making it work #1

a new series on ai, agent-building, and whatever comes next!

Apr 03, 2026

This is the first post in a new series called making it work — where I document what I’m actually building with AI, what broke, what surprised me, and what it means if you’re a marketer who keeps hearing “agents” and nodding like you know what that means. (I was you. Recently. It’s fine.)

Happy Friday :) I stopped being skeptical of the agent hype.

It was a Tuesday, and I had just been reminded (again) about an event I’d already forgotten to RSVP to, which I had already forgotten once before, which I had initially forgotten because I didn’t write it down, which I didn’t write down because I was in the middle of something else, and around here is where the spiral of human failure usually ends with me sending a slightly embarrassed “so sorry, is it too late to?” message.

So I built an agent to handle it. And it worked. And then I couldn’t stop.

I’ve been learning a lot about agents this week!

The first thing I made was an event reminder agent in Gumloop.

Gumloop is, and I say this as a genuine compliment, very pink. That’s not why I chose it, but it’s not nothing. My girlfriends love it.

More importantly, it makes building an agent feel easy. It even helps you think through what you’re building.

The event reminder agent does exactly what it sounds like: it reminds you about events. It monitors my calendar, Luma, and Partiful, identifies upcoming commitments I haven’t acted on, and pings me with enough lead time to act.

A sticky note that learned to talk.

The anatomy of an agent

Agents have an anatomy. Every agent, regardless of what it does or how complex it looks, is made of the same basic parts:

A trigger (what starts it)
Instructions (what it should do and how to behave)
Tools and access (what it can see and act on)

These determine the output (what it produces or does with what it knows). That’s it. Dead simple. Every tool, every demo, every breathless “I automated this” post maps back to this structure.

The second thing I built was a competitive intelligence agent that outputs a battle card. On Zapier.

This one has been helpful. The agent pulls from a set of sources (transcripts, competitor websites, G2 reviews, recent press, recent job postings) and synthesizes what’s changed, and outputs a structured battlecard: positioning, key differentiators, objection responses, and messaging shifts.

Then I built a battlecard agent to ensure all that information is ready to go and published on Notion. (You can swap Notion out for Google Drive or whatever internal knowledge base your company or sales team might use)

What it surfaces is what I care about: not just what competitors say about themselves, but the delta. What changed, what they’re doubling down on. That’s the part that’s usually missed when competitive intel is done manually and inconsistently.

Patterns require repetition. Agents are very good at repetition.

Gumloop or Zapier?

Zapier is the thing that connects your CRM to your email, your spreadsheet, your Slack, and the twelve other tools you said you’d consolidate.

It’s reliable, it’s established, and it has a decade of integrations baked in. Zapier has also recently moved hard into the agent space. Making it easy to have your automations and agents in the same place, which is a genuinely different capability than just triggering actions between apps.

Gumloop, by contrast, feels built for the AI-native age. Where Zapier connects tools, Gumloop orchestrates AI-native thinking. I also prefer the UI. It’s also easier to build agents that scrape Luma, LinkedIn, or X with Gumloop.

Zapier, however, has a stack of established integrations. So, I wonder if I’ll hit a snag with Gumloop at some point. Gumloop ultimately has to add integrations at lightning speed to expand its library and destroy Zapier’s moat.

Notion’s Custom Agents

Been finding out so many new things you can do on Notion custom agents, like:

A Q&A custom agent for your internal knowledge base. You might have your answer before your engineer even gets to your questions.
Weekly project updates
Weekly team updates (calibrate against OKR)
Feature and bug tracking

See you next week!

Reply if you’re already building something. I want to hear what’s working and what absolutely isn’t! Also, let me know if this series is any good! See you on Sunday :)