NEW ON BEYOND THE BID
Last Tuesday, Brian Turmail of AGC of America and Christene Marie joined me for Part 2 of What’s Next for Construction?, on planning for 2027. When it wrapped, I had 90 minutes of raw footage and no editor.
So I tried something new. I put one AI in charge of another.
The first pass: Claude in Tella
Tella is a video platform I use for podcasts and talking-head videos. Think of it as Loom, upgraded. It connects directly to Claude, so I uploaded the recording and let Claude run a full production pass:
Cold opens and hooks
Cut the “ums,” “ahs,” and dead air
Motion graphics and digital b-roll
It got the job done. But when I watched it back, the edit was flat. The zoom cuts were weak, and the cold opens ended without a transition sound. It was technically finished but had no energy.
The gap: who checks the work?
I had two options: fix it by hand, or keep rewriting prompts until Claude got it right. I didn’t want either. I wanted someone to hold the edit to a standard so I didn’t have to.
That’s a project manager’s job.
The fix: Muse as the PM
Muse is Meta’s AI assistant. I gave it access to my Claude account and to Tella, then gave it my notes on what was wrong with the edit.
Then I stepped back. Muse took it from there:
Directed the work. It turned my critique into specific instructions and had Claude run a second editing pass in Tella.
Cleared the roadblocks. It handled sign-ins and stayed with Claude until the edits were done.
Checked the work. It compared Claude’s summary of what it changed against what actually happened in Tella, without touching the raw footage.
It’s the same setup you’d run on a job site. The super doesn’t frame the wall. He makes sure it got framed right.
One AI doing the work is a tool. One AI checking another’s work is a system.
The bigger point
Most people use one AI to plan, do, and review a whole project. That’s like asking your framer to inspect his own framing.
The better setup is a hierarchy. A heavy-duty model like Claude does the work, and a lighter one like Muse supervises it, catches what’s off, and sends it back until it meets the standard. You set the standard once and stop babysitting.
This is still an experiment. But it’s the first time I’ve watched an AI hold another AI accountable, and the edit came out better for it.
Steal this for your company
Think about the AI output you always end up fixing yourself: estimates, client emails, daily logs. Write down the three things you correct every time. That list is your QA checklist. Next time, give it to a second AI and have it check the first AI’s work before it reaches you.
Want to go deeper?
The weekly workshop also introduces our Contractor Operating System, a six-month cohort that builds six core systems into your business: Vision & Strategy, Finance & Reporting, Marketing & Sales, Production, Operations & Automation, and HR & Leadership. It starts January 7, 2027. We’re taking 20 companies, and only one per trade in each area, so you won’t sit next to your competition.
Join us live every Tuesday
Christene Marie of The Knowing Agency and I run a free live workshop every Tuesday for builders working through exactly this kind of operational mess. One real problem each week. Systems, AI, and growth. Bring your questions and we build the session around them.
Every Tuesday · 12:00–1:30 PM PT / 1:00–2:30 PM MT / 3:00–4:30 PM ET · Live Online
THE VALUE VAULT
SkillMatch — AI Estimate Tool
Answer 26 questions. Get a prioritized list of the AI skills that will have the biggest impact — personalized to your revenue, team, and project mix.
Beyond the Bid — YouTube
Every episode, uncut. Watch full conversations with the operators, consultants, and builders who've already figured it out.
Claude — AI Writing Tool
Draft review responses, client communications, and SOW language using your own project context.
CompanyCam — Job Site Photo App
Snap a photo and it’s filed by project and date automatically. Turn your photos into reports and customer updates in minutes. Beyond the Bid readers get 14 days free, then 25% off the first two months.
What’s the one AI output you always end up fixing yourself?
Grant
