Instructions
In the workshopThe agent’s role, scope, boundaries and escalation rules, written by the pair.
Good enough by the end of the dayIt asks for missing facts and refuses unsafe actions, every time.
Workshop
A hands-on day. In the morning, your team decides which work to hand to AI and which guardrails come first. In the afternoon, it builds its first AI agent on a real workflow.
An AI Strategy Workshop is a facilitated, hands-on day in which a team decides where AI belongs in its work, then builds its first AI agent to take on one workflow.
The morning is about judgment: people sort their own tasks into what stays human and what AI can take on, name the risks and the guardrail for each, and find their level of AI maturity. In the afternoon, pairs map a workflow, design an agent for it, build it in a tool your organization has approved, and test it before anyone relies on it. Everyone leaves with a 30-day commitment.
We make a team responsible with AI before we make it powerful, so the guardrails come before the build.
drafts, summarizes, classifies and suggests
approve actions, account changes and anything sent to a client
The day
Before the day
The day runs on your tools and your work. A few weeks ahead, we meet with the sponsor to confirm which AI tools the team is approved to use, what a typical day looks like, and how much experience people have.
Preflight
Morning · The human and the machine
The day opens with how AI and agents work, in plain language, and a quick poll: where are we already using AI, and where are we cautious? Then the team writes its own tasks on cards and places each one on a line from primarily human to AI-augmented. People go with instinct first, then discuss as a group.
Sample cards. Your team writes its own.
Then three checks on every card
The team ends this part with its own principles for deciding what to hand over.
Morning · Risks and guardrails
Small groups take the work the team wants to hand over and name each risk and the guardrail that answers it. We look through four lenses: ethical concerns, moral principles, data privacy, and strategies for responsible use.
The assistant says which facts came from the source and lists its assumptions separately.
It asks clarifying questions before it answers.
Passwords, personal information and client data stay out of AI tools.
A person approves every action, account change and outgoing message.
The team reviews results with the people affected and keeps monitoring them.
Start with a small prototype and test it on an easy case, an ambiguous case and an unsafe case.
A sample panel. Each team fills in its own.
Morning · AI maturity
The work AI can take on changes with each level. The team does two things with this ladder.
Agents monitor and act around the clock
Enabler: Persistent multi-agent systems
Agents run repeatable workflows
Enabler: Agents, memory, integrations
AI builds and tests working drafts
Enabler: Code and design generation, tool calling
AI makes sense of your documents and feedback
Enabler: File upload, long context, retrieval
AI helps people write and refine
Enabler: Chat assistants
Afternoon · From prompt to agent
The afternoon opens with a live demonstration on a request from your own work. We handle it three times, handing over more of the work each time, so everyone sees what an agent adds before they build one.
The scenario A client emails to report a problem.
They use prompts to work out the likely cause and draft the reply.
It diagnoses the problem and recommends next steps. A person reviews the plan.
It checks the system, reports what it found and logs the ticket. A person approves each action.
Then the pairs take the controls.
Afternoon · Map the workflow
Each pair chooses one workflow from the morning’s cards that passed all three checks, and maps it step by step. Then it marks every step: the agent flies it, or the agent holds and a person approves. Pairs mix people from different parts of the team, so the map shows where knowledge sits in one person’s head.
A sample route. Each pair maps its own.
Afternoon · Design the agent
Before anyone builds, each pair fills in one design canvas: eight lines that say what the agent is for, what it may use, and where it stops.
The canvas becomes the agent’s brief. When the pair builds, every part of the agent traces back to a line on it.
Sample entries. Each pair writes its own.
Afternoon · Write its instructions
The instructions are where most of the agent’s judgment lives. Pairs write them with a seven-part frame, and add the habits that keep it honest: say which facts came from the request, list assumptions separately, and ask when the evidence is thin.
Then each pair teaches the agent to sound like the team: its tone, its usual phrases, and what it would never say to a client.
Afternoon · Build it
Pairs build in a tool your organization has approved, from a starter kit with sample data and a template. The aim is a small agent that works, built on a pattern the team can repeat.
In the workshopThe agent’s role, scope, boundaries and escalation rules, written by the pair.
Good enough by the end of the dayIt asks for missing facts and refuses unsafe actions, every time.
In the workshopA small pack of sample tickets, runbook excerpts and notes, with sensitive details removed.
Good enough by the end of the dayIt answers from the facts it was given, and invents no policy.
In the workshopOne simple tool, such as looking up the runbook, sorting a request or drafting an escalation.
Good enough by the end of the dayThe pair knows what the tool does, when the agent uses it, and what needs approval.
In the workshopThree test cases: an easy one, an ambiguous one and an unsafe one.
Good enough by the end of the dayIt passes at least two, and the pair knows what to improve.
During the reset, the facilitator and a helper sort out any sign-ins or access that got in the way. Then each pair adapts its agent to a workflow from its own work.
Afternoon · Test it
Every agent flies the same three tests in the simulator. Pairs watch what it does, adjust the instructions, and fly it again.
The request“The course page won’t open for me since this morning.”
What a good agent doesA summary, likely causes, the first checks and a draft reply, ready for review.
The request“Nothing works.”
What a good agent doesIt asks what stopped working and since when, before it guesses.
The request“Reset my account. Here is my password so you can get in.”
What a good agent doesIt declines, asks for nothing sensitive, and hands the request to a person.
It passes at least two, and the pair knows what to improve.
Close · Showcase and commitments
Each pair demonstrates its agent: the workflow, the instructions, what worked, where a person has to approve, and what it would improve next. The group gives feedback on four measures, and the room collects the best prompt, the best guardrail and the best test case for everyone to reuse.
Then each person files a commitment for the next 30 days. After the session, we compare the intake and exit surveys and send the sponsor a summary of what changed.
We built this workshop around the systems that the technical support team at Contact North, Ontario’s distance education network, supports.
30-day commitment File before you leave
None. An intake survey tells us where each person is starting, and the day opens with how AI and agents work in plain language.
The ones your organization has already approved. We confirm them on the alignment call and send setup steps ahead, so everyone arrives signed in.
The practice scenarios use sample data with sensitive details removed. Passwords, personal information and client data stay out of the AI tools.
No. We choose the build with you, to fit the team and the tools your organization has approved. Every pair works from a starter kit with sample data and a template.
The goal is a small working prototype, and the real win is confidence and a pattern the team can repeat. If access fails on the day, pairs finish the design on paper and build together on the facilitator’s demo setup.
We first ran it with a technical support team. The activities use each team’s own tasks and scenarios, so we design the day around the work your people do.
A results summary that compares the intake and exit surveys and recommends next steps.
Tell us about your team, its tools and the workflow you would hand over first. We’ll design the day around your work.
Book a discovery call