Daily field briefing
POLR AIPOLR AI

AI + AEC Today

Daily AI intelligence for the built world

Weekly digest

Weekly AI in AEC stories

← Archive

Edition · Friday, July 24, 2026

AI & Construction

Weekly Pulse · Industry actionability

The field is piloting, not just experimenting.

60%
ExperimentalPilotingField-ready
  • AI is moving onto the job site to measure real work, not just office paper
  • New tools let a firm fence in what an AI agent is allowed to do
  • Where an AI model came from is turning into a real business question
  • Governments are betting big money on AI for building problems

Practical Use of the Day

Fence in an AI agent before you let it loose

The big lesson this week is guardrails. Before you let an AI agent do real work, write down its limits. List what it may do on its own, like draft a reply or pull an RFI status. List what it must never do without you, like send money or promise a date. Then test it on a few old cases first. Start small and watch it.

Try this prompt

You are helping me set safe limits for an AI agent that will handle routine construction office work. I will describe one task, like answering vendor emails or checking invoices. First, list the steps the agent can safely do on its own. Second, list the steps where it must stop and ask a person, and say why. Third, give me three test cases I should run before I trust it, including one tricky case that should force a human handoff. Keep it plain and short. Here is the task:

Tool — Claude (or any capable LLM)

Stories

4 curated
ConTechModerate

Sodex raises 4M euros to map job sites live with AI

Sodex is a company from Austria. It just raised 4 million euros for its site mapping tool. The tool turns your machines, like dozers and loaders, into rolling sensors. As they drive, they record the ground and how much dirt got moved. AI then does the math and shows your progress on a web page in near real time.

Why it matters to GCs

Today, someone has to walk the site and measure by hand, which takes time and can be off. Live maps and dirt counts let you bill sooner, spot slow areas faster, and argue less over how much work got done.

tech.eu·Jul 13·Read source →

LinkedIn
AI FrontierModerate

US says China's Moonshot copied Anthropic's top AI

The White House says a Chinese firm, Moonshot, copied a top AI model from Anthropic, the maker of Claude. The trick is called distillation. You use a strong AI to train a weaker one, without asking. Moonshot's new model, Kimi K3, came out July 16. The US now says sanctions could follow. Some experts doubt copying alone could explain how good the model is.

Why it matters to GCs

More firms now build their tools on top of AI models. Where a model came from, and whether it is safe and legal to use, is starting to matter. This fight is a sign that AI has become a big power struggle between the US and China.

TechCrunch·Jul 22·Read source →

LinkedIn
AI FrontierModerate

OpenAI launches a safer way to put AI agents to work

OpenAI put out a new product called Presence. It helps big companies set up AI agents that answer calls and chats. You set the rules first: what the agent can do, and when it must ask a person. The agent gets tested before it goes live. OpenAI says its own agent already handles 75% of support calls with no human help.

Why it matters to GCs

Your back office is full of calls and emails an agent could take, like vendor questions, invoice checks, and RFI status. The real news here is the guardrails. You can pen in what the agent may do, so it does not go off script with your money.

Help Net Security·Jul 22·Read source →

LinkedIn
AI & BusinessLow

US to spend $5B on AI research, including tougher building materials

The US government will spend $5 billion on science that runs on AI. Fifteen agencies will use big federal computers and data to run tests faster. One goal is to make building materials that last longer. Microsoft is adding $40 million in computing credits to help.

Why it matters to GCs

This will not change your next job. But stronger, longer-lasting materials could mean less repair work and fewer callbacks down the road. It also shows the government is betting big on AI to solve real building problems.

The Daily Reporter·Jul 23·Read source →

LinkedIn