Record a Workflow for AI Implementation

Why recordings are useful

People describe workflows in a cleaner way than they actually happen.

They say, "I update the project board after the call." Then you watch the recording and see the real workflow:

1. Open Fireflies.
2. Search for the right transcript.
3. Copy notes into a Google Doc.
4. Check the Notion playbook.
5. Search Slack for a decision from last week.
6. Update ClickUp.
7. Draft a Slack status.
8. Leave two things out because they might change scope.

That is the workflow. Not the sentence.

A recording helps an AI implementation partner see the real handoff, the hidden judgment calls, and the tools involved. It also gives agents better context. A transcript or screen recording can be turned into an agent-ready workflow map much faster than a generic form response.

What to record

Record one workflow that feels repetitive, fragile, or annoying.

Good examples:

1. Turning a sales call into CRM notes.
2. Turning client meeting notes into project tasks.
3. Creating an invoice approval packet.
4. Summarizing support tickets into product feedback.
5. Updating a weekly status report.
6. Reconciling a spreadsheet with a finance tool.
7. Moving an approved request from Slack into a project board.
8. Creating an onboarding checklist from a customer intake.

Do not record a tour of every tool. Record a single handoff.

The recording script

Use this script while recording:

This workflow starts when...

The first place I go is...

The information I need is...

The part I copy or rewrite is...

The decision I make here is...

If this happens, I escalate to...

The final output should look like...

The output goes into...

Before I consider it done, I check...

You do not need to sound polished. The messy parts are the most useful parts.

What to show on screen

Show the actual tools if you can safely do so.

Useful screen details:

1. The source record or message.
2. The document, board, CRM, spreadsheet, or ticket where work happens.
3. The places you search for context.
4. The template or SOP you reference.
5. The field names that must be filled.
6. The final output.
7. The approval step.

If you cannot show real data, use a sanitized example. Keep the structure realistic.

What to say out loud

Narrate your thinking. This matters more than the clicks.

Say things like:

1. "I ignore this section because it is usually boilerplate."
2. "This is where I check whether it is a scope change."
3. "If the client mentions budget, I flag it for Rob."
4. "I only update this field after the project manager approves."
5. "This template is old, but we still use the second section."
6. "The task title has to match this naming convention."

Those sentences become decision rules. Decision rules are what make AI safer.

Audio-only recordings

Audio is useful when screen recording is not possible. Use audio to explain:

1. What starts the workflow.
2. Which systems are involved.
3. What gets copied between systems.
4. What judgment calls a person makes.
5. What output is expected.
6. Who reviews the output.
7. What would make the workflow risky.

Audio is not as precise as a screen recording, but it is still far better than "we need AI for ops."

How to protect sensitive information

Do not reveal:

1. Passwords.
2. API keys.
3. Bank information.
4. Private customer records.
5. Employee health, compensation, or HR details.
6. Legal documents unless explicitly approved.
7. Anything your company would not want in an implementation brief.

If the workflow involves sensitive data, describe the shape of the data instead of showing the data itself.

Example:

This column contains customer email addresses, so I am hiding the values. The important part is that I match the email to the CRM account and then update the renewal status.

That gives enough context without oversharing.

How an agent uses the recording

After the recording is submitted, an agent can turn it into:

1. A workflow summary.
2. A list of systems involved.
3. Inputs and outputs.
4. Decision rules.
5. Risk flags.
6. Missing context questions.
7. First proof of value.
8. A 7 day implementation plan.

Use the agent handoff prompt to process the recording transcript or summary.

The best recordings show friction

Do not hide the awkward parts. The awkward parts are where AI can help.

Good friction:

1. "I have to search three places."
2. "I always forget which template is current."
3. "The board field names do not match the intake form."
4. "I have to rewrite this for the client."
5. "Finance needs the same data in a different format."
6. "The team asks the same question every week."

That is implementation signal. A clean demo is less useful than an honest walkthrough.

Before you submit

Check that the recording answers:

1. What starts the workflow?
2. What tools are involved?
3. What does the human decide?
4. What should AI draft or automate?
5. What should stay under human review?
6. Where does the finished work go?
7. What examples prove the output is good?

If the recording answers those questions, it is useful.