This is the stage everything else exists to serve. Input flows in; Output flows
out; but in the middle sits the Transformation Agent — the thing that actually
does the work. In the old org chart this was a person you hired, trained, and
managed. In the agentic org chart it's a worker you design. Understanding what
that worker is made of is the difference between deploying agents and just hoping
a chatbot figures it out.
What the agent IS
A Transformation Agent is three things stacked together:
- The model — the reasoning engine, the raw intelligence that reads the input
and decides what to do. On its own it's a brilliant generalist with no hands
and no job description.
- Skills — the judgment you give it: the instructions, standards, examples,
and voice that turn a generalist into your specialist. A Skill is how you
encode "here's how we do this, specifically." It's the training you'd give a
hire, written once and applied every time.
- Tools — the hands: the ability to send an email, write to the database,
call an API, search the web, move a file. Skills are what it knows; Tools are
what it can touch.
Model plus Skills plus Tools — that stack is the worker. Strip any one and it's
not an employee, it's a demo. A model with no Tools can think but not act. Tools
with no Skills act but without your standards. The Architect's craft is composing
all three into a worker that does one job to your bar.
How an Architect builds it
You're not writing code so much as writing a job description that executes:
- Define the job. One agent, one clear responsibility. "Qualify inbound leads"
beats "handle sales," the same way a real role beats a vague mandate.
- Author the Skill. Hand it your criteria, your examples of good and bad, your
voice. This is where your expertise gets encoded and made repeatable.
- Grant the Tools. Give it exactly the access the job needs — CRM read, email
send — and nothing it doesn't.
Worked example
Take the support-ticket input from the last module. The Transformation Agent for
it is: a capable model; a Skill that holds your support voice, your refund
policy, and three worked examples of how you'd answer a tier-one question; and
Tools to read the customer's order history and draft a reply.
Now it reads each ticket, reasons over it with your policy, and produces a
response in your voice — the exact work a junior support hire would do, encoded
once and running every time a ticket arrives. You didn't hire. You designed the
worker.
Pro tip — let the agent help you compose the worker
You don't need to memorize every Skill or every Tool that exists, and you
definitely don't need to learn them in a specific order. The faster move:
- Grab the latest Skills primer at
fastframe.work/skills and the Tools primer
at fastframe.work/tools. Each one is a short
tour of what's in the toolbox today.
- Paste both into your agent, along with the context of the job you're trying
to do (one paragraph — what the worker is for, what it needs to touch,
what good output looks like).
- Let the agent help you (a) pick the right Tool for the job, (b) walk you
through getting the API keys and stashing them safely, and (c) author the
Skill to a high standard — or, if you're already inside Optimus, just
use one of the FAST Skills that ships with your membership. The hard
composition is already encoded; you're delegating the assembly itself.
That's the architect's leverage in one move: you stop being the person who
learns every tool, and start being the person who designs the worker that uses
them.