AI Agents for Construction: Introduction and How to Get Started

AI Agents Are Not Another System to Learn

Every construction business runs on the IP (intellectual property) it has built up, the people who deliver the work, and the systems they work in. AI agents do not land on the systems side of that equation. They land on the labour side.

Doug Vincent walks through what that means in practice. How an agent plans and executes a task on its own, what has to be built around it before it can touch real project work, and how to find a first use case without betting the business on it.

What You'll Learn

• What separates an AI agent from Claude and ChatGPT, and why goal-driven beats answer-driven.
• How an agent actually works, breaking a goal into steps and calling a model many times to finish it.
• The five components every agent needs before it can be trusted with project work.
• Why a construction agent has to hold memory at the user, project and business level.
• How one site photo becomes an issued EOT notice in two to three minutes.
• The use cases already working today, from inbox monitoring and registers to tender packs and meeting minutes.
• Which tasks belong to an agent, and which still need human judgment.
• Why permissions break first when you move from one user to a whole team.
• Where AI still fails, including quantity takeoffs from drawings and open-ended document requests.
• How to keep token costs under control by routing different tasks to different models.
• The two steps to getting started, and the three ways teams are deploying agents now.

Chapters

00:00 Introduction: What are AI agents in construction?

‍00:44 Doug Vincent's background and why he founded Mastt

‍01:55 The agenda: what this webinar covers

‍02:51 What is an AI agent? Agency, goals, tools, and guardrails

‍04:00 AI agents vs Claude, ChatGPT, and AI chat tools

‍04:44 The five categories of AI tools and where agents fit

‍05:50 Where AI agents fit in construction today

‍07:44 The agent reasoning loop: how does an AI agent work?

‍10:33 Construction AI agent example: creating an EOT notice

‍12:55 Five core components: governance, tools, memory, skills, LLMs

‍14:41 AI skills explained: reusable workflows in Claude

‍16:47 Why memory matters: user, project and workspace memory

‍19:15 How an enterprise AI agent connects business systems

‍20:40 Challenges with AI agents today: permissions and privacy

‍24:17 Australian data residency and offshore AI processing

‍26:16 Drawing takeoffs, context windows and the ContractorOS benchmark

‍29:01 Token pricing, model routing and AI agent costs

‍32:39 What is possible with AI agents today?

‍34:16 When should you use a workflow instead of an AI agent?

‍35:45 AI agent example: an EOT from site via WhatsApp

‍37:43 AI administration vs human project management judgment

‍38:59 How to onboard an AI agent like a new employee

‍40:44 How much time can AI agents save?

‍41:24 How to get started: discovering your AI agent use cases

‍43:20 Operationalizing use cases with an agentic harness

‍44:43 Three ways to deploy: hobby, in-house build, or licensed platform

‍47:52 The future of AI agents in construction

Who It's For

• Project owners and asset managers deciding where AI fits in delivery.
• Owner's representatives and PM consultants carrying heavy contract administration loads.
• Project managers and project controls professionals losing hours to registers, notices and reporting.
• Program managers looking for capacity across a portfolio rather than one project.
• Directors and business owners weighing up whether to build, buy or wait.

See What a Construction Agent Actually Does

You have seen what an agent can do. The next question is what you would hand over first.

Mastt's AI Agent ships with the permissions model, memory layers and templates covered in this webinar already built, so you are not assembling a harness yourself. Check out Mastt AI Agent and see it running on real project work.