Something is happening in software procurement that will shape the next five years for every AEC firm, and it is easy to miss because each individual step looks like progress.

Two forces are colliding. Frontier models are converging, so raw model capability is becoming a commodity that everyone can access. At the same time, AI has made software radically cheaper to build. Put those together and you get the result we are all now living through: a niche AI tool for every workflow, arriving faster than anyone can evaluate them. A tool that reads drawings. A tool that drafts RFIs. A tool that reviews specs. A tool for submittals, for schedules, for safety forms. Each one demos beautifully. Each one solves a genuine problem for a genuine team.

And each one puts the same quiet question to your firm, whether or not you have noticed it: do you keep buying point solutions, one workflow at a time, or do you invest in the foundation that lets AI actually compound? This post is about why the first path is a trap.

The Flood is Real. AI is Pouring Fuel On It.

Software sprawl was already a problem before AI. The average enterprise now manages around 300 SaaS applications, according to Zylo's 2026 index, and roughly half of those licences go unused. The waste runs to about 18 million dollars a year for a large organisation, and Gartner estimates that close to a third of global SaaS spend is toxic, money spent on tools that sit idle. This is the baseline, and it was built one reasonable purchase at a time.

AI is now accelerating it hard. Spending on AI-native SaaS grew 108 percent year over year in the same index, and vertical tools, construction among them, are among the fastest-growing categories in the market. When building software is cheap, the market fragments into a tool per task, and the pace of new arrivals outstrips any firm's ability to assess them. None of this means the tools are bad. It means that treating "buy the tool" as a strategy no longer scales, because there is now a tool for everything.

The Tech Stack You Never Designed

The cost of a point solution is never just its subscription. It is the integration to your systems of record. It is the pipeline that feeds it your data. It is the maintenance every time either side changes. It is the training, the vendor management, and the renewal, and SaaS renewals now commonly rise 10 to 20 percent a year, negotiated from a position of weakness because switching costs are high and the vendor knows it.

Multiply that by a dozen and you have a system nobody architected. Expensive to run, brittle at every seam, and impossible to reason about as a whole. This is not hypothetical for our industry. Large contractors already run 15 to 20 platforms with minimal integration between them, as we covered in our analysis of the AEC technology gap. Adding an AI tool per workflow does not fix that fragmentation. It deepens it, faster.

Every Tool is Another Copy of Your Data

This is the part that should concern a regulated industry most, and it is the part the demos never mention.

Each point tool needs your project data to function, so each one takes a copy, held on infrastructure you do not control, under a security posture you did not set. Every copy is a new attack surface and a new data-exposure risk. Shadow IT makes it worse: Gartner projects that by 2027, three-quarters of employees will acquire or build technology outside their IT function's visibility, up from 41 percent in 2022. People will adopt AI tools faster than anyone can govern them.

In AEC, a single project carries confidential designs, commercial terms, and client data. A dozen vendors each holding a slice of that, under a dozen different security regimes, is not a strategy. It is a liability waiting for an audit finding or a breach. Reliable, safe, secure AI in a regulated field cannot be assembled from sprawl.

Point Tools Accumulate. They Do Not Compound

The deepest problem is the quietest one. All of this spending buys capability that never adds up.

Twelve tools that do not share a data model, an ontology, or a memory cannot build on one another. The knowledge each one extracts from your projects stays trapped inside that vendor's silo. The drawings tool learns nothing the specs tool can use. You cannot ask a question that spans two of them without building yet another integration to join their outputs by hand. You are not accumulating AI maturity. You are renting a dozen disconnected fragments of it and paying an integration tax to make them speak.

Maturity, it turns out, is not a thing you can purchase one subscription at a time. It is a property of a foundation, and a foundation has to be built or owned, not assembled from other people's products.

The Fork

So every firm faces a choice, named or not. Keep buying point solutions, and accumulate fragmentation, cost, and risk, with capability that resets to zero at every vendor boundary. Or invest in AI-readiness: the owned foundation that lets any tool or model plug in and compound on top of it.

The first path always feels like progress, because something is always happening, a new tool, a new demo, a new logo on the stack. The second is quieter and it is the one that actually gets you somewhere. What that foundation is, and what a firm has to own to be genuinely AI-ready, is what we will lay out next.

To be clear, we are not against tools. We use plenty, and the good ones earn their place. The argument is narrower and more important than "tools are bad." It is that you cannot buy your way to AI maturity one point solution at a time, and in a regulated industry the attempt carries real risk. If you want to look at your own stack and the foundation underneath it, that is a conversation worth having.

Book a working session →