The Core Mental Model
The Core Mental Model
You can understand almost any workflow with four ideas:
- Steps — the individual actions.
- Constants and variables — what stays the same vs. what changes each time.
- Composability — combining steps so they flow together.
- Named workflows — packaging the whole thing so you can reuse it.
Steps: the smallest repeatable actions
A step is one basic action you do every time.
Examples:
- Pull unread emails from the last 24 hours.
- Extract action items from meeting notes.
- Summarize a long document into five bullets.
- Flag anything related to a specific client.
- Reformat notes into a weekly update template.
A good step is: small, clear, repeatable, and easy to verify.
If you cannot explain what "done" looks like for a step, it is not yet clear enough to delegate.
Constants and variables: the template underneath your work
This is the key insight most people miss.
Every repeatable workflow has a template hiding inside it:
- Constants are the parts that stay the same every time: the structure, the quality criteria, the format, the rules.
- Variables are the parts that change: the specific client, the new data, the particular property, the date range.
Example — the realtor website workflow:
| Constants (same every time) | Variables (change each time) |
|---|---|
| Website template and layout | Property address |
| Design style and tone of voice | Interior/exterior photos |
| Sections: before/after, narrative, neighborhood | Listing price and details |
| Quality standard for images | Neighborhood data |
| Publishing process | Design visualization choices |
When you identify the constants and variables, you have parameterized your workflow — turned it into a reusable template. The constants become the instructions. The variables become the inputs you swap in each time.
This is the core act of building an AI workflow: extracting the template from your manual process.
Composability: chain steps so they flow
Composability means connecting steps so the output of one becomes the input of the next, without confusion.
Example sequence:
- Gather updates from email, chat, and project board.
- Filter for urgency and impact.
- Summarize into a short priority brief.
- Route top items into your daily plan.
Each step is simple. Together, they produce a high-value output you would have spent an hour assembling manually.
Named workflows: package it for reuse
Once your steps work together, give the workflow a name:
- Morning Briefing
- Weekly Risk Scan
- Client Prep Pack
- New Property Proposal
- Candidate Screening Report
You do not re-explain every detail each time. You invoke the workflow by name, provide the new inputs (variables), and review the output.
You have gone from doing a task to running a system.
Autonomy levels: not all steps are the same
A critical nuance: within any workflow, different steps deserve different levels of agent independence.
- Full autonomy steps: The agent can handle these without your input. Gathering data. Formatting. Applying your standard template.
- Guided steps: The agent proposes, you approve. Design choices. Tone of messaging. Priority ranking.
- Human-only steps: These require you. Final client communication. Strategic decisions. Anything with relationship or reputational risk.
When you design a workflow, label each step: agent handles, agent proposes / I approve, or I handle.
This keeps you in control without bottlenecking the mechanical work.
Quick exercise (10 minutes)
Pick one repetitive task from your last week:
- Name it.
- List the steps you did.
- For each step, identify: is this a constant (same every time) or does it involve variables (changes each time)?
- Label each step: full autonomy, propose-and-approve, or human-only.
- Give the whole thing a reusable name.
You just designed your first agent workflow — on paper.