The cost of building software has fallen close to zero. This fall we stood up fourteen working project sites from the public bid documents of awarded public works projects. Each one indexes the full set of drawings and specifications, drafts a submittal register from the specifications that runs to hundreds of lines, and screens the project for permits. A new one takes about a day. Not long ago that was months of work for a funded team.
That changes what is worth paying for. When anyone can stand up the tools, the tools stop being the product. Every contractor, engineer, and testing lab in the country can open the same frontier model we use tonight. Most have. In this year's AGC and Sage survey, six in ten construction firms said they are using AI or planning to. In last year's RICS global survey, fewer than one in a hundred had it running across the firm.
The gap between those two numbers is the business we are in.
A subscription does not close it. We wrote in June that the model is now a capable junior: fast, well read, and wrong often enough to need a supervisor. A junior with no supervisor and no assignment summarizes meetings, and that is where most firms have stopped.
The value arrives when someone who knows the work gives the junior a real job. That means knowing which submittal the specification actually requires, what a controller checks before a pay application goes out, why a density test that fails at two in the afternoon matters to a pour scheduled for six the next morning, and who in the firm needs to hear about it first. That knowledge takes years on projects to earn. Building well with these models is its own craft, and it takes a different set of years.
Very few people hold both. The people who know the work are busy doing it. The people who know the models have mostly never read a specification, closed a month, or signed a test report. We have done the work, and we build with these models every day. That combination is the scarce thing, and it is what we sell.
So Hesiod leads with AI strategy. We sit down with a firm and trace how work really moves from the field to the front office. We find every workflow with a gap in it and group the ones a single solution can close. Then we build, inside the systems the firm already runs. Where a firm has no system for a job, we bring one of ours. Where it has one it trusts, we connect to it. We stay through adoption, until the new way is simply how the firm works.
Done right, the AI disappears into the work. The submittal arrives already checked against the specification. The test result reaches the superintendent before the pour. Nobody opens a chat window, and nobody calls it AI.
This reaches past general contractors. An engineering firm's hours go to reports, reviews, and proposals. A geotechnical and materials testing lab runs on field tickets, lab results, and reports that have to reach the project team the same day. The documents differ from firm to firm. The shape of the problem is the same, and so is the fix.
The firms that move first will carry more work with the same team. We take on a few a year.