The 10-Prompt Rule: Why AI Adoption in Construction Is a Training Problem, Not a Tech Problem
Trunk Tools has usage data from thousands of construction workers, and it points to one brutally simple threshold: users who write 10 prompts in their first week stay forever, and users who don’t churn. That finding has reshaped how CEO Dr. Sarah Buchner sells. She now refuses to close a deal unless the customer lets her team train their people, because in her words, real AI is far more about change management than picking the right technology.
Ask people why AI disappoints on construction projects and you’ll hear about hallucinations, accuracy, and hype. Sarah Buchner hears something different in her data. Trunk Tools, which raised a $40 million Series B led by Insight Partners last year, has watched thousands of field and office users interact with its text-based tools, and the pattern is remarkably clean. Ten prompts in week one, and the person becomes a user for good. Fewer than that, and they quietly drop off.
On the latest Bricks, Bucks & Bytes, Buchner explained what that cutoff taught her about why so many AI rollouts stall, and why the fix has almost nothing to do with model quality. For any contractor budgeting an AI purchase this year, her takeaway moves money from the software line to the training line.
The 10-prompt cliff
Getting over the hump of new and scary
The threshold isn’t magic, it’s psychology. “People have to get over the hump of, this is new, this is scary, I don’t like this,” Buchner said. Ten prompts is roughly the amount of use it takes for someone to see the tool answer real questions about their real project. Below that, the discomfort wins and the tab never gets opened again. Above it, the habit forms and holds.
Notice what the threshold is not: a measure of the software. The same product, the same models, the same project data produce opposite outcomes depending entirely on whether a person pushed through their first week. Which is why Buchner keeps repeating a line that should be pinned above every ConTech procurement desk: real AI is way more about how you do change management than about picking the right technology.
Owen’s analogy on the show lands well. Hand someone Excel with no instruction and they’ll manage a few basic formulas, then plateau far below what the tool can do. The software was never the constraint. The same dynamic showed up in BuildOps’ survey of 600 trade contractors, where training, not fear, emerged as the main barrier to faster AI adoption.
Refusing to sell without training
Churn insurance, disguised as principle
The 10-prompt finding pushed Buchner to an unusual commercial position: Trunk Tools won’t sell its platform unless the customer agrees to let her team train their people. Part of that is affection for the industry. The other part is candidly self-interested. “Churn would eat me up alive if I didn’t force our customers to let me train the people.”
The training itself is hands-on to a degree she never expected to fund. Forward-deployed agent builders. People physically in the field walking superintendents through outputs. “That’s not AI, that’s not sexy either. It’s literally just completely necessary for our industry.” The trust-building move is always the same: show the superintendent a submittal the agent processed, an RFI it wrote, a drawing discrepancy it caught, and let them react. Once they see it, she said, the response is usually some version of “this is really good,” and usage follows.
A lot of startups come out making crazy claims, dump software, and then leave. That’s just not how you implement change management in construction.Dr. Sarah Buchner, CEO, Trunk Tools
The adoption playbooks that actually survive the jobsite
Field-tested lessons from the operators rolling out AI across construction, weekly.
Join 3000+ ReadersTrain now, switch vendors later
Why the workforce is the durable asset
Buchner’s pitch to hesitant executives has an unusual escape hatch built in. Commit now, let her train the whole organization over the next couple of years, and if a bigger provider builds something better, switch. “You can switch within a day, because it’s all the same interface.” Agentic products are simple on the surface; most of the machinery runs in the background.
The asymmetry is the argument. Software is swappable in a day. A workforce that thinks AI-first, that knows how to collaborate with an agent, check its output, and fold it into a workflow, takes years to build. Firms waiting to identify the winning vendor before training anyone are protecting the cheap, replaceable asset while neglecting the expensive, durable one. That’s the gap she sees widening in the market right now, and it maps onto the pattern we covered in why customer readiness, not vendor velocity, sets the pace of ConTech AI.
What sticking looks like versus stalling
A rollout checklist drawn from the episode
Pull the threads together and the difference between AI rollouts that stick and the ones that stall looks less like a technology gap and more like an operating decision made before the contract is signed.
| Dimension | Rollouts that stick | Rollouts that stall |
|---|---|---|
| First-week usage | Every user pushed past 10 prompts with support | Login credentials emailed, usage left to chance |
| Training | Hands-on, in the field, mandatory | A webinar link and a PDF |
| Trust building | Real outputs from the user’s own project shown early | Generic demos from sales decks |
| Vendor role | Forward-deployed teams through rollout | Software dumped, vendor gone |
| Executive framing | Change management program with a tool attached | Procurement exercise with a login attached |
It’s a retention threshold Trunk Tools observed across thousands of users of its text-based AI tools. People who write at least 10 prompts in their first week keep using the product long term; people who don’t almost always churn. Buchner treats it as the point where a new user has seen enough real value to get past the discomfort of a new tool. (Source)
Two reasons, by Buchner’s own account. Untrained users don’t know how to work with agents or assess their output, so the deployment falls flat, and failed deployments become churn that damages the vendor. Making training a condition of sale protects the customer’s rollout and Trunk Tools’ revenue at the same time. (Source)
Less than assumed. BuildOps research covered on Bricks & Bytes found 78% of trade contractors already use AI in some form, and the biggest blocker to going further was training rather than fear. Buchner’s field experience matches: once superintendents see an agent’s output on their own project, skepticism tends to flip quickly. (Source)
Trunk Tools staff who work directly inside customer organizations, building agents around specific workflows and training users in the field. Buchner admitted she never expected to run hands-on training at this scale, but considers it necessary infrastructure for AI in construction rather than a temporary crutch. (Source)
Buchner argues the risk is smaller than it looks because agentic interfaces are simple and largely interchangeable; her claim is that a trained organization could swap providers within a day. The costly mistake is the opposite one: waiting for a clear winner and leaving the workforce untrained, which delays the gains and deepens the skills gap. (Source)
The episode’s implicit answer: weight it toward people, not licenses. Fund structured first-week onboarding designed to get every user past 10 prompts, insist on vendor-led field training as a contract term, and measure early usage rather than seat counts. Software that nobody prompts is the most expensive kind. (Source)
Related Articles
78% of Trade Contractors Use AI: Adoption Is Way Ahead of What You Think
“You Can Only Go As Fast As Your Customers”: The Quiet Brake on ConTech AI
