I built three AI agents that run my sales pipeline

Hi everyone,

A year ago, everyone got carried away calling any AI automation an "agent."

But, an agent is not just a five-step workflow that runs the same way every time. An agent is an LLM that can loop through tools, call different tools depending on what it finds, loop as many times as it needs, and decide on its own when the task is done. No fixed path, no predetermined steps.

That distinction changes everything.

Deep Research was the first real glimpse of agentic capability. You gave it a question and it went off on its own. Browsing, reading, comparing, circling back. It decided what to do next based on what it found. That felt different.

Then OpenAI launched Operator. Soon after, they killed it and turned it into "Agent". It was clunky, limited, frustrating. It felt like they rushed it out the door.

Claude Code is the best example of agentic behavior I've seen. You give it a goal, it reads files, writes code, runs tests, fixes errors, reads more files, and keeps going until the job is done. It even proactively asks questions, if it thinks you’ve missed something.

I've watched it loop through multiple tool calls on a single task. That's an agent.

This is where most of my agent building time goes right now. Notion has calendar, email, and custom MCP integrations. All my context already lives there. So I built a three-agent sales system.

Here's the hack: Notion agents can't talk to each other directly yet.

So I built shared context files that all three agents read from and write to.

The SDR updates the file after qualifying a lead. The AE reads it before prepping for call two. The CRO reads everything and writes its analysis back. They "communicate" through shared documents.

An orchestrator concept is coming to Notion. When it does, these three agents will become a real team, just like the Claude Code agent teams I discussed recently over on my YouTube:

For now, the shared file system works.

I also work with Lightfield as a secondary CRM alongside the Notion system. Different tools for different angles on the same pipeline.

The models powering all of this are getting serious. GPT-5.3 Codex is phenomenal at big, complex coding tasks. Opus 4.6 (what I'm running right now) excels at lots of small decisions fast. GPT-5.3 Instant just came out. Both families are good at different things.

Here's what I'm exploring beyond Notion:

I'm building a lot with this right now. Their agent builder lets you chain tools, add memory, and monitor every decision the agent makes. If you want full control over how your agent thinks, start here.

Built on an integration platform. It's proactive: it talks to the team on Slack, pulls data from your tools, and surfaces things before you ask. Worth watching.

Automations that can run agents inside them. If you need something that fires on a schedule or event and then makes its own decisions, this is interesting.

My setup: Telegram bot, cron jobs, and Tailscale tunneling. It runs tasks overnight, sends me results in the morning. I've been running this for months and it handles everything from content creation to deal tracking.

You don't need to be technical to build and run AI agents.

Just start here:

1. Find the idea — Write one sentence about what the agent does. Then write a paragraph covering what tools it needs access to and what "done" looks like. Keep it specific.

"Qualify inbound leads and draft a response email" is better than "help with sales."

2. Pick your tool — Zapier is the simplest. Make is in the middle. n8n gives you the most control. All three have agent builders now.

3. Describe it — Put your sentence and paragraph into the tool's agent builder. Let it generate the first version.

4. Connect your tools — Log into your email, CRM, calendar, whatever the agent needs. Most platforms handle this with OAuth. Five minutes.

5. Test it — Run it at least 10 times. Look at every output. Give feedback on each run. Optimize daily for the first week.

Most important: keep your agents specific. An agent that does one thing well beats an agent that tries to do five things poorly. My SDR agent only handles lead qualification to first call. That constraint is what makes it reliable.

March 12. 2:30 PM GMT. Claude Code for Founders.

50 spots. More than 10 are gone. It’s free, hosted on Zoom, for founders and executives.

I'm going to open my terminal and show you how I build these systems.

Reply to this email to secure your spot!

The best AI outputs come from detailed prompts. But typing long, context-rich prompts is slow - so most people don't bother.

Wispr Flow turns your voice into clean, ready-to-paste text. Speak naturally into ChatGPT, Claude, Cursor, or any AI tool and get polished output without editing. Describe edge cases, explain context, walk through your thinking - all at the speed you talk.

Millions of people use Flow to give AI tools 10x more context in half the time. 89% of messages sent with zero edits.

Works system-wide on Mac, Windows, iPhone, and now Android (free and unlimited on Android during launch).

Try Wispr Flow free →

Quick shoutout to my wife, Rebecca, who polishes every edition of this newsletter.

First, I dictate my thoughts for the week into Claude Code, where I’ve created a newsletter-writing Skill. Then, a decent draft appears in the content calendar. From there, she does the final polish and fact-check, then publishes it.

She was a copywriter for Notion before joining me. Last week we hit a 50% open rate, that's her craft showing.

We're not prompt engineers anymore. We're becoming orchestrators. We decide what agents to build, what context they need, how they coordinate, and when to step in.

We're becoming AI Operators.

If you're exploring agent tools and found something good, reply and tell me. I test everything.

Tim