AI SDR agent: find leads and book demos from one chat prompt

An AI SDR agent is software that does a sales development rep's job from a single request: it works out who buys your product, finds the people showing buying signals right now, writes a personal first message, and sends it across LinkedIn, X and email. You tell it the outcome you want. It does the prospecting.

The 50 second video above shows the whole loop as one chat with the PumpGTM agent. Everything in it (the people, the companies, the numbers) is a fictional example made to show the flow. None of it is a customer result.

You start with the outcome, not a lead list

The prompt in the video is something founders actually say: "I need 20 demos with engineering leaders this month and I'm out of hours." No filters, no CSV, no sequence builder.

That works because PumpGTM runs inside the AI you already use. You add it to Claude or ChatGPT through MCP, and from then on the chat is the interface. If you want the step-by-step version of this, the AI-first GTM playbook walks through it with prompts you can copy.

It reads your site and picks who buys

The first thing the agent does is read your website and say back what you sell, in one line. In the example it is "AI code review for fast-moving teams."

From that it picks the buyer: Head of Engineering and CTO, at AI startups with 20 to 200 people. You can correct it before anything else happens. Getting this right matters more than any message, which is why a real AI SDR starts here and a mail-merge tool skips it.

It keeps only the people showing signals

Next it scans profiles that match and looks for reasons to reach out this week: the company is hiring engineers, just raised funding, or the team is growing fast. These hiring and funding signals are the difference between a cold message and a timely one.

In the video, 13,200 profiles become 248 with signals and 6 best fits for today. The point is the shape, not the numbers. Most of the list gets thrown away on purpose.

It also skips people who look right on paper but are not buyers. Two get dropped in the demo: a recruiter (not a buyer) and a company with 5,000 people (too big). That kind of lead scoring before outreach is what keeps reply rates up.

Each message opens with something true about them

The example opener comes straight from the company's job posts:

Saw Mergeloop is hiring 6 engineers. Teams growing that fast drown in code reviews. Our AI reviewer clears each PR in under two minutes. Worth 15 minutes this week?

It names a real reason, connects it to the problem, gives one line of proof and ends on a question. You see the messages before outreach begins, and you can edit them.

It sends on every channel and brings back replies

Then it sends across LinkedIn, X and email at the same time, from your own accounts, each one personal. Reaching the same person in more than one place is the whole idea behind multi-channel outbound.

Replies come back to you, and the goal of every conversation is a demo on your calendar. That last frame of the video, a calendar filling up, is the job a first SDR hire would be doing. If you are weighing that hire, here is how an AI SDR compares to a human SDR.

Try it in your own chat

PumpGTM works inside Claude and ChatGPT. Add it from the MCP page, or start with the AI-first GTM playbook if you want to run the first audience by hand before you automate it.

Transcript

There is no voiceover in this video. It is a music-only screen of a chat with the PumpGTM agent. This is the on-screen text, in order.

"What should we sell today?" The user types: "I need 20 demos with engineering leaders this month and I'm out of hours. Help!!!"

The agent replies: "Connecting to PumpGTM..." then "Connected to PumpGTM." "Read your site: AI code review for fast-moving teams." "Picked who buys: Head of Engineering, CTO, 20 to 200 people, AI startups."

"Scanning 13,200 profiles for buying signals..." Signals: Hiring engineers. Just raised funding. Team growing fast. Counters: 13,200 scanned, 248 showing signals, 6 best fits today. "Scanned 13,200 profiles. Kept the rare few."

"Finding you the best fit leads..." Lead cards show each person's signal and fit score. Two are skipped: a recruiter (not a buyer) and a company with 5,000 people (too big).

"Drafting high-converting messages..." Example message, with the opener taken from the company's job posts: "Saw Mergeloop is hiring 6 engineers. Teams growing that fast drown in code reviews. Our AI reviewer clears each PR in under two minutes. Worth 15 minutes this week?"

"Finding and sending messages across every channel..." LinkedIn, X and email inboxes, 593 personal messages in parallel.

"Replies are coming in..." "Booking meetings into your calendar..." The calendar fills with demos.

"Book demos on autopilot without spending on a GTM team." Works inside Claude and ChatGPT. pumpgtm.com

Questions founders ask

What is an AI SDR agent? Software that does a sales development rep's job from one request. It picks who buys, finds people showing buying signals, writes personal messages and sends them across LinkedIn, X and email.

Are the people and numbers in the video real? No. Mergeloop, the leads and every number are fictional examples made to show the flow. They are not customer results.

Do I need a lead list to start? No. You describe the outcome you want. PumpGTM reads your site, proposes the buyer, and you correct it before it finds anyone.

Does it send anything without me seeing it? No. You see the audience and the messages first, and outreach starts when you approve it.

Which AI tools does it work with? Claude and ChatGPT, plus Claude Code, Codex and Cursor. You add it as an MCP server.

Which channels does it use? LinkedIn, X and email, from your own accounts, in one coordinated sequence.