How To Create An AI Agent, Today!

Last fall at Hubspot's INBOUND conference, Hubspot's Founder and CTO Dharmesh Shah unveiled his pet project: ​Agent.ai​.

I signed up immediately.

My opinion is that when a successful tech person who founds a multi-billion dollar tech company shares free signups to their new tech pet project, you should sign up immediately. Anyone's hobby that's on the bleeding edge of tech has the potential to be the next new hotness, but the odds get better if that pet project is led by someone more visionary (and well-financed) than your cousin.

Agent.ai is designed to be "A marketplace and professional network for AI agents and the people who love them." Think of it like LinkedIn for robots: it's a way to find, rate, and "hire" AI apps that do a range of different things.

Of course, if you want to have a social network of robots, you have to make it easy for new robots to join. A key feature of the platform is called their Agent Builder, which is basically a wizard to help non-coder people like me build AI-powered apps.

Again, if the CTO of an easy-to-learn billion-dollar software lets you into his hobby project for free, you sign up.

During the holiday break, I gave myself room to play around with the Agent Builder and figured out how it worked (I know, we enjoy holiday breaks differently, let me be me). I ended up designing a ​Brand-Relevant Editorial & Narrative Direction Assistant, aka BRENDA​.

Thanks again to all the people who filled out the survey on how this newsletter should evolve for 2025. The most popular ask was for practical resources and guides, so here's a practical breakdown of how to build an AI app if you don't know a lick about coding.

Step One: The Problem

Always start with the problem.

One of my strategic initiatives for 2025 is to improve Learn to Scale's backlink count. I intend to go out to more popular websites and do guest blog posts that link back to www.learntoscale.us to improve my SEO ranking and traffic.

A challenge with executing this plan is that I'll have to read a bunch of posts on another person's website, figure out their vibe, then come up with an article that bridges my goals (enabling small marketing agencies to grow) and their goals (probably something adjacent to that)...and I'll have to do that for dozens, maybe hundreds of blogs.

The problems:

1. Scanning someone else's blog to get a feel for what their audience wants takes time and I could easily get it wrong.
2. It's hard to stay within the guardrails of my own brand if I have to write hundreds of unique articles.
3. It's hard to connect my brand with someone else's brand.

Step Two: The Ideal Solution

What I would love is to put in the web address for the target blog, put in my brand guardrails, then have AI suggest some topics. I can then take the best prompts to kickstart my article writing.

Step Three: Build a MVP

I know this is the part that is intimidating, but let me reassure you that this is a lot less challenging than you think. The Agent Builder really only has three things you need to fill out:

1. The Settings, like the name and description for your AI agent. Your LinkedIn profile is more complicated.
2. The Trigger, which for everyone here should be "Manual Only." This means that it only does something when you use it. Don't overthink this.
3. The Actions, which are the things you want it to do. The Agent Builder comes with several templates that builds out all the actions you need for a simple AI agent and all you need to do is tweak it.

For BRENDA, I started with "Chat with a website" template. What this template does is takes in a user-supplied URL and a user-supplied question. The agent then scans the target website, throws it into AI with your question, and then spits out the analysis.

HMMM! Sounds like a pretty close match to my ideal solution, right?

The only real tweak I needed to do was add more context during the "throw it into AI with your question" step. Here's the actual language within BRENDA and some explanations of why I included it:

Here's Why: The Agent Builder has already scanned the target blog and stored it as a variable and also took the user-supplied brand guardrails and stored that input as another variable. I'm just summoning them for the AI.

Here's Why: This is the core function of the app and also why AI is blowing up the world: I don't know coding but I can write what I want in simple English and the AI can figure out the rest on its own.

This step could be where the agent stops, but I have two tricks to make this sooooo much better:

Here's Why: Prompting an AI to think of best practices, common mistakes, myths, etc. to build a scorecard drastically improves the AI generation.

Here's Why: User experience matters. I don't need to see all the work the AI is doing, but I want to know what the AI thought the target blog was about to feel confident with its topic suggestions.

I've also found that AI (similar to humans) actually produce better results when you tell it to revise its work, especially when you give it a pile of best practices from the scorecard.

That's it. You've now created an AI Agent. See how well I did and give BRENDA a try.​

Was this breakdown useful, interesting, or eye-opening? Or just meh? Let me know what you thought of this kind of content.

A TL;DR from the CRO

Is January the season for hygge or hobbies...or both?

-Roman Noodles, Chief Ruff Officer

Last summer, I did a webinar with Krysta Van Ranst from BuildingPPL about how easy it is to get state grant funding for training. During the webinar (watch the recording here!), she applied and submitted a grant application for one of my grant-approved programs in less than 10 minutes.

This week, we completed the final phase of that program with a team offsite focused on "Building BuildingPPL for the Future": making their mission more tangible, charting buying journeys, prioritizing how to evolve their services, and penning a 3-year service offering & staffing roadmap.

Her company got that all funded by the Express Grant program, but I feel like I should have paid her for the wonderful experience. Her team was so welcoming, engaged, passionate, and psychologically safe that they could have really deep conversations on sensitive topics with ease. It was a masterclass in how to build a company to be proud of.

Starting my 2025 with that offsite reminded me why I do what I do: I want other companies to be like BuildingPPL.

If you want to have a workplace where you can laugh, cry, do great work, learn from each other, and be true to yourself, make sure to follow Krysta on LinkedIn.​

​

Hanging out with cool people,

Dan from Learn to Scale​

​Opt-out from the newsletter | Unsubscribe from all emails | Update your Preferences | www.learntoscale.us, Boston, MA 02119

​

​PS. You can't make something delicious without breaking some eggs.​