aioperants
Book a call
// ai agents for small business

Put AI to useful work.

We help you choose the right AI tools and set up agents that do real jobs in your business. Each agent gets a role, the skills and tools to do it, the knowledge it needs, and clear rules for when to ask you.

Start your AI brief →What is an agent?

We are independent. We don’t build AI models; we help you adopt the ones that fit your work.

Role

Booking assistant

Answers availability questions, books and reschedules, and sends reminders so fewer people miss appointments.

Today
Skills
Answer availabilityBook and rescheduleRemindersWaitlist offers
Tools
Booking systemWhatsApp BusinessSMS and email
Knowledge
Services and durationsOpening hoursCancellation policy
Asks you first
RefundsWaiving a deposit
  1. WhatsApp: “Any slot Saturday?”Done
  2. Offered 11:30 or 14:00Done
  3. Booked 11:30, deposit link sentDone
  4. Cancellation: refund askedNeeds you
  5. 12 reminders sent for tomorrowDone
A day in the roleSteps marked “Needs you” wait for your approval.
// how we work with AI
  1. firstTransparency

    You see what each agent can access, what it did and what it cost. No black boxes.

  2. → thenEfficiency

    Agents take on the repetitive work, measured against how long it took before.

  3. → alwaysHumans above the tech

    People set the rules, approve what matters and make the decisions. The agent works for your team.

// what an agent is

Three kinds of AI help. Only one does the job for you.

Most businesses already use a chat assistant. An agent is different: you give it a goal and it works through the steps, using your tools, until the job is done or it needs a decision from you.

e.g. ChatGPT, Claude, Gemini in a browser

Chat assistant

You ask, it answers. It drafts, summarises and explains, but you copy the result and do the work yourself.

You do: every step

e.g. Zapier, Make, n8n

Automation

Fixed rules: when this happens, do that. Reliable for steps that never change, and it breaks when something unexpected arrives.

You do: handle every exception

e.g. Claude Cowork, ChatGPT workspace agents, Hermes

Agent

Given a goal, it plans the steps, reads and writes in your tools, and adapts to what it finds. It stops and asks when a step needs your approval.

You do: approve, review, decide

  1. 1Role

    The job it holds, written like a job description. “Booking assistant”, not a mascot with a name.

  2. 2Skills

    Step-by-step instructions for each task, written once and reused, in your way of working.

  3. 3Tools

    The apps it may use: email, calendar, CRM, accounting, booking. Only with the access the job needs.

  4. 4Knowledge

    Your prices, policies, standard answers and past work, so replies sound like your business.

  5. 5Rules

    What it may do alone, what needs your approval, and when to hand over to a person.

We set up all five. The model inside matters less than most people think; a well-defined role with good skills and the right access is what makes an agent useful.

// the platforms

What the major platforms offer small businesses.

Each platform has strengths. The right one depends on the email and files you already use, how much you want to host yourself, and the jobs you want done. These products change monthly, and we recheck before every recommendation.

// anthropic · cowork and small business workflows

Claude

Claude is a strong writer and analyst, and Claude Cowork lets it work on files and across apps on your computer. In May 2026 Anthropic released Claude for Small Business, a package that switches on inside Cowork.

Best for
Owners and teams who want an assistant that does multi-step office work: finance, sales follow-up, documents.
What it can do
  • Connectors to QuickBooks, PayPal, HubSpot, Canva, Docusign, Google Workspace and Microsoft 365
  • Ready-made workflows such as invoice chasing, month-end close, cash-flow forecasting and campaign prep
  • Skills: reusable instructions for your own procedures, in an open format
  • Asks before anything is sent, posted or paid
Where it runs
Anthropic’s cloud, with the Claude desktop app for Cowork.
How you pay
Per-user monthly subscription. Team and Enterprise plans don’t use your data for training by default.
Watch for
Connectors cover popular tools; industry software usually needs a custom connection.

Checked 11 Oct 2026 · Source: Anthropic: Claude for Small Business ↗

Not sure which fits?

That is the first thing we answer. Most small businesses start with the assistant that works best with their email and files, then add agents for specific roles.

Ask us in your brief
// roles we set up

Agents with a job title. Not a mascot.

We define each agent the way you would hire for a position: what it is responsible for, what it can touch, and what it must check with you. Start with one role; add others when the first one earns its place.

// role

Administrative assistant

Sorts the inbox, drafts replies, files documents and keeps the calendar in order.

  • Triage email by urgency
  • Draft replies in your tone
  • File attachments
  • Daily summary

Asks you first: Replies to new contacts, Deleting anything

// role

Lead generator

Researches prospects that match your best customers, qualifies them and drafts a first message.

  • Research a company
  • Score fit against your criteria
  • Draft first outreach
  • Log in CRM

Asks you first: Every first message, Adding to mailing lists

// role

Analyst

Pulls your weekly numbers from sales and accounting and explains what changed and why.

  • Weekly sales report
  • Margin by service
  • Spot unusual changes
  • Cash-flow outlook

Asks you first: nothing, it only reads

// role

CRM expert

Keeps customer records clean, logs every email and call, and makes sure no follow-up is missed.

  • Log emails and calls
  • Update deal stages
  • Merge duplicates
  • Follow-up reminders

Asks you first: Merging or deleting records

// role

Booking assistant

Answers availability questions, books and reschedules, and sends reminders so fewer people miss appointments.

  • Answer availability
  • Book and reschedule
  • Reminders
  • Waitlist offers

Asks you first: Refunds, Waiving a deposit

// your role

A role of your own

Quote follow-ups, supplier orders, document checks, social posts. If you can explain the job to a new hire, we can usually define it as a role.

Describe it in the brief

Where agents are not the answer

  • Decisions that need your judgement, like pricing a large job or handling a complaint.
  • Work with no clear right answer that you could explain to a new hire.
  • Tasks that happen a few times a year. A checklist is cheaper.
  • Steps that never change. A simple automation does those for less.
// what is possible

up to 80%less time on the right tasks, with a large lift in what your team gets done. When it is implemented correctly.

Anthropic’s study of 100,000 real tasks estimated AI cut the time per task by about 80%. Controlled workplace trials found smaller gains, from 14% to 56%. Results depend on the task and on how the change is introduced. Read the study ↗

// implemented correctly

The technology is the easy part. People make it work.

Most AI projects that disappoint fail on adoption, not on the model. We treat every agent as a change in how your team works, and plan it that way.

The right task
Repetitive, rule-based work with a clear “done”. We measure how long it takes today before we start.
Your team involved
The people who do the job today help write the agent’s skills and rules. They know the exceptions.
Training
Hands-on sessions on your real work: how to review, correct and adjust the agent.
Gradual trust
Every step approved at first. Approvals are removed only where the agent has proved reliable.
A named owner
One person in your team owns each agent, its rules and its results.
// how we work

One role first. Measured before it grows.

We start with the job that costs you the most time, prove it works on your real tasks, and only then expand.

  1. 01

    Brief

    Your business, where time goes, your tools and your rules, collected in one place.

    You get: a shared AI brief
  2. 02

    Assess and choose

    We map the work and compare platforms for your tasks, data and budget.

    You get: a recommendation with running costs
  3. 03

    Build the role

    Skills written, tools connected, knowledge loaded, approval rules set.

    You get: a working agent in a test setup
  4. 04

    Pilot

    It runs on real work with you approving each step, then fewer steps as trust builds.

    You get: time saved, errors, costs, measured
  5. 05

    Train, measure, support

    Your team learns to work with it. We track results and running costs, and keep it current as tools change.

    You get: a monthly performance and cost report

Tool selection

Independent advice on Claude, Gemini, ChatGPT and open-source agents for your work.

Custom skills

Your procedures written as reusable skills, in the open format most platforms now read.

Automations and connections

CRM, booking, accounting and messaging linked so the agent can act, with access limited to the job.

Deployment

On the vendor’s cloud, or self-hosted on a server you own when data needs to stay with you.

Change management and training

Your team prepared, trained and involved, so the agent is used rather than worked around.

Performance and cost monitoring

Time saved, quality and token spend tracked every month, with budgets and alerts.

// measured, not assumed

You see what it saves. And what it costs.

Every agent gets a baseline before it starts and a monthly report after. If the numbers don’t justify it, we change the setup or switch it off.

Booking assistant · SeptemberDelivered monthly
Team hours saved
11.5 of 14 h/week before
Tasks completed
412
Approved without changes
94%
Handed to a person
23 as the rules require
AI usage (tokens) this month$38 of $60 budget
Alert at 80%. The agent pauses non-urgent work at 100%, never customer replies.
// what is a token?

The unit AI is billed in.

AI models read and write text in small pieces called tokens: a short word, part of a longer word, or a punctuation mark. Providers charge per token, for everything the agent reads (your instructions, its knowledge, the customer’s message) and everything it writes.

Can I book a cleaning on Saturday?

9 tokens. Each model splits text slightly differently.

3.5–4
characters of English per token
130–160
tokens for 100 words
~2,000
pages of text in a million tokens

Prices are set per million tokens, and writing costs more than reading. A larger model costs more per token than a smaller one, so we use the smallest model that does each job well, and set a monthly budget with alerts.

// ai brief

Tell us where the time goes. We’ll design the role.

Eight short steps, about 10 minutes. Skip anything you are unsure of; we’ll cover it on the call. As you answer, your role card takes shape. Answers are saved in this browser.

Start your AI brief →

Opens full screen. About 10 minutes; your answers are saved in this browser.

// before we start

Common questions.

Can you just train our team, without building an agent?

Yes. Training and tool selection work as a standalone engagement, shaped around your team’s real tasks.

Will an agent send messages or spend money on its own?

Only what you allow. We set which actions run alone and which wait for your approval. Sending to new contacts, payments and deletions start as approval-only.

What happens to our business data?

We review each provider’s data, retention and training terms for your plan before connecting anything, and give the agent access only to what its role needs. Where data must stay with you, we can self-host.

Do we need new software?

Usually not. Agents work inside the tools you already use. If a tool has no way to connect, we tell you during the assessment.

What does it cost to run?

AI subscriptions and usage are billed by the provider, in your name. Usage is counted in tokens. We estimate running costs during the assessment, set a monthly budget with alerts, and report the actual spend every month.

Are you partners of these vendors?

No. We are independent. Product names identify the tools; they don’t imply partnership or endorsement.