Let’s talk

Agents

Work that needs judgement calls,
done independently and brought back to you

Why did Germany drop in September? Which keywords should we chase next? A Workflow can’t make those calls. An agent can.

For a whole function, such as SEO, an Agent Team splits the work between three to five agents and checks what comes back. Nothing goes out without your yes.

Book a Discovery Call30 minutes. Free.
See an Agent at work
A mechanical hand directing three figures on a small stage: a painter, a writer and a scientist
Image credit: ChatGPT

For open-ended work that needs
intelligence between steps

  • Traffic questions

    Find out why traffic moved: your own changes first, then tracking, then outside causes.

  • Search priorities

    Weigh rankings, demand and competitors’ pages, and bring back where to focus.

  • Campaign post-mortems

    Compare ads, site and sales results with past campaigns, and size each likely cause.

  • Competitor moves

    Read what changed on competitors’ pages and offers, and say what it means for you.

  • Customer signals

    Read reviews, tickets and survey answers about one question, and bring back the evidence.

  • Pipeline questions

    Look through the CRM for the deals or segments that changed, and explain what moved them.

When every step can be written down in advance, a Skill or a Workflow does the job for less. We tell you which.

Give it a goal.
Get the finished product.

Three jobs, and the path an agent took through each: what it did, what it left alone and what it brought back for your yes.

Goal: replace the spring ads that have gone staleExample

The steps it chose

  1. Found the tired ads

    6 of 24 ads have lost a third of their clicks since launch.

    Done

  2. Checked how often people saw them

    2 were shown too often to one audience: a budget call, not new ads.

    Left for you

  3. Read what is working

    Your best ads lead with the price, and carousels beat video.

    Done

  4. Wrote the replacements

    2 new versions for each of the 4 tired ads, in your voice, with claims cleared for each market.

    Done

  5. Built them in Canva

    8 ads from your brand kit, sized for each placement.

    Ready

What it brought back

8 new ads to replace 4 tired ones, built in Canva, with a change list: what to pause and what to upload.

Waiting for your yes
Pausing 4 ads and uploading 8
What it used
The ad account, Canva and your brand kit, the brain pages “Voice” and “Claims by market”
Left for you
The 2 over-shown ads: a budget call.
Review

Your team checks the ads and approves the change list

Your judgement · 20 min
Ready

New ads go live

Work moves forward
By hand
240 min
With an Agent
20 min
Time freed up per refresh
220 min

Agent Team
A whole function, run by agents

An Agent Team is three to five agents for one function, run by one command for one owner. Here is a full brand and content team, as we would build it in your AI assistant: five agents, one command, one head of brand.

The head of brand asks: launch the spring collection with a blog post, a launch email and a week of socialExample

Owner

Head of brand

Asks, then approves
The team’s command

Opens the Brain map, writes a brief for each agent, then checks and combines their work

Runs again any part that misses its brief
  1. First

    Content strategist

    Sets the angle, the audience and the mix of channels from the calendar and past results.

    ReadsCalendarAudiencesPast winners

    Works inNotion

  2. Side by side

    Copywriter

    Writes the blog post and the launch email in your voice.

    ReadsVoiceProducts

    Works inGoogle Drive

  3. Side by side

    Social editor

    Turns the story into a week of posts for each channel.

    ReadsVoiceChannel guide

    Works inGoogle Drive

  4. Side by side

    Designer

    Makes the visuals in Canva from your brand kit.

    ReadsBrand rules

    Works inCanva

  5. Last

    Brand guardian

    Checks every piece for voice, claims by market and brand rules, and sends back what misses.

    ReadsClaims by marketBrand rules

Ready

Launch pack: one blog post, one launch email, seven social posts and four visuals, checked against your brand

Nothing is published or scheduled until the head of brand says yes

  • One owner

    One person directs the team and approves its work. Agents for different roles or owners are separate Agents.

  • Three to five agents

    Each does one part of the function’s work, with the tools its part needs.

  • One shared context

    Every agent reads the same pages: your Team Brain, or a context page we write for the team.

Your agent,
briefed like a colleague

You keep four things, yours from the day it goes live.

  • The agent

    One plain file: its role, what it’s responsible for, how it works to your standards, the tools it may use and its rules. For an Agent Team, one per agent, plus the command that runs them.

  • The context

    One short page on your brand, products and audience, so the agent knows what your team knows. If you have a Team Brain, we check the pages it reads instead.

  • The try-outs

    Real questions from your team, each run twice. The owner reads how it worked as well as what it found, before it goes live.

  • Trained owners

    Training for the owner and your AI Ambassador: asking it well, checking its answers and keeping its file current.

Built around
your real work

We do the building. Your team brings the questions it asks today, the standards a good answer meets, and the final say.

  1. 01

    We define the role

    We scope the agent with its owner: the goal, the questions it takes, the tools and sources it may use, and its limits.

    From your team

    Bring the questions you ask today, and an answer you’d trust.

  2. 02

    We write the agent

    We write its file as a role in your team, in your words. For an Agent Team, each agent and the command that plans and checks their work.

    From your team

    Help us fill gaps in how the work should be done.

  3. 03

    We try it on real questions

    Each question runs twice. The owner reads its path and its answer, and we adjust the file.

    From your team

    Judge whether its answers are useful.

  4. 04

    We install it and train the owner

    We install it for everyone who will use it and train its owner and your AI Ambassador.

    From your team

    Name who owns it and who keeps it current.

Nothing goes out
without your yes.

An agent reads and drafts. Every send, publish or change to a record waits for a person’s yes. Once it’s live, you can lift that for one named action at a time, in writing.

Prepare

The agent brings back its answer and drafts

Agent at work
Approve

You approve or change it

Your judgement
Ready

Work goes out

Work moves forward

  • A named owner

    Every agent reports to one person. None runs a job with nobody named.

  • One AI Ambassador

    The one person who changes its file. It can be anyone on your team. Until you choose someone, it’s you.

  • 15 days of small adjustments

    From each Agent’s delivery, or from an Agent Team’s, at no cost.

  • Ongoing support

    Your team owns the agents from go-live. After the 15 days, changes and new tools or outputs come under the Fractional AI Operator.

    Explore the Fractional AI Operator

A few questions
you may have

Do we need an audit first?

No. If you know the question or the function you want covered, you can start with an Agent or an Agent Team. We work out the detail together in its scoping, at the start of the build.

When is a Skill or a Workflow better?

When you can write down every step before it runs. Then a Skill or a Workflow does the job, for less. We recommend an agent only where the steps depend on what it finds.

Explore Skills Explore Workflows
Do we need a Team Brain?

No. Without one, we write a short context page for the agent or the team. Several agents working from scattered knowledge are fragile, so where yours isn’t written down, we may advise a Team Brain first. That’s advice, never a condition.

Explore the Team Brain
Does it run on its own?

It runs when someone asks for it. Where the job needs it, it can also start on a schedule. Either way, every send, publish or change waits for a person’s yes.

Have you built agents before?

Yes, on our own work. We’re new to selling them, and we say so rather than borrow someone else’s results.

Which plan do we need?

A Claude Team or Enterprise plan, or ChatGPT Business or Enterprise. Every run uses your plan’s included usage, and we build the agent to fit within it. If you’re on individual plans, we show what the upgrade costs before you commit.

How do we judge whether it’s worth building?

By the hours it gives back. We check whether those hours, at what your team’s time costs, pay for the agent within six months. Where it does work your team doesn’t do today, we judge the likely business effect instead. These are estimates, not promises.

Who owns it?

You do, from the day it goes live.

Which question
keeps your team waiting?

Bring one question your team keeps chasing, or one function stretched thin, to a free, 30-minute Discovery Call. We’ll look at whether an Agent or an Agent Team fits. If one does, your proposal follows within one working day.

Book a Discovery Call30 minutes. Free. A conversation about your work.Explore the Team Brain
A woman in Victorian lace wearing a black glass visor, with yellow, mint and red reflections
Image credit: ChatGPT