A Few Projects // BrXnd Dispatch vol. 053

Building an AI personal assistant and some other fun projects

You’re getting this email as a subscriber to the BrXnd Dispatch, a (roughly) weekly email at the intersection of brands and AI.

In this edition, I'm excited to share some of my recent AI projects that, while not directly marketing-focused, demonstrate the potential of AI to solve real-world problems and streamline workflows. These examples showcase how marketers and brands can think creatively about applying AI to their own challenges. By exploring these projects, I hope to inspire you to consider new ways AI could transform your marketing processes and strategies.

Marshall

The first one is something I’ve been working on for almost a year now, and I call it Marshall (named after McLuhan, of course). It’s a personal assistant that I have slowly been adding various features. Here’s a partial list:

• Receive texts and emails

• Sort tasks into projects automatically

• Reply to comments on Linear, my project management system

• Do research

• Forward my receipts to my expense tracking system

• Summarize emails

It probably does some other stuff, too, but I mostly use it for this. When I run into a problem a few times, I try to add it to Marshall. He also tries to reply in the famously obtuse style of Marshall McLuhan, which makes me laugh whenever it happens. 

One of the recent additions is to connect it to a separate research service I built so that I could comment on a message in my task manager, and Marshall would call the research service and then leave the answer as a comment. Here’s what that looks like:

The general architecture of Marshall is something my friend Lance from Langchain has been calling a control flow agent. I like the distinction because everyone’s first question when you talk about agents is whether it can book flights or write recipes—totally open-ended tasks. In contrast, Marshall operates within a more defined scope. It can perform a specific set of tasks, with the AI deciding which of these tasks to execute, but it doesn't venture beyond this predefined list. This approach results in a deterministic workflow where the AI has some decision-making autonomy within fairly rigid boundaries. From my observations and discussions with others, this control flow approach appears to be the only one that really works and is deployed in production. The more open-ended agent systems remain aspirational, likely requiring significantly more powerful models than the ones available today.

While Marshall is my personal assistant, imagine the potential of a similar AI system tailored for marketing teams. It could automate campaign reporting, track competitor activities, manage content calendars, or even assist in real-time social media engagement. The key is to identify repetitive tasks in your marketing workflows that could benefit from AI-powered automation and decision-making.

Obsidian Tools

I’m a big fan of Obsidian, a note-taking tool with a robust open-source ecosystem. Fundamentally, Obsidian isn’t complicated: it’s just a text editor you can use to write markdown. What makes it different is that rather than storing all your notes in a database, it just stores them as files on your computer. While that doesn’t sound revolutionary, it lets you do neat stuff like sync all those files to Github.

Anyway, I’ve been using Obsidian for a little over a year now and have quite a collection of notes. One of my thoughts was that getting a summary note at the end of each day would be nice. So, I built and open-sourced it. 

Fundamentally, the code does three things:

1. Email to Obsidian: Set up an email address that can catch emails and add them as notes to your Obsidian.

2. Daily Summaries: Writes a Daily Summary note each day of your new notes + changes.

3. Weekly Summary: Writes a weekly summary note each week from your Daily Summaries.

The summaries are broken down like this:

Daily Summaries

Daily Summaries are automatically generated each day and provide a concise overview of the day's activities, insights, and key information. They typically include:

1. Overall Summary: A brief paragraph summarizing the day's main events, discussions, and developments.

2. Interesting Ideas: A bullet-point list of noteworthy concepts or thoughts that emerged during the day.

3. Common Themes: Recurring topics or patterns observed across various activities or discussions.

4. Questions for Exploration: Open-ended questions or areas identified for further investigation.

5. Possible Next Steps: Actionable items or potential follow-up tasks based on the day's events.

6. Notes: Detailed breakdowns of specific meetings, conversations, or activities that occurred during the day.

7. URLs: A list of relevant links mentioned or accessed during the day, with brief descriptions of their content.

Weekly Summaries

Weekly Summaries are generated at the end of each week, providing a higher-level overview of the week's activities and insights. They typically include:

1. Overall Summary: A comprehensive paragraph or two summarizing the week's main themes, developments, and achievements.

2. Strategic Implications: Key insights or decisions that could have long-term impacts on projects or strategies.

3. Challenges & Opportunities: A breakdown of obstacles faced and potential areas for growth or improvement identified during the week.

4. Key Developments & Trends: Significant progress made on projects or notable trends observed in the industry or work environment.

5. Long-Term Implications: Potential long-range effects of the week's events or decisions on future projects or strategies.

6. Goals for Next Week: Specific objectives or tasks to focus on in the coming week based on the current week's outcomes.

These summaries serve as a quick reference for tracking progress, identifying patterns, and maintaining focus on important goals and ideas over time.

If you want to try it out, you’ll have to be technical and run the code yourself (it works “very” quickly with Vercel). You can find the repo on GitHub.

Why do I share all this?

I know the vast majority of you aren’t super technical, and ostensibly, this is a newsletter about the intersection of marketing and AI, so why do I keep talking about these projects?

There are a few reasons:

1. I’m having the time of my life building things. The magic of solving problems in these ways, because I have AI as a fuzzy interface, is absolutely amazing to me, and I hope some of the excitement rubs off on all of you.

2. Since I started this newsletter and the BrXnd Marketing X AI Conference and basically all the work I’ve been doing, I’ve worked hard to differentiate my approach from the rest of what’s happening in the world by being grounded in building. There’s too much talk and not enough doing, and in this moment of AI explosion, understanding by building fingerspitzengefühl is critical. 

3. I’m out there talking to brands and marketers all the time, and I know that one of the biggest challenges is that folks don’t immediately see the opportunities for AI. I hope that sharing some of these projects sparks imagination for the kinds of problems you might be able to solve.

4. I want to communicate that you don’t need to wait to build powerful tools with AI. One of my pet peeves is all the conversations about AGI. It’s not that I’m not interested, but I’m so blown away by what we have right now that the need to incessantly talk about what’s coming next seems a bit ridiculous to me. I think if we never got a better model than GPT4/Claude 3.5, we would still have ten years of runway to find amazing ways to integrate these things into our lives and work.

Finally, lots of folks have asked why I don’t turn some of these things into software as a service (SaaS). After all, that’s where I’ve spent much of my career. The answer to that is somewhat more complicated. Still, it essentially boils down to not seeing the path to building a defensible SaaS business (something investor James Cham and I discussed in our conversation). These tools I’ve built are cool and incredibly useful, but I also built them fairly quickly and was able to see a tremendous amount of value without an immense amount of work. That’s amazing if you’re an end user or a brand, but it’s less amazing if you’re trying to build a defensible SaaS product. That’s not to say no one will win in the world of AI software, but for now, I’m much more focused on solving problems with code + AI + expertise than trying to deliver it as a service to individuals and businesses. 

I think that’s it for this week. As always, thanks for reading, subscribing, and supporting. If you have questions or want to chat, or if you’re at a brand and want to go deeper with AI, please be in touch.

Thanks,

Noah