Accuracy Expectations and Trust-But-Verify

Mike and Luigi outline customer accuracy thresholds and the need for human-in-the-loop verification.

Play episode from 32:06

chevron_right

Transcript

chevron_right

Transcript

Episode notes

The UFC arrives in AEC. And we’re live from the AEC Octagon.

In the very first episode of our brand-new Bricks & Bytes Debates series, we are trading PR statements for sparring gloves. We’re putting two industry heavyweights head-to-head to battle out one of the hottest topics in construction: Has AI actually solved estimating?

In this no-holds-barred contester series, our contenders took the gloves off to represent the extreme Bull and Bear cases for Artificial Intelligence:

• In the Red Corner: Mike "Mighty" Powers (Co-Founder, BuildVision), arguing that LLMs have solved the extraction problem and the industry needs to embrace the "Waymo" era of automation.

• In the Blue Corner: Luigi "The Stinger" La Corte (Co-Founder, Provision), arguing that estimating is more than just counting symbols - and that without 100% accuracy, AI is still just a student driver.

Moderated by the referee of reason, Patric "Praying For Exits" Hellermann, this debate moves through three grueling rounds:

- Round 1: The Tech. Can LLMs actually count and measure complex drawings, or are they just guessing?

- Round 2: The Accuracy Gap. Is 80% accuracy "good enough" to change the industry, or is "trust but verify" just a nice way of saying "do it yourself"?

- Round 3: The Liability. In a world of design-bid-build, who takes the fall when the AI misses a scope gap?

Tune in to uncover:

• The "Waymo" Strategy: Why waiting for perfect AI is a mistake, and how to leverage the "self-driving" mindset to reduce takeoff times by 90% today.

• The Hidden Liability Loophole: Why architects and engineers could solve the estimating crisis instantly by sharing models—and the legal reason they never will.

• Symbols vs. Language: Learn the fundamental technical limitation preventing LLMs from reading 2D drawings like a human, and when we can expect "Visual Language Models" to catch up.

• The Subcontractor’s Dilemma: Why "Design Intent" matters more than counting linear feet, and why subs might actually be wasting time on granular takeoffs.

• A "Bribe" for Better Data: One listener’s hilarious (but practical) suggestion for how to finally get the Revit model from the design team.

Who will deliver the knockout punch? Tune in to decide for yourself.

Chapters

00:00 Intro

01:48 Introduction to Bricks and Bytes Debates

04:07 Opening Statements and Context Setting

06:51 Technical Feasibility of AI in Estimating

09:57 Practical Applications and Limitations of AI

12:59 Customer Expectations and Accuracy in AI Solutions

15:57 The Role of Human Oversight in AI Estimation

18:43 Future of AI in Construction Estimation

38:23 Navigating the Self-Driving Debate