CRM Copilot

Super excited to have been working with @GrantMargerum this last few weeks. He's building a YC Backed startup that's trying to make CRMs easier to update with new data. With permission, I want to share some aspects of the project.. Let's start with design!

Our first step during project kickoff was to wireframe the idea out in more detail Grant had already spent some time thinking about the idea so we hit the ground running, solving for UX flows right away

It didn't take very long before we had what looked like an interface that could plausibly go live

And kept on nudging the details until it started to feel a bit more well thought out

With enough context, I went into "design mode" and took the input I got from the client along with the wireframing we did and spent some time iterating. The look we were going for was pastel colors that was simple, clean and still unique in the enterprise/sales app universe

With the goal of launching for YC Demo week, we're jumping straight into "build mode" now that we have a look that both the client and I are very happy with

Getting the design language done is super critical for build velocity. The last thing you want to think about when building out complex APIs and frontends is how rounded you want a button. It can be distracting and inefficient. So with the designs ready, it's build build build!

If you enjoyed this, give me a follow! I build software for founders and share some details along the way. This app is already being built out and can't wait to share more as it's ready!

Figma -> Localhost Localhost hits different

In just a bit more than a week, we have a pretty complex project taking shape here The stack is Express, Typescript, Next, Supabase/Postgres, Clerk & Railway. This isn't just a frontend you're looking at.. it actually works!

We're using AI to extract contacts from the given email conversation. The contacts can be saved to Hubspot in just one click. This is a big step in the right direction because we have all the critical pieces working, and now it's a matter of finessing it into a production build

The world is changing so fast. The future is going to be way smaller teams, executing incredibly fast. When you're working with a founder with a clear vision, it's crazy what you can accomplish in a week or two.

The destination is not fixed After building out some core functionality, we realized we were building in the wrong direction. 2 things became obvious to us:

1. Our MVP direction would require a lot of heavy lifting on the user's end by having to manually paste in the activity. 2. We didn't generalize the read/write into the CRM enough for it to be useful for real world use cases. Here is the pivot, it's much simpler!

Changlog: 1) We give the user recommendations. Without any manual input from them. 2) We develop recommendations by analyzing the emails/documents they give us access to via APIs

This is why I like working with clients by billing time, not scope. As we learn more, we can change the direction, the scope, and the goals of the project. We're not building against "our plan". Our plan is evolving to match reality. Nothing set in stone.

Touchdown Made the first generalized save to hubspot

What do I mean by generalized? The previous MVP we decided on would deal with very specific objects. But we decided to scrap that! We're now going a step further and letting the user decide what they want to update in hubspot. This is a lot harder, but we're now alot closer!

Layering on complexity slowly For anyone who has built integrations, you know how complex it can get. And integrations into CRM systems like hubspot/salesforce, are another level So, I'm trying my best not to "spook" the codebase with added complexity One abstraction at a time

Building an AI agent for @GrantMargerum 1. It is blowing my mind what agents can already do 2. The future of software is going to be very very interesting 3. I think agents will become as fundamental to software as a function call is today 🤯

@LifeOfPhat @GrantMargerum We are trying to solve for a very specific use case so we're building our own.

We've been pushing some boundaries on the Quantic project Not much has changed on the frontend, it looks the same But how it solves the problem has been reworked from the ground up to use an AI Agent CRMs are INCREDIBLY complex. continued..

And extracting data from your emails to update them is a VERY hard problem to solve. It's kinda easy to solve the 'happy path'. Where you find one kind of data and update the CRM But VERY very hard to solve for all the edge cases. And there are hundreds, maybe thousands of them

This is where an Agent helps us solve this problem in a way that would be massive uphill battle with a deterministic control flow (how normal software works) The agent can understand the nuance of what to update in the CRM, and what to leave out.

It's a tiny human, doing slightly human things. Saving the real breathing human a sh*t ton of time. If I had some extra money, I'd invest in this app myself. I have a ton of faith in @GrantMargerum to take this to fruition in the marketplace.

For one thing, he's extremely customer obsessed. And if I've learned anything, that is the number one requirement to win in business. So yeah, watch this space, if you're a customer or investor.

This thing is now generating recommendations using AI Agents for real It's like magic

Tech we used for the Quantic project: - Custom AI Agent - Clustering (embeddings/dbscan) - Queuing - NoSQL - OAuth - Typescript - React/Next Frontend - Design - Integrations into Salesforce/Hubspot/Gmail/Gcal/more - Cronjobs (doh') Crazy stack. Game changing value for the user.