AI agentsSales

AI agents for sales: the skills, the job and the agent

Most sales teams use AI one prompt at a time: a follow-up here, a company summary there. Every rep asks differently, so customers get five styles and the CRM gets five formats.

A skill describes one sales task in detail, the way your best rep does it. Several skills in order make a job, such as the follow-up after a meeting. The job running on its own, stopping for the rep before anything reaches a customer, is an agent. Here is that ladder for sales.

  1. 01

    6 skills

    One task each, described in detail.

  2. 02

    Connectors

    Reach into the systems the tasks need.

  3. 03

    One job

    The meeting follow-up

  4. 04

    One agent

    The follow-up agent

From skills to agent
The skills

Claude skills for sales: the 6 to write first

meeting-prep

Meeting prep

Use when: a customer meeting or demo is booked and the rep wants to walk in prepared.

Reads the deal, the contacts and the latest notes and emails, and writes who is coming, where the deal stands, what we have promised and three questions to ask.

Output: One page to read five minutes before the meeting.

call-notes-to-crm

Meeting notes for the CRM

Use when: a meeting or call has ended and the notes need to go into the CRM.

Turns rough notes into the CRM’s fields: summary, decision makers, objections, next step with a date and a suggested deal stage.

Output: A note ready for the deal, and a suggested stage the rep changes themselves.

follow-up-email

Follow-up email

Use when: a customer meeting, demo or call has ended and the customer should get a summary.

Writes the email in the rep’s own voice: what we heard, what we agreed and the next step with a date. Anything it cannot confirm goes on a list to check.

Output: A draft under 150 words, ready to read and send.

proposal-draft

Proposal draft

Use when: a deal is ready for a proposal.

Takes the customer’s needs from the notes, picks the right package and writes the proposal in your template. Any price, discount or start date nobody has approved is marked as an open question.

Output: A proposal draft in your template, with the open questions at the top.

pipeline-review

Weekly pipeline review

Use when: someone asks for the pipeline review, the forecast or which deals are stuck.

Goes through every open deal, flags those with no next step, a close date in the past or no activity for 14 days, and sums up value per stage.

Output: One page with the stuck deals and one action for each.

lead-qualification

Inbound lead check

Use when: a new lead comes in from the website, an event or a colleague.

Checks the lead against your ideal customer (industry, size, role, need), grades it A, B or C with one line on why, and suggests the next step.

Output: A grade, the reason and the next step, ready for the CRM.

One skill, written out

What a good skill looks like

An example written for this page, following Anthropic’s guidance for writing skills. Adapt it to how your team works.

follow-up-email/SKILL.md
---
name: follow-up-email
description: Writes the follow-up email after a customer meeting, with what we agreed and the next step. Use when a sales meeting, demo or call has ended or someone asks for a follow-up or recap email.
---

# Follow-up email after a customer meeting

## What you deliver
One email draft, at most 150 words, in this order: Subject, Thanks (one line), What we heard, What we agreed, Next step, Sign-off.
Below the email, a short list headed "Check before sending" with anything you could not confirm.

## Steps
1. Read the meeting notes and the deal: company, contacts, stage, what we promised before.
2. What we heard: the customer's problem in their own words, at most three points.
3. What we agreed: every promise from both sides, with who does what.
4. Next step: one action, one owner, one date. If the notes have no date, suggest two times and add it to "Check before sending".
5. Subject: the customer's company and the next step, for example "Acme AB: pilot start 1 December".
6. Read the email as the customer would. Cut anything they did not say or agree to.

## Rules
- Write as the rep: first person, short sentences, no "I hope this finds you well".
- Never add a price, a discount or a delivery date that is not in the notes. Put it under "Check before sending" instead.
- If you cannot tell who attended or what was agreed, leave [brackets] where the gap is. Never guess.
- Attach nothing. Name the document the rep should attach.
- This is a draft. A person reads it and presses send.

## Example
Subject: Nordbyggen AB: pilot start 1 December

Hi Sara,

Thanks for today. What I heard: your project leads spend Fridays chasing time reports, and the board wants one view of every site before the new year.

We agreed that I send the pilot plan by Wednesday and that you check the start date with Johan.

Next step: a 30-minute call on Thursday 12 November at 10:00 to go through the plan. Does that work for you?

Best,
Erik
The description uses the reps’ words
Reps rarely say “use the follow-up skill”. They say “send them a recap” or “follow up with Acme”. The description names those moments, so Claude picks the skill up when it is needed.
A rule for every promise
The costly mistake in a follow-up is a price or a date nobody agreed to. The skill moves those to a list to check instead of writing them into the email.
The example sets the length
A real email under 150 words shows the tone and the level of detail better than any rule. Every follow-up comes out the same way, whichever rep is behind it.
The job

From skills to a job: The meeting follow-up

Today the follow-up email and the CRM note take a rep half an hour after every meeting, and in a busy week they wait until Friday. With the skills and the right connectors it takes ten minutes per meeting, in Claude or ChatGPT.

  1. 01Ask the AI for the deal and its latest notes from HubSpot, and give it your notes from the meeting.
  2. 02Run call-notes-to-crm on them.
  3. 03Run follow-up-email on the result.
  4. 04Read both and add what only you know, such as who really makes the decision.
  5. 05Send the email and save the note on the deal.
The agent

From job to agent: The follow-up agent

Starts when

Every day at 16:00, for the meetings held that day.

Then
  1. 01Lists today’s meetings in your calendar and keeps those with attendees from outside the company. via Google Calendar MCP
  2. 02Reads the notes you saved from each meeting in your meeting notes folder. via Google Drive MCP
  3. 03Reads the deal, its contacts and its latest notes for each customer. via HubSpot MCP
  4. 04Runs call-notes-to-crm, then follow-up-email, for each meeting.
  5. 05Creates each follow-up as a draft in your Gmail. Nothing is sent. via Gmail MCP
  6. 06Adds the meeting note to the deal once you have confirmed the text. via HubSpot MCP
A person approves

The rep reads each draft, fixes what is wrong and presses send in Gmail. The note goes on the deal only after the rep has confirmed it. Nothing reaches the customer before that.

Result

A follow-up waiting in Drafts and a CRM note ready for every customer meeting, by 17:00 the same day.

In Claude, make it a scheduled task: describe the job once, choose daily at 16:00, and it runs with your skills and connectors even when your laptop is closed. It goes through the day’s meetings at a set time, not the minute a meeting ends, which is all a same-day follow-up needs. Scheduled tasks are on every paid Claude plan. In ChatGPT Work, a scheduled task does the same on the same daily cadence.

With Deployed OS, the skills and connectors belong to the company, not to one rep. Every rep gets the same follow-up skill in Claude, ChatGPT and Copilot, connects HubSpot and Gmail with their own login, and every call the agent makes is logged. Admins decide which groups may let the AI write in the CRM at all.

Pitfalls

Where it goes wrong

A CRM nobody trusts
If stages, close dates and next steps are out of date, the agent reads the mess and writes it up with confidence. Agree on what each stage means before you build on it.
Promises in the draft
A follow-up that invents a discount or a start date becomes a promise to the customer. Tell the skill that prices and dates go on the list to check, never into the email.
One prompt per rep
When every rep writes their own follow-up prompt, customers get five styles and the sales lead gets five CRM formats. Write one shared skill and let the team improve it.
Drafts nobody reads
If reps press send without reading, the checkpoint is gone. Keep drafts short enough to read in a minute, and put the list of things to check where nobody can miss it.
Questions

AI agents for sales: common questions

The recurring work around the selling: meeting prep, the follow-up email, the CRM note, the proposal draft and the weekly pipeline review. It does each in the same order every time and stops for the rep before anything reaches a customer.

The ones every rep needs after every meeting: the follow-up email and the CRM note. They run many times a week, so you learn fast what the skill gets wrong. Then meeting prep, then the pipeline review.

It cannot send. The Gmail and Outlook connectors create drafts that wait in Drafts, and in Oneflow it creates a draft contract from your template that nobody receives until you send it. In the CRM it adds a note to a deal in HubSpot or Pipedrive, or logs an activity in Salesforce, after you confirm the text. It does not change stages or deal values. We recommend a person approves every one. With Deployed OS, admins decide which groups may let the AI write at all.

Both. Skills are written in plain language, and the same connectors work in Claude, ChatGPT and Copilot through Deployed OS. Use the assistant your reps already have open.

Yes. Pipedrive and Salesforce are read the same way, and Outlook drafts work like Gmail drafts. Only the write step differs: Pipedrive gets a note, Salesforce an activity on your own activity list. The skills stay the same; only the agent’s connectors change.

It depends on the size of your team. Get in touch and we will walk you through it.

Other rolesAI agents
Let’s talk

Start with one job.

Tell us which task eats your team’s week. We’ll show you the skills that do it, and the agent that runs it.

Poyan Karimi
You’ll hear from Poyan Karimi
Co-founder · Deployed AI