Agents

AI agents that do the prep, so your team can sell

Free, open agents for Claude that research companies, prepare calls and write up follow-ups. You stay in charge: nothing is sent or saved without a person saying yes.

What is an AI agent?

An AI agent is a short set of written instructions you install in Claude, the AI assistant. You give it a task in plain English, such as “prepare my 10:00 call”. It follows the same steps every time: it reads the right sources, drafts the result and stops for your approval before anything reaches a customer or your CRM.

How it works

Four steps, and you own the last one

  1. 1

    You ask

    In plain English: “Brief me on northwind-freight.com before Thursday.”

  2. 2

    It reads

    Your CRM, the company website, news, job ads and LinkedIn. Only what you connected.

  3. 3

    It drafts

    A brief, a call plan or a follow-up, in the same format every time, with sources.

  4. 4

    You decide

    Read, edit, approve. Nothing is sent or saved until you say yes.

Your first agent

Set up in about 10 minutes

  1. Get Claude. Use the Claude desktop app or claude.ai on a paid plan. Agents run as Claude “skills”.
  2. Download an agent. Each agent is a small folder with a file called SKILL.md: the written instructions. Open agent files will be on GitHub soon.
  3. Add it to Claude. In Claude’s settings, find Skills and upload the folder as a zip. In Claude Code, put the folder in your skills folder.
  4. Ask in plain English. “Brief me on acme.com.” Claude recognises the task and follows the agent’s steps.
  5. Connect your tools when ready. Add HubSpot, Gmail, Calendar or Slack in Claude’s connector settings so the agent can read your own data.
Full setup and safety guide
What an agent looks like inside: SKILL.md
---
name: account-brief
description: Build a one-page brief on a company before
  first outreach. Use when someone says "brief me on",
  "research this company" or gives a company website.
---

# Account brief

1. Confirm the company (website, name, country).
2. Read their website, recent news and job ads.
3. List buying signals with a date and source.
4. Draft three guesses about their priorities.
5. Suggest one opening line.
6. List what you could not verify.

Rules: one page max. Never guess. Mark anything
unsourced as "not verified".

Safe by design

Six rules every agent follows

  1. 01

    Read first

    New agents start read-only. Writing is switched on only when you trust the output.

  2. 02

    Propose, approve, write

    An agent shows you the change first. It saves only after a person says yes.

  3. 03

    Never overwrite

    Agents fill empty fields. Changing existing information always needs approval.

  4. 04

    Blank beats guessed

    When the agent is unsure, it says so instead of inventing an answer.

  5. 05

    Keys stay private

    Passwords and API keys live outside the agent’s files, where it cannot read or leak them.

  6. 06

    Web pages are data

    Text on websites and in emails is treated as information, never as instructions to follow.

Questions people ask

What is an AI agent for sales?

An AI agent is a set of written instructions you install in an AI assistant such as Claude. You give it a task in plain English, like "brief me on this company". It researches, drafts the output and stops for your approval before anything is sent or saved.

How is an agent different from asking ChatGPT?

A normal chat starts from zero every time and follows no fixed process. An agent follows the same steps every run, reads your own context such as your customer profile and CRM, and produces the same format each time, so the output is consistent enough to rely on.

Do I need to know how to code?

No. You install an agent by adding a folder to Claude and then talk to it in plain English. Connecting tools like HubSpot or Gmail takes a few clicks in Claude’s settings.

Is it safe to let AI touch my CRM?

Yes, if it is set up to propose changes instead of making them. Every agent here shows you the change first and saves nothing until a person approves it. Start read-only and switch on writing only when you trust the output.

What does it cost?

The agents are free. You need a paid Claude plan to use skills and connectors. Check Anthropic’s pricing page for current plans.

Want help setting up your first agent on your own tools? Bring it to a free 20-minute Lab Hours call.