43
Issue 3 | Fall Edition | 2026      BUILD MANITOBA
WCA EDUCATION SPOTLIGHT
GUARDRAILS MATTER
Garry encourages companies to 
establish clear guardrails before 
relying on AI-generated content. “We 
do not want made-up information,” 
he said.
Rather than allowing AI to fill in 
missing details, he instructs it 
to identify gaps, ask questions 
and insert placeholders when 
information is unavailable. Among 
his standing instructions:
	
Never fabricate information. 
	
Flag all gaps explicitly. 
	
Documents are legal records.
	
One topic per chat, similar to 
one email per subject / topic.
	
Ask before assuming. 
	
Do not override missing 
guardrail data with estimates.
These simple rules dramatically 
reduce the risk of AI “hallucinating” 
details that were never discussed.
PROTECT YOUR DATA – AND YOUR COMPANY
As AI becomes more common in 
construction, both presenters  
stress the importance of using it 
responsibly – especially when  
handling confidential information.
Before uploading drawings, contracts 
or other project documents, companies 
need to understand where that 
information is going. Public versions 
of AI tools pose the greatest privacy 
risk, while enterprise versions offer 
significantly stronger protections. Even 
AI won’t replace construction professionals, but it is changing how they work. Whether 
it’s organizing meeting notes, comparing subcontractor bids or drafting company 
policies, today’s most successful users aren’t asking AI to think for them – they’re 
asking it to handle repetitive work so they can focus on applying their own experience 
and judgment.  
LEE’S PROMPT CHEAT SHEET
Before asking Copilot, ChatGPT or another AI tool  
to tackle a task, give it five pieces of information:
1. ROLE
Tell it who it is.
“I am a project 
manager at 
a Manitoba 
construction 
management firm...”
2. GOAL
Tell it exactly what 
you need.
“Compare these two 
subcontractor quotes 
and flag every scope 
gap and exclusion.”
3. CONTEXT
Explain why you 
need it.
“I need to determine 
whether the low bid 
reflects missing scope 
before the pre-award 
meeting.”
4. SOURCE
Tell it what to  
work from.
“Use the two 
attached 
subcontractor  
quotes for the same 
project scope.”
5. EXPECTATIONS
Set the rules for  
the answer.
“Present the results in 
a table. Flag missing 
scope. Do not make 
assumptions or award 
recommendations.”
the webinar, she demonstrated how 
AI can draft a policy in minutes, 
covering everything from confidential 
information and acceptable use to 
meeting recordings and consent 
requirements for AI transcription tools.
AI can help build the policy, but it can’t 
own it. Every organization still needs 
legal review, management approval 
and employee buy-in before putting it 
into practice.
then, Garry cautions against uploading 
highly sensitive information. "If you 
wouldn't email it, don't put it into AI," 
he said. "If you definitely don't  
want the information shared in a  
court setting, don't upload it to  
AI. Ever. Period."
Lee encourages companies to take the 
next step by developing a formal AI 
policy that outlines approved software, 
acceptable uses, privacy expectations 
and employee responsibilities. During 

View this content as a flipbook by clicking here.