Workflow 009: The Role Builder

I've hired hundreds of virtual assistants over seven years of running an assistant agency.

The hardest part was almost never finding people. It was that clients couldn't describe who they needed.

A role builder is a skill that interviews you about the gap you're trying to fill, decides whether that gap needs a person or an agent, then writes the matching document: a job description if it's a human, an SOP an agent can be built from if it isn't. Mine runs on Claude and takes about twenty minutes of answering questions.

The decision comes before the document. That ordering is the whole workflow.

What this replaces

Job descriptions copied from the internet, and the interview where nobody can say what the job is.

The version I saw constantly: a client knows they're drowning, so they ask for an assistant. We hire someone excellent. Three months later it isn't working, because half the role was repetitive work nobody enjoyed and the other half was never clearly owned.

Nothing was wrong with the person. The role was never defined, so there was nothing to succeed at.

Blueprint

Answer questions about the gap. Get back a decision and a document.

Trigger: manual, whenever you're about to hire or about to automate. Input: what's falling through, what you're still doing yourself, what kind of person you imagine, and how the work fits with everyone else. Output: a hire-or-automate recommendation, plus either a job description with roles and responsibilities, or an SOP ready to become an agent.

I now run this before the interview rather than after. It takes twenty minutes and it changes what the interview is even for.

Organize

What has to exist first:

• Clear intake questions about the gap, not the job title
• Past role definitions that led to placements that actually worked
• Your SOP library, so a role can reference documented processes
• An honest list of what's currently falling on you

The intake is the part worth investing in. Most hiring briefs describe a wish: someone proactive, organized, good with people. Those words screen out nobody. The questions have to get at what the work actually is.

Systematize

You are my role builder. Before I hire anyone or build any agent, I run
this with you.

Interview me first. Ask a few questions at a time, covering:
- What is falling through right now, specifically
- What I'm still doing myself that I shouldn't be
- What the outcome of this role would be if it worked
- Which parts repeat, and how often
- Which parts touch other people: clients, team, candidates, partners
- Which parts need judgment, and what happens when the judgment is wrong
- What kind of person I imagine, including personality and working style
- What tools and access this work touches

Push back when my answers are vague. "Proactive" and "organized" are not
answers, ask what that looks like in the actual work.

Then make a recommendation, splitting the work outcome by outcome:

HIRE when the output is touched by other people and needs care and context,
or when judgment is required and being wrong is costly.

AUTOMATE when the work repeats and produces the same output every time,
with low judgment.

Say clearly which outcomes fall on each side, and why. If the role splits,
say so. Most do.

Then write the matching document.

If it's a human role, write:
- The outcomes this person owns
- Roles and responsibilities, specific enough for a recruiter to screen against
- Decisions they can make alone, and what they escalate
- The personality and working style that fits, tied to the actual work
- What good looks like at 90 days

If it's an agent, write it as an SOP: trigger, steps in order, decision
points, definition of done, and where a human reviews.

If the role splits, write both, and say which agents this person should
build once they're in the seat.

How to set it up

1. Save it as a skill.
2. Run it on a role you've hired for before, ideally one that didn't work out. The gaps will be obvious.
3. Correct the output and fold the corrections in.
4. Use it before every hiring conversation, not after.
5. Feed the human-role output to whoever screens candidates, and the agent output to your SOP builder.

Assign

What the AI does: run the interview, split the work, make the recommendation, write the document.

What stays human: the answers, and the final call.

The test I keep coming back to is simple. If the output is touched by other people and needs care and context, that's a person. If the work repeats and produces the same result every time, that's an agent.

Most roles split down that line, which is why running this before hiring matters. You end up hiring for what's left, and what's left is almost always the judgment part.

Here's the part that surprised me about doing this for clients.

Sometimes the answer is that they don't need to hire at all. And that's a fine outcome. They walk away with the definition either way, which is the thing they'd have needed before doing anything.

The version I like most is when the role splits and the new hire's first job includes building the agents for their own repetitive work. The assistant isn't competing with the agents. She builds them.

Rule of thumb from this build: define the outcomes, split them, then hire for what's left.

Integrate

The role builder sits in the middle of a chain that already exists:

• The intake questions come from what's falling through, often straight out of a brain dump
• The human-role output goes to whoever screens candidates, so they can call green light or red light against something specific
• The agent output goes to the SOP builder, which turns it into a file an agent gets built from
• Whatever the new person builds gets documented and folds back into the library

One workflow, two doors, and both of them lead somewhere.

Tomorrow: the onboarding skill. What happens in the first week after someone says yes.