From Agency Work to Product Success

This episode we're joined by Stu Green, a product designer, agency founder, and serial app builder who's sold not one but two successful SaaS products.

We dig into the realities of building your own product versus running an agency, the role AI plays in modern product development, and whether the flood of AI-built apps is a threat or an opportunity for professionals.

Plus, we check out Bleet, an app that turns your meeting transcripts into social media content, and Paul shares how AI-powered personas are changing the way he approaches user research.

App of the Week: Bleet

You know you should be posting on LinkedIn. You've told yourself that every week for the past 6 months. But then you sit down, stare at the blank post box, and realize you have absolutely no idea what to write about. So you close the tab and promise yourself you'll do it tomorrow. You won't.

Bleet is an app built by Stu Green (and collaborator Nick) that solves this by mining the conversations you're already having. It takes your meeting recordings and transcripts, extracts the key topics using AI, and helps you turn them into social media posts. And the thing that sets it apart from just asking ChatGPT to write something for you is that it pulls your actual words and phrases from the conversation, piecing them together into posts that genuinely sound like you rather than generic AI slop.

How It Works

You connect your meeting recordings or transcripts (or even just speak a thought into the app), and Bleet will surface a list of topics you covered. From there, you pick the ones you want to post about and hit "create." You can dial in how much creative liberty the AI takes, from near-verbatim to lightly polished.

So you sit down for 10 minutes once a week, pick a handful of topics, schedule them up, and you're done. A single meeting can generate enough content for almost a week of daily posts.

What About Client Confidentiality?

The number one concern people raise is about sharing sensitive client information. Bleet strips out client names, specific people, and identifiable details. It focuses on the general topic and the ideas discussed, not the specifics of who said what in which meeting. And of course, you review everything before it goes anywhere, so if something feels too close to the bone, you just skip it or edit it.

Topic of the Week: Building Products vs. Running Agencies

Stu Green has lived both lives. He's run agencies, built products from scratch, and sold 2 SaaS businesses. So what's the difference between building for clients and building for yourself? Quite a lot, as it turns out.

Start by Solving Your Own Problem

Both of Stu's successful apps, a project management tool and HourStack (a time management app), started the same way: he needed something that didn't exist. The project management tool grew out of running his own consultancy. HourStack came from juggling small children and fragmented work hours, and wanting a way to visualize and stack little blocks of productive time.

If you're genuinely your own best customer, there's a good chance others like you exist. And if even 2 or 5 or 10 of them show up, you've got the start of something real.

The Myth of "I One-Shotted This"

AI has made it dramatically easier to build apps, but Stu is refreshingly honest about the gap between a demo and a product. Sure, he cloned entire apps in a single prompt and it looked great. But behind that impressive facade? Hours of iteration, hosting setup, video infrastructure, S3 servers, and a stack of decisions that require real product-building experience.

The people posting "I built this in one shot" on X are technically telling the truth, but they're showing you the Hollywood set, not the house behind the door. Getting from prototype to something you can actually charge money for still takes professional knowledge. You need to know what questions to ask, which answers are good, and when you're being led down a rabbit hole.

Two Tiers of AI Tools

Paul and Stu landed on a useful mental model: there are essentially 2 categories of AI building tools.

1. Tools for everyone: Platforms like Lovable or Figma Make that let anyone create a basic app or prototype. Great for personal use, proof of concepts, and quick experiments.
2. Tools for professionals: Things like Cursor and Claude Code that enhance a developer's ability to build production-quality software faster and better, but still require real expertise to use well.

Think of it like desktop publishing in the '90s. When it arrived, everyone panicked that graphic designers were finished. Instead, regular people made terrible flyers with Comic Sans, and the professionals used the same tools to produce better work, faster. AI-built apps are following the same pattern.

The 3-Stage Development Model

Paul offered a framework for thinking about where AI fits in the build process:

1. Prototype and proof of concept: Anyone can do this with AI tools. Great for validating ideas quickly and cheaply.
2. The production build: This still needs a professional. Security, scalability, accessibility, solid architecture: these are non-negotiable if people are paying to use your product.
3. Post-launch iteration: Once a professional has laid a strong foundation, less technical people can step back in and make tweaks and improvements with AI assistance, because they're working within a well-built structure.

A Revenue-Sharing Model Worth Considering

Stu floated an interesting agency model: instead of charging a client the full upfront cost to build their app, what if you took partial ownership? The client pays a smaller retainer and upfront fee, you build and host the product, and you share in the revenue. If the app takes off, everyone wins. If it doesn't, your exposure is limited.

The key is picking partners carefully. They need to bring the marketing and audience side of the equation, because your job is the infrastructure and development. It's a model that silverorange, a Canadian agency, used successfully with e-commerce clients years ago, and it still holds up.

When to Sell

Stu sold both his apps when they hit what he calls "the plateau," that point where growth flattens and your churn rate starts catching up with new customer acquisition. At that stage, you either invest heavily to push through (hiring, scaling infrastructure, customer success teams) or you sell to someone who wants a product with proven recurring revenue.

For Stu, as a creative who'd rather build new things than manage database consultants and customer support, selling was the obvious choice. He used brokers both times, people who handle the paperwork, the letter of intent, and protect both sides of the deal. They take a cut, but they also sent chocolates, so it all evens out.

Finding the Right Ideas

With everyone building apps now, how do you pick the ones worth pursuing? Stu's answer is to not go it alone. Find partners who are excited enough about the idea to invest their time and audience. If you pitch an idea and nobody wants in, that's useful information. If someone does, you've got both validation and a distribution channel on day one.

He tested this with an AI running coach concept, reaching out to local running coaches in Jacksonville. When they responded with polite indifference, he moved on rather than sinking months into a product nobody was asking for.

Read of the Week: AI-Powered Personas

Paul shared his latest obsession: using AI to breathe new life into user personas. He's written 2 articles for Smashing Magazine that walk through the process:

1. Functional Personas With AI: A Lean, Practical Workflow: How to build genuinely useful personas that focus on what people are trying to do, not just demographic data.
2. AI In UX: Achieve More With Less: Broader lessons from using AI across user research, design, development, and content creation.

The approach: take all your research (surveys, interviews, call logs, analytics) plus deep online research from tools like Perplexity, feed it into AI, and generate highly detailed personas, far more detailed than the traditional single-page variety. Then load those personas into a project in ChatGPT, Claude, or Gemini, with instructions to answer questions from the persona's perspective.

The result is something you can consult in every meeting, on every decision. A product team can upload photos of next season's lineup and ask "what would our audience think?" A web team can test wireframes against the personas. Real user research still matters, of course, but this approach makes research-informed thinking available at a frequency and scale that traditional methods never could.

Marcus's Joke

"I tried to steal spaghetti from the shop, but the female guard saw me and I couldn't get pasta."

Courtesy of comedian Masai Graham. And yes, it's exactly as bad as you think.

Find The Latest Show Notes