Learn AI From the Ground Up

Step 1: Pick Your AI Tool

Go to claude.ai (Anthropic) or chatgpt.com (OpenAI). Both are free to start. Create an account with your email or Google login. That's it — no software to install, no credit card required.

Step 2: Start Simple

Don't overthink your first message. Type something you'd normally Google, but phrase it like you're talking to a helpful colleague. Instead of "datacenter cooling best practices," try: "I'm building a 10MW datacenter in Texas. What cooling approach should I consider given the climate?"

Step 3: Have a Back-and-Forth

The real power of AI is the conversation. When it gives you an answer, follow up: "Can you make that more specific to direct-to-chip cooling?" or "Give me that as a comparison table." The AI remembers everything in the conversation, so each reply builds on the last.

Step 4: Try These Starter Prompts

• "Summarize this email and draft a reply" (paste the email)
• "I have a meeting about [topic] tomorrow. Give me 5 smart questions to ask."
• "Explain [complex concept] like I'm explaining it to my team"
• "Write a professional email to a client about [situation]"

What to Expect

AI won't be perfect on the first try — and that's fine. Think of the first answer as a draft. Ask it to revise, expand, shorten, or change the tone. The more specific your feedback, the better the result. Most people get genuinely useful output within 2-3 exchanges.

What AI Actually Is (No Jargon)

AI — specifically the kind you're hearing about (large language models like Claude and ChatGPT) — is software that's been trained on enormous amounts of text. It can read, write, summarize, analyze, translate, and generate content. It's a very powerful assistant. It is not a sentient being, it doesn't "think" the way you do, and it cannot replace judgment, experience, or relationships.

What AI Is Good At

• Drafting: Emails, reports, proposals, SOPs — anything you stare at a blank page for
• Summarizing: Turn a 40-page document into a one-page brief in seconds
• Research: "What are the current ASHRAE guidelines for datacenter humidity?" — instant, sourced answers
• Repetitive tasks: Formatting data, generating checklists, creating templates
• Brainstorming: Need 10 ideas for a presentation? AI gives you 10 in 5 seconds

What AI Is Bad At

• Judgment calls: It doesn't know your client, your team, or your project history
• Physical work: It can write an inspection checklist — it can't do the inspection
• Accountability: It can draft a report, but someone still signs off on it
• Real-time data: It doesn't know what happened on your jobsite today (unless you tell it)

The Real Picture

AI doesn't replace people — it replaces the tedious parts of what people do. The project manager who uses AI to draft daily reports in 2 minutes instead of 30? They're not being replaced. They're getting 28 minutes back. Every. Single. Day. The people who will struggle are those who refuse to learn what AI can do — not because AI takes their job, but because someone who uses AI will simply be more productive.

Why Most Prompts Fail

The #1 mistake: being too vague. "Write me a report" gives you a generic report. "Write a 2-page executive summary of the HVAC commissioning results for Building C, highlighting the 3 failed AHUs and recommended corrective actions, for an audience of the owner's rep and GC" gives you something you can actually use.

The CRISP Framework

Use this structure for consistently good results:

• C — Context: Who are you? What's the situation? ("I'm a QAQC manager on a 50MW datacenter project")
• R — Role: Who should the AI be? ("Act as a senior mechanical engineer reviewing commissioning data")
• I — Instructions: What exactly do you want? ("Compare these two cooling system designs")
• S — Specifics: Format, length, tone, audience ("Create a comparison table, 1 page max, for a non-technical owner")
• P — Parameters: Constraints or requirements ("Use only data I provide — don't make assumptions")

Advanced Techniques

• Chain prompting: Break complex tasks into steps. "First, list all deficiencies. Then, prioritize them by impact. Then, draft the corrective action plan."
• Few-shot examples: Show the AI what you want by giving it an example. "Here's a sample RFI response I wrote. Write the next one in the same style."
• Negative instructions: Tell it what NOT to do. "Don't use jargon. Don't add filler. Don't make up data I didn't provide."
• Iteration: Treat the first output as a draft. "Make it shorter." "Add more technical detail." "Change the tone to be more direct."

Prompts You Can Steal Right Now

• "Rewrite this email to be more professional but keep it under 150 words"
• "I'm going to paste a spec section. Summarize the key requirements as a checklist I can use in the field"
• "Create a meeting agenda for a 30-minute commissioning kickoff with the MEP subcontractors"
• "Turn these bullet points into a client-ready narrative paragraph"

What Is an AI Agent?

An AI agent is an AI that doesn't just respond to one question — it takes a task and works through multiple steps to complete it. Think of it as the difference between asking someone a question and asking someone to handle a project. Agents can read emails, search documents, fill out forms, generate reports, and chain tasks together automatically.

Where to Start: Claude Projects

The easiest way to build something agent-like is with Claude Projects (claude.ai). Upload your company's templates, SOPs, and reference docs into a Project. Now every conversation in that Project has access to your materials. Ask Claude to "draft a punchlist report using our template and the deficiency photos I just uploaded" — and it does it in your format, every time.

Level Up: Zapier + AI

For true automation — things that run without you — use Zapier (zapier.com) with AI steps:

• Email triage: When a new email arrives, AI reads it, categorizes it (RFI, submittal, scheduling), and routes it to the right person
• Daily report generation: Pull data from your project management tool at 5 PM, have AI draft a daily report, email it to stakeholders
• Document processing: When a new spec PDF is uploaded to Google Drive, AI extracts key requirements and adds them to a tracking spreadsheet

Building Custom Agents with Claude Code

For teams ready to go deeper: Claude Code lets you build agents that interact with your own systems — reading from databases, calling APIs, generating documents with your branding. This is the professional tier, and it's where AI stops being a tool and starts being a team member. You'll need some technical support to set this up, but the workflow design — what the agent does and when — is something any operations person can define.

The 3 Rules of Good Agents

• Human in the loop: Start with agents that draft, not agents that send. Review outputs until you trust the system.
• Narrow scope: An agent that does one thing well beats an agent that does ten things poorly.
• Measurable value: Track time saved. If an agent saves 5 hours a week, that's $12,000+ a year in recovered productivity.

The Documentation Problem

QAQC professionals in datacenter and commercial construction spend 40-60% of their time on documentation — writing inspection reports, logging deficiencies, responding to RFIs, compiling closeout packages. The inspection itself takes 30 minutes. Writing it up takes two hours. AI flips that ratio.

Inspection Reports in Minutes

Here's the workflow that's changing QAQC on datacenter projects:

• In the field: Take photos, speak voice notes into your phone describing what you see
• Back at the trailer: Upload photos and transcribed notes to Claude or ChatGPT
• Prompt: "Generate a formal inspection report for the CDU piping installation on Level 2, Zone B. Use these photos and notes. Format per our company template. Flag any items that don't meet the approved submittals."
• Result: A complete, formatted report in 2-3 minutes. Review, adjust, sign.

Punchlist Management

Upload photos of deficiencies with brief descriptions. Ask AI to generate a formatted punchlist with location codes, responsible parties (based on the trade scope you provide), priority levels, and recommended corrective actions. AI can also cross-reference against previous punchlist items to flag recurring issues by trade or system.

RFI Responses

Paste the RFI question plus the relevant spec section. Ask AI to draft a response that references the correct specification clauses, applicable codes, and recommended resolution. An experienced QAQC engineer still reviews and signs — but the drafting that used to take 45 minutes takes 5.

Commissioning Documentation

AI excels at turning raw commissioning data (test results, TAB readings, functional test logs) into narrative reports. Upload the data, tell AI the system type and project context, and get a draft commissioning summary with pass/fail determinations, exception lists, and recommended retests — all referencing the correct ASHRAE standards.

The Jobsite Is Going Digital — Fast

In 2026, the construction teams winning work and delivering on time are the ones using AI as a daily tool — not as a future buzzword. You don't need to be a tech company. You just need to know which tools solve which problems.

Daily Reports & Logs

Tools like Claude and ChatGPT can generate daily construction reports from minimal input. Speak your notes into your phone at the end of the day, paste the transcription, and get a formatted daily report with weather, manpower counts, work completed, issues encountered, and next-day lookahead — all in your company's format. Teams doing this consistently save 30-45 minutes per day, per superintendent.

Schedule Analysis

Upload your P6 or MS Project export (CSV or PDF) and ask AI to identify critical path risks, flag activities with negative float, or summarize schedule impacts from a delay. AI won't replace your scheduler, but it can give your PM a plain-English summary of a 3,000-line schedule in 60 seconds.

Safety & Toolbox Talks

Generate tailored toolbox talks based on the actual work happening that week. "Create a 10-minute toolbox talk for electricians pulling wire in an active datacenter white space — focus on arc flash, lockout/tagout, and overhead hazards." Specific to your scope, your site, every time.

The Tools Worth Your Time in 2026

• Claude / ChatGPT: General-purpose AI for reports, emails, analysis, document review — the Swiss Army knife
• Procore + AI features: Photo documentation AI, automated RFI routing, quality observation logging
• OpenSpace / DroneDeploy: AI-powered visual progress tracking from 360 photos and drone captures
• Gamma: Turn project data into polished client presentations in minutes
• Zapier: Connect all your tools — when a punchlist item is logged, automatically notify the sub, update the tracker, and add it to the weekly report