information, tearing down the traditional walls between the back office and the field.

What people will learn the hard way, if we’re not careful about how we deploy some of these tools, is that you can outsource design, can outsource administration, to some extent – even to AI! – but there is a physical reality to actually managing the chaos that is a mega- or Gigaproject and getting it across the finish line with your sanity intact. The only way to do that is to have deep ownership in every possible facet of the project and plan out all of it – plan the execution, plan the recovery, plan the what-ifs, the contracts, the forces majeure – and keep planning when things don’t go to plan. To replace or alleviate the pure brute force this takes will require a true living model of the project, not a frozen blueprint. It's what will allow an owner to run a lean, sharp team instead of renting (or even hiring) a giant one. It's what will enable the E and the PC of an EPC to actually work together with fewer barriers between office and field. And it's the only thing that will make this grand industrial renaissance more survivable and repeatable so we build the muscle memory of how to build again.