I’ve spoken with several AEC software company founders who are bewildered about where they’re headed. If every client can now build its own software with AI, why would anyone buy ours? What is the value we can provide in the future?
The worry is understandable. With AI coding tools, a project manager can build a tool for a very specific need, even for a single project, in an afternoon. As one podcast guest recently put it, he no longer has to wait for a software company to release something that nearly fits his needs.
Predicting the impact of exponentially developing technology is hard. My view is that we’ll need software companies for years to come. In the longer run, the way we build, buy, and use software will change beyond recognition. That doesn’t mean the AEC software industry will die, but it will look very different.
We have been here before
Construction professionals building their own software is nothing new. In the 1980s and 1990s, many developers in our industry were architects and engineers without much formal training in programming. We learned by doing, borrowing code from others and adapting ideas from magazines and, later, the internet. We rarely started from scratch, though. We built on CAD systems, then BIM platforms, and office tools.
AI now lets us build from the most basic elements, and it also makes extending existing platforms far easier. As software exposes APIs and MCP interfaces, clients can add features and even let AI agents operate their tools without waiting for the vendor.
However, the fact that everyone can code doesn’t mean everyone wants to. Most people won’t be interested, but every organization has a few individuals eager to put these new tools to work.
The first version is the easy part
I develop software with AI myself. Getting to a working first version, a minimum viable product, is surprisingly fast and deceptively satisfying. In traditional enterprise software, however, the real work starts after the prototype: back-end databases, integrations, backups, user management, cybersecurity, and scalability.
AI tools for testing, refactoring, and cloud deployment are improving rapidly, making even this work easier. Moreover, many AI-generated tools won’t need a full enterprise back end. With open APIs and MCP interfaces, project teams can create ephemeral software, such as disposable scripts or single-purpose interfaces that complete a task on one project and are then discarded. Such tools don’t carry the maintenance burden.
A tool that outlasts its project is a different matter. When an enthusiastic engineer builds it, it can become orphaned the day that person leaves, and our industry has plenty of abandoned macros and spreadsheets to prove it. An organization that builds internal software at scale has to take on an unfamiliar role: managing its own software maintenance and IT operations.
Where software vendors hold the line
If project managers can build their own interfaces and scripts on the fly, the value of commercial software shifts from front-end features to infrastructure and trust. Commercial software has always been more than code. It requires product management, sales, support, and productization. Customers also buy continuity, a guarantee that the software will be supported and secured for years to come.
A firm can generate dozens of custom tools in a week, but those tools still need structured, reliable data to act on. This matters more as construction becomes data-driven and data becomes the raw material for both AI and new services. Software companies will increasingly act as data stewards, maintaining the standardized formats, common data environments, and project schemas that keep complex ecosystems aligned.
Security and liability set another boundary. In an industry defined by strict legal obligations, custom-built micro-apps may not meet those demands. Vendors provide audited, secure environments that support compliance and keep a clear record of where data came from and how it has changed.
People matter, too. As an architecture firm’s R&D manager, a role we might call CTO today, I saw how differently professionals with the same background used the same software. Vendors, who understand that psychology, build tools people actually adopt, and AI does not attend design meetings or walk construction sites.
Instead of shipping fixed feature sets, successful vendors will build platforms. By offering well-documented APIs and MCP endpoints, they give client-built AI agents a safe place to operate without compromising core systems. Custom AI coding will handle the specific needs of individual projects, while software companies provide the foundation that keeps the industry connected and compliant.
The job, not the tool
Looking further ahead, it helps to ask why customers use software at all. They want a job done, and the best technology is the one that makes it easiest. Nobody wants a BIM tool as such; they want a coordinated design. Every hour spent learning menus, workarounds, and file formats is overhead that adds nothing to the result.
Computing has always moved in this direction, from hardwired machines to machine code, programming languages, and now plain-language instructions. Over the next decade, I expect the traditional application, something you buy, install, and learn, to lose its central role. We will increasingly describe what we need and let an environment respond with the data it has. Incidental knowledge, such as where a command hides in the menu, will fade, while essential domain knowledge, the ability to judge the result, becomes more important.
Construction will not be the first mover, as liability, regulation, and legacy systems slow change. Yet our industry may benefit more than most, because every project is a prototype that generic software has never fit perfectly. Software companies will not disappear, but they will change beyond recognition. The ones that thrive will provide the data, platforms, and trust everyone else builds on.