Agenda
We'll start simple, build confidence, then peek around the corner
Foundations
How LLMs work, training stack
Better Results
Prompting, custom instructions
Tools
What's available today
Agents
Autonomous AI systems
AI Trends
What's coming next
BC Transit
Vision, governance, roadmap
How to Participate
Ask Along the Way
Feel free to interrupt with questions at any point during the session.
No question is too basic!
Questions appear on screen and get answered by our AI assistant in real-time.
What We're Not Here To Do
How the Garden Works
(or how the sausage is made)
How AI Responses Are Shaped
Seven layers influence every answer you receive
Vector Space: Transit Words
Similar concepts cluster together. Watch the relationships animate, or click any word.
Watching vector relationships...
Why this matters:
Words with similar meanings have similar vectors. This is how AI "understands" that subway ≈ metro ≈ underground, even though they're completely different strings of characters.
Vector Bias: Where AI Learns Stereotypes
Training data reflects societal biases. Watch how professions cluster with gender.
Watching bias patterns...
The problem:
"Doctor" is closer to "man" than "woman". "Nurse" is closer to "woman". This isn't truth - it's training data reflecting historical bias.
Why it matters:
AI reproduces and amplifies these biases in hiring tools, content generation, and decision support systems.
Getting Better Results
Practical tips for working with AI
Spend more time than you think you should with...
Explain the Constraints
Define the Output / Format
Be Clear with Instructions
Double Pro Tip: Prompt your AI to make your prompt!
Exercises to Try at Home
1. Basic Summarize
Open Copilot, drag in a document, and ask: "Summarize this document for me"
2. Summarize with Context
Context: "I am a director of ___ at BC Transit..."
Task: "Evaluate the document. Identify issues, risks, and opportunities..."
Format: "Create an executive summary with prioritized key points..."
Try both approaches and compare the results!
Custom Instructions
Example: My personal setup (available in most AI tools)
Identity & Context
Director of IT Business Services at BC Transit. Lead Privacy, Data & Analytics, Portfolio Management, Enterprise Architecture, AI Governance. ~50 staff, $34M portfolio.
Scale recommendations accordingly, not Fortune 500 solutions.
How I Work
- Technical background but focus on business strategy
- Audience: senior executives in public sector
- Value directness, precision, structured output
Response Style
- Be practical. Minimal fluff. Canadian English.
- Known problems: respond succinctly
- Open problems: ask clarifying questions
- Push back on flawed reasoning
Output Defaults
- Tables for comparisons, prose for narratives
- Start reviews with critical issues first
- "Draft" means 80% ready, not skeleton
Domain Calibration
- Don't suggest "consult privacy" when I am the authority
- Consider public sector procurement constraints
- Avoid corporate buzzword density
Key Takeaway
The more context you give AI about who you are, how you work, and what you need, the better the results.
This is available in ChatGPT, Claude, Copilot, and Gemini.
BC Transit Style Guide
For branded content generation - copy/paste into your custom instructions
Primary
Secondary
Neutrals
BC Transit Style Guide (for branded content only)
Primary Colours:
Blue RGB(9,62,113) | Green RGB(54,181,79)
Secondary Colours:
Magenta-pink RGB(212,15,139) | Orange RGB(243,143,30)
Red RGB(217,26,51) | Cyan RGB(3,171,206)
Neutrals:
Cool Gray RGB(122,135,142) | White RGB(253,253,253)
Typography:
Helvetica Neue (primary), Myriad Pro (alt),
Arial/Segoe UI (fallback)
General Tips & Tricks
Analyze & Extrapolate
"Analyze this spreadsheet and give me trends. What patterns do you see? Extrapolate for next quarter."
Visualize Data
"Create an infographic based on this data. Visualize the key metrics in a way executives would understand." Use JSON for better results.
Meta-Prompt
Ask AI to write your prompt. "Help me write a prompt that will get you to summarize legal documents effectively."
Use Templates
Provide a template or example output. "Write a status report following this format: [paste example]"
Chain Your Tools
Take output from one AI and pass to another. Summarize in Copilot → refine in Claude → visualize in Gemini.
Summarize Long Chats
When context gets full, ask: "Summarize our conversation so far" - then start a new chat with that summary.
Tools
What's available today
Mainstream Commercial Tools
Because I'm Asked Regularly "What Should I Use?"
Microsoft Copilot
- Integrated with M365
- Uses OpenAI GPT
- BCT approved for confidential data
OpenAI ChatGPT
- Synonymous with AI
- Good all around
- Most obedient model
Google Gemini
- Strong multimodal capabilities
- Deep Google integration
- Good for research tasks
Anthropic Claude
- Strongest reasoning
- Best for long documents
- Most "thoughtful" responses
What Can You Use Where?
Data sensitivity determines which tools you can use
Microsoft Copilot
BCT approved for confidential data
- Internal documents and reports
- Budget and financial discussions
- Strategic planning content
- Employee-related discussions
- Personal information (PI) - avoid or anonymize when possible
ChatGPT, Claude, Gemini, etc.
Public/non-sensitive use only
- General research and learning
- Public information queries
- Personal productivity (non-work)
- Code examples and technical help
- No confidential, sensitive, or personal data
When in doubt, use Copilot - but always minimize personal information exposure.
What Can Copilot Do Today?
Copilot Chat (Free) vs Copilot Full (Licensed)
Copilot Chat (Everyone has this)
- General Q&A and brainstorming
- Summarize documents you upload
- Draft emails and content
- Web search and research
- Analyze data you paste in
Copilot Full (Licensed users)
- Everything in Chat, plus...
- Teams: Meeting summaries, catch-up on missed meetings
- Outlook: Draft replies, summarize threads
- Word: Generate docs, rewrite sections
- Excel: Analyze data, create formulas
- PowerPoint: Generate slides from prompts
Agents
I Keep Hearing About Agents...
What is an AI Agent?
An AI agent is an AI system that can take a goal, break it into steps, use tools or information, and keep working until the task is done.
Example: Agentic Fire Detection
An AI that monitors, reasons, and acts autonomously
Camera Feeds
Weather Data
News & Events
AI Agent
Monitors, reasons,
decides action
The agent pulls in multiple signals, applies rules and context, and decides whether to act.
Decisions are contextual, not hard-coded rules.
AI Trends
What's coming next in AI
What's Happening Right Now
February 2026 - The frontier model race continues
GPT-5.2
400K context window, unified routing that auto-adjusts reasoning depth, 100% AIME math score
Claude Opus 4.5
First model above 80% on SWE-bench coding, 1M token context, strongest long-document reasoning
Gemini 3 Pro
3x faster than Gemini 2.5, 60-70% cost savings, embedded across all Google Workspace
The era of "just make it bigger" is over - smart beats big in 2026
Claude Cowork
January 2026 - Anthropic's "SaaSpocalypse" moment
What is it?
- AI agent that works directly on your computer
- Reads, edits, and creates files in folders you choose
- Multi-step task execution with planning
- Integrates with Slack, Figma, Canva, Box, Salesforce
- Claude Code capabilities without the terminal
Why it matters
- AI moves from "chat assistant" to "digital coworker"
- Completes work autonomously, loops you in on progress
- $285B market impact - disrupting legacy SaaS tools
- Sets new bar for what "AI at work" means
Available now on Claude Desktop (Pro plan) - Windows & macOS
140K+ GitHub stars - Viral in late Jan 2026
Connects to WhatsApp, Slack, Discord, iMessage, Teams
Feb 14, 2026: Creator hired by OpenAI
Project moving to open-source foundation
Browse web, read/write files, run commands - "The gap between imagination and reality has never been smaller"
Key Industry Shifts
Reasoning-First Models
Models now "think" before answering. Trade speed for accuracy on complex problems.
Agentic Workflows
AI that plans, executes, and iterates autonomously. The biggest trend of 2026.
Efficiency Over Scale
GPT-4 level performance at 10% of the cost. Running out of training data forces innovation.
Enterprise AI Adoption
AI has moved from experimentation to early mainstream
of OECD firms using AI in 2025 - more than double since 2023
of companies using "physical AI" (robots, automation) - projected 80% within 2 years
adoption at large enterprises vs ~17% at small firms - gap creates opportunity
Mid-market firms now deploying AI faster than large enterprises due to agility
Source: Global AI Adoption Index 2026, Deloitte State of AI in the Enterprise 2026
AI Enters Regulated Industries
Healthcare, finance, and government - new frontiers with new risks
ChatGPT Health (OpenAI)
Health-oriented AI for triage, patient education, care navigation
- Signals push into regulated domains
- Raises liability and safety questions
- Potential "digital front door" for healthcare
EU AI Regulation Package
Omnibus package harmonizing AI governance
- Clearer rules for EU operations
- Global firms adopt EU standards as baseline
- Higher bar for transparency & oversight
Execution risk (data, governance, talent) now more pressing than model capability
What's Next for BC Transit?
Vision, governance, and roadmap
AI Vision
Reimagining the future of transit with AI by empowering our people, improving how we serve riders, and creating a smarter, safer, more connected experience for the communities we serve.
AI Mission
We use AI to enhance public services, strengthen our teams, and deliver better outcomes for riders. Our mission is to apply AI in ways that improve efficiency and safety, enable more personalized and responsive services, and uphold strong ethical and responsible practices.
AI Centre of Excellence
BC Transit's single governance and standards body for AI
2026 Priority Areas
Policy Framework
Acceptable use, data handling, vendor assessment
Opportunity Intake
Clear pathway from idea to deployment
Agent Governance
Guardrails before agentic AI arrives
AI Playbook
Reusable patterns, prompt templates, training
Core members: IT, Security, Privacy, Procurement, People & Culture + Business Area Representatives
"Full" Copilot Roadmap
(high level)