AI Won’t Fix a Badly Run Company: AI 101 for Construction Leaders | Ep. 388

This is not a tech seminar. We’re not chasing shiny objects, we’re not worshipping the robots, and we’re not pretending AI is going to magically fix a poorly run company. This is the first episode in a new series on AI for construction companies, and it answers one question: how can contractors use AI to save time, communicate better, reduce confusion, and get real work done? Dylan Davis is my collaborator on the Construction Genius GPT course, where we teach construction professionals to use AI on real work. Dylan brings the AI expertise. I bring the construction translation.

What you’ll learn:

• – What AI actually is (a prediction machine) and where it fits in everyday construction work
• – Why context, not prompting, is the last competitive lever left
• – The What-Why-How framework for getting useful output every time
• – Why the first answer is the start, not the finish, and how to iterate like a pro
• – Which model to use for which task: ChatGPT, Claude, or Gemini
• – How to keep your company data safe, and the one thing Dylan won’t let AI do
• – Three moves to make on day one of your AI journey

Listen or Watch the Episode

A Prediction Machine, Not a Magic Box

Strip away the jargon and AI is a model that predicts what comes next based on what you ask it. That one small capability, predicting the next word, scales up into writing RFIs, drafting change orders, summarizing meetings, and analyzing fifty past projects to estimate the next one. You don’t need to understand the back end. You need to understand what to feed it.

Start Small. Start Simple. Don’t Boil the Ocean.

The most common day-one mistake is asking AI to automate all of finance. Too much, too fast. Start with one email, one report summary, one small task. Build your ability with the tools, learn what they can and can’t do, then add complexity. And when you don’t know how to do something in the tool, ask the tool. Talk to it like you’d talk to a friend. No special language required.

Where It Gets Risky

AI is chaotic by design. Same prompt, same model, three different conversations, three slightly different answers. If a task has to come out identical every single time, don’t hand it to AI directly. Have AI write code that does it the same way every time instead. Where it shines is generative work: drafting, summarizing, spotting patterns and anomalies across a stack of Excel sheets.

Where AI Fits in Everyday Construction Work

RFIs drafted from your inputs, in your structure and your tone. Estimates predicted from your last fifty projects. Change order drafts. And the sleeper use case: meeting transcripts. You don’t need to clean up a messy transcript before handing it to the AI. Feed it the crosstalk and all, tell it what you want, and it will pull out the summary, the lines of responsibility, and who’s doing what by when.

Talk, Don’t Type

Dictation is the highest-leverage habit change in this entire episode. Get the ChatGPT or Claude app, and speak into it. Think of yourself as a large language model: you have deep expertise that comes out when you’re prompted correctly. Dictate what you want, then ask the model to ask you questions about what you said. It will pull out insights you didn’t know you were carrying.

“When you talk to an AI, talk to it as if you’re teaching a new employee.”

Smart, capable, and clueless about your business until you teach it. The more context you give the engine, the more effective it becomes.

Context Is the Last Competitive Lever

This is the strategic heart of the episode.

“There are three levers any company can pull to separate themselves from their competitors when it comes to AI. And two of those levers have disappeared.”

Everyone has access to the same models now. Prompting matters less every month because the models keep getting smarter. What’s left is context: sharing the right information at the right time. Dictate it, drop in files, or build a dedicated project inside ChatGPT or Claude that holds your company’s documents and references them every time. The contractor with the best context wins.

The What-Why-How Framework

Three questions to answer every time you ask AI to do anything. What exactly do you want, including binary pass-fail criteria so the AI can grade its own output. Why are you doing this, so the AI can fill the gaps you forgot to mention. How do you want it delivered: paragraphs not bullets, PowerPoint not PDF, the constraints around the output.

One update worth noting: stop telling the AI what you don’t want. Negating terms like “never do this” tend to steer newer models the wrong way. Use prescriptive terms. Say what you want, not what you don’t.

The First Answer Is the Start, Not the Finish

This came from one of Dylan’s clients, and it’s the mindset that separates the people who get value from AI from the people who call it hype and walk away.

“When you interact with an AI, the first response that you get back from the AI is where things start, not finish. This is the beginning, not the end.”

Bad output? Don’t blame the AI. Blame yourself. Give it feedback, iterate five or ten times, and when you finally land on the output you want, tell the AI to turn the whole conversation into a prompt so the process is repeatable forever. If you’re not willing to iterate, AI won’t work for you. And the cost of walking away is real.

“When you have ten other competitors that have accepted the fact that it is useful and they’re leveraging it, and you postpone for another year, that’s a year you’ll never get back.”

Promote Your AI Like a Project Engineer

You hire a project engineer out of college with a shiny degree. Clueless about how to run a project, but not stupid. You give them time in the seat, rotate them through estimating, operations, maybe business development, and over time you delegate more. Treat AI exactly like that. Start it on small tasks, feed it the context of your business, and promote it from project engineer to project manager to project executive. Start with an hour a day in the tools. The end state is a tab that’s always open and a standing question: can AI assist with or do this for me?

Which Model for Which Task

The answer changes by the week, but as of this conversation: Claude for writing tasks that need to sound human. Gemini for processing high volumes of PDFs at lower cost. ChatGPT and Claude’s newest models for data analysis across Excel sheets and for code. ChatGPT for images and infographics. Don’t get overwhelmed by the choices. The ecosystem matters more than the pick: get comfortable now so you can flex as the models change.

Is Your Company Data Safe?

If you’re on a paid business plan with ChatGPT, Claude, or Gemini, the security level is similar to Gmail, Outlook, or Dropbox, with SOC 2 and similar compliance behind it. Go into settings and toggle off training on your data; every major provider has the switch. Dylan’s one real boundary isn’t about what AI can see, it’s about what it can do. Let it read the bank account. Don’t let it send the money. Access, yes. Action, carefully.

Three Moves for Day One

Don’t get overwhelmed by model choices; use whatever you already have a subscription to. Make a list of the time sucks in your day and break each one into its steps, because those steps are what you’ll hand to the AI. And start dictating. The first couple of days feel weird. Push through. You’ll never want to type again.

Where This Is All Headed

Dylan’s five-year picture: most knowledge work augmented or automated, ambient AI working alongside you at all times, and physical-world AI catching up faster than people expect because robots learn from simulated worlds. Whatever happens, the data centers still need to be built. Contractors, take note.

Subscribe and Connect

If this episode hit home, share it with someone on your team who needs to hear it. Subscribe to Construction Genius wherever you get your podcasts. Leave a rating and a review.

Get on the waitlist for the Construction Genius GPT Course: https://ericanderton.activehosted.com/f/255

Connect with Dylan Davis on LinkedIn: https://www.linkedin.com/in/dylantdavis

Dylan’s website: https://offerings.gradientlabs.co

Dylan’s YouTube channel: Dylan Davis AI https://www.youtube.com/@dylandavisAI