ProductLed Podcast

Day 55, and I'm past the halfway point of building an AI product in 100 days. This one looks back at the first 50 days and lays out the plan for the last 45. The biggest lesson so far: I thought building the app would be the hard part. It isn't. Figuring out what to build, how it fits into the business, and what the next offer should be is still way harder. The PLG analyzer now gets about 8 out of 10 recommendations right. I'm not willing to trade quality for cost or speed, so getting that to 99% comes first. The money side hasn't worked yet. The analyzer was meant to lead to working with a ProductLed operator, but most of the people using it run small companies, and they're hard to serve at that price point. Meanwhile the free product costs about $1,300 a month to run, developer included. So the plan for the next 45 days is one new enterprise client, won through blind calls: analyses of companies I already want to work with. Every one of them also trains the analyzer to give you better recommendations. IN THIS EPISODE 00:00 Day 55, and what I'm building 02:09 Your feedback, from "about as good as ChatGPT" to three scans in a row 03:05 Building the app was never the hard part 04:01 Highlights: shipping, Matt Pocock's skills, new friends 05:20 Lowlights: why 8 out of 10 isn't good enough 06:58 Building an app has no training plan 08:33 The game plan for the last 45 days 09:36 Why the operator offer hasn't worked 10:45 A free product that costs $1,300 a month 11:22 Pay for itself, fit the business, fix quality 13:08 How my advisory work actually runs 14:24 Blind calls: account-based marketing that trains the AI 15:29 One enterprise client and 99% good outputs 16:36 Next: turning internal workflows into products 17:54 What's your take? MENTIONED Run the free PLG analyzer: https://productled.com Matt Pocock's skills: https://github.com/mattpocock/skills Wes Bush on LinkedIn: https://www.linkedin.com/in/wesbush/ What's the biggest challenge you've hit building an AI product?

In the last episode I told you I'd hit a wall building an AI product in 100 days. This is what happened after I asked for help. So many of you replied, and a handful of those replies changed how I build. One pointed me to Matt Pocock's skills on GitHub, which exposed an embarrassing habit: session after session I'd been telling Claude "ship it" and assuming the change went live. It didn't. Others told me I was trying to boil the ocean. So the PLG analyzer now finds one opportunity worth your week instead of three, and a judge agent decides what matters most rather than a formula weighing everything the same. My own rating went from a 4 out of 10 to a 6. Next up is cost and speed. A free assessment still costs 40 to 60 cents and takes about two minutes, and I want it under 60 seconds. I close on something that has little to do with the build: using AI to write without ending up sounding 80% like yourself. IN THIS EPISODE 00:00 Day 39, and why updates are now every two weeks 00:47 Stuck, and what happened when I asked for help 02:01 Matt Pocock's skills, and the 11 I use all the time 02:37 I said ship it, and nothing shipped 03:41 Stop trying to boil the ocean 04:16 Why a formula misses the biggest problem on your site 04:57 From three opportunities to one worth your week 05:46 Why an LLM product needs a judge agent 06:58 Eric's idea: a playbook behind every recommendation 08:14 The next two weeks: cost and speed 09:20 From a 4 out of 10 to a 6 10:08 What I learned 10:42 Sound like yourself when AI writes for you MENTIONED Run the free assessment: https://productled.com Matt Pocock's skills: https://github.com/mattpocock/skills Wes Bush on LinkedIn: https://www.linkedin.com/in/wesbush/ What's one skill or tool someone recommended that changed how you build?

An honest look at building an AI product in public and why it currently earns a 4 out of 10. The project analyzes websites for product-led growth opportunities in under 60 seconds, but speed, model costs, and messy code are only part of the challenge. The real puzzle is turning years of expert pattern recognition into useful recommendations that survive every edge case.

The podcast has been quiet for a couple of months. This episode is why. I'm building a new AI-first product at ProductLed, in public, over 100 days. It launches November 19th. You'll be able to use it while I build it, and rip it apart, and I'm sharing everything as I go, including the revenue. Before any of that, I wanted to talk about how I'm going to attack it. I recently finished a full Ironman. It ends with a marathon, and that's after the 3.8km swim and the 180km bike. Training for it changed the way I take on anything large, and I'm running the same structure on this build. I call it full Ironman mode, and this episode walks through all 16 parts of it. Some of it is obvious. Most of it is not. The part that surprised me most was standards, because when you look at the standards you set for a goal, they should make the goal inevitable. That is the difference between hoping you finish and knowing you will. Whatever your next 100 days hold, this should be useful. IN THIS EPISODE (04:15) Get crystal clear on your vision (04:47) Name the core problem you're actually solving (05:28) Define the end game, and my three success criteria for this build (07:42) Tap your network for people who have already done it (08:54) The identity shift, and the line I write every morning (10:35) Pick the date (11:39) Why one why is never enough (12:53) Get a coach or an advisor (14:37) Create space, and audit what pulls you away (17:40) Find peers in the trenches (18:50) Name the price you're willing to pay (19:54) Resources are accelerants (20:48) A daily plan you don't have to think about (22:09) Standards that make the goal inevitable (23:05) A reward you only get if you finish (24:57) Write your own rules MENTIONED Conquer 100, the documentary about the Iron Cowboy Mickey Allen, CEO at Foldspace, advising on this build What's your next 100 days going to be? Let me know.

Tibo Louis-Lucas, a French indie hacker who built TweetHunter and Taplio, now runs a portfolio of product-led SaaS. He discusses shifting to revenue-first validation. He explains focusing deeply on one platform for creators, an unusual profit-share distribution play, no-call DM feedback loops, and how AI sped up his build-and-iterate style.

Roeland Delrue, CEO and co-founder of Aikido Security, built a developer-friendly cybersecurity unicorn by solving his own security pain. He explains why they rejected sales-heavy buying, embraced transparent pricing and fast time-to-value. Roeland discusses building a multi-product platform, leveraging compliance tailwinds, and where AI is already reshaping pentesting and detection.

Jaleh Rezaei, co-founder and CEO of Mutiny who rebuilt an eight-figure SaaS into an AI-native company. She explains why running SaaS and AI together caused friction. Short recounts of the 90-day pivot: shrinking the team, migrating customers, and refocusing culture on speed. Talks product-led growth, tactile AI demos, and why experience generation plus analytics form a defensible AI moat.

Jeremy Clarke, bootstrapped founder who grew WebMerge to $5M ARR and sold it for nine figures, now building Quinn and a small venture studio. He recounts turning a PDF tool into an integration-driven growth engine. He explains why staying lean, handling support, and delaying hires mattered. He contrasts WebMerge’s margins with the tougher AI landscape and outlines Quinn’s assistant-focused positioning.

Jeff Wang, CEO of Windsurf who steered the company through a high-stakes crisis and acquisition. He recounts the 72-hour leadership scramble, why Windsurf’s autocomplete, context engineering, and agent breakthroughs drew big tech interest. They discuss moving to multi-agent workflows, the shift from free extensions to enterprise GTM, and how token economics and playbooks drive real AI adoption.

Edith Harbaugh, co-founder and CEO of LaunchDarkly and former engineering leader, talks about the origins of feature management and the TripIt insight that sparked the company. She covers creating a new category, balancing product-led growth with enterprise sales, keeping a free tier, returning to lead the company, and positioning LaunchDarkly as runtime control for the AI era.