Oct 30, 2025 · 49m · mixergy
#2283 How Dan Shipper uses AI to build artisanal AI apps
0:00 / 0:00
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Every founder Dan Shipper joins Andrew Warner to discuss building artisanal AI applications, operating a mostly bootstrapped media and software business, and leveraging multimodal LLMs for personal introspection and organizational development.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Andrew holds 33.3% of the talking time here. How this is scored →
Andrew as informed peer 3.9 Guest teaching 3.5 Guest disagreement 1.3 Andrew pushing back 2.3
speaking balance: gold is Andrew, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 43:27 Dan rejects autonomous driving analogy for knowledge work
Dan forcefully pushes back on Andrew's comparison between self-driving cars and AI knowledge workers, arguing that road distributions are static while economic work dynamically rewrites its own rules.
Hardest push from Andrew ▶ 24:05 Andrew criticizes Spiral for generic outputAndrew openly critiques Dan's software Spiral, telling him it outputs generic social posts lacking genuine taste and emotional introspection.
Biggest teaching moment ▶ 41:30 Dan breaks down smuggled intelligence in benchmarksDan explains the hidden human labor required to formulate, prompt, and evaluate LLM task benchmarks, demonstrating that evals themselves smuggle massive amounts of human intelligence.
Andrew holds their own ▶ 31:21 Andrew cites Sujan Patel on unbundling Excel templatesAndrew validates and expands upon Dan's Excel SaaS thesis by recalling a concrete founder interview where spreadsheet templates were turned into full standalone B2B products.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Andrew as informed peer | Guest teaching | Guest disagreement | Andrew pushing back | Why |
|---|---|---|---|---|---|---|
| Every's Business Metrics and Product Ecosystem | 4 | 2 | 1 | 2 | Andrew recaps Every's product suite from memory and probes into their bootstrapping claims. Dan clarifies their 'sip seed' funding structure and ARR numbers. | |
| Organic Product Incubation and AI-First Culture | 3 | 4 | 1 | 2 | Andrew asks how Dan validates what to build among competing Whisper-based apps. Dan explains Every's internal AI-first experimentation workflow. | |
| The Genesis and Differentiation of Monologue | 3 | 4 | 2 | 3 | Andrew pushes on whether Every does market research or deliberate differentiation. Dan pushes back gently, arguing differentiation emerges organically from personal taste and multilingual usage patterns. | |
| Serving Audience Needs through Creator-First Alchemy | 5 | 3 | 1 | 2 | Andrew cites Rick Rubin's creative philosophy to explore creator-first vs customer-first mindsets. Dan agrees and shares personal habits around journaling with GPT-4V OCR. | |
| Deciphering Complex Philosophy with ChatGPT Voice | 3 | 2 | 0 | 2 | Dan discusses using ChatGPT voice mode to parse Kierkegaard's dense philosophical writings. Andrew sympathizes and complains about Siri's limitations on Apple Watch. | |
| Mobile AI Ecosystems and AI as a Mental Health Tool | 3 | 3 | 1 | 2 | Andrew shares his skepticism of AI as a therapist, rating it a 2/10. Dan explains how he uses it as an asynchronous journaling assistant between human therapy sessions. | |
| Mining Meeting Transcripts and Granola for Cultural Insights | 4 | 4 | 0 | 2 | Dan reveals his prompt recipe in Granola to auto-generate unspoken cultural handbooks from meeting transcripts. Andrew remarks on Dan's deep introspection and marketing savvy compared to Granola itself. | |
| Critique of Spiral and the Need for Taste in AI Writing | 6 | 1 | 1 | 6 | Andrew delivers direct critical feedback on Spiral, calling its output generic and lacking heart compared to Dan's personal taste. Dan graciously confirms that this exact critique motivated Spiral v2. | |
| Architecting an Agentic Ghostwriter with Interviewer Agents | 3 | 5 | 1 | 1 | Dan outlines the architecture of Spiral v2, detailing the CLI-inspired interface and the interactive interviewer agent that extracts authentic context. | |
| In Defense of AI Wrappers and the Excel SaaS Analogy | 6 | 4 | 1 | 1 | Dan defends AI wrappers by drawing an analogy to how B2B SaaS unbundled Excel spreadsheets. Andrew builds on this by referencing his past interview with Sujan Patel on spreadsheet-based time tracking tools. | |
| Collaborative Prompt Sharing in Claude Projects | 4 | 3 | 1 | 2 | Andrew inquires about collaborative prompt maintenance within Claude Projects. Dan describes their shared copyediting projects and discusses upcoming file-use and agentic capabilities. | |
| Scaling Team Size and Preserving the 'Every Smell' | 3 | 3 | 1 | 2 | Andrew notices Every has grown beyond its publicized sub-10 headcount. Dan details how hiring dedicated readers preserves company culture across 15 to 20 team members. | |
| Defining Culture: Explicit Playbooks vs. Implicit Values | 4 | 6 | 2 | 2 | Andrew brings up Hugh Hefner writing explicit company playbooks. Dan counters that implicit culture is more authentic, then educates Andrew on 'smuggled intelligence' in LLM benchmarks. | |
| Dynamic Knowledge Work vs. Autonomous Driving Constraints | 5 | 4 | 5 | 5 | Andrew challenges Dan's job-creation optimism by citing Waymo replacing drivers. Dan rejects the analogy, arguing driving has bounded static rules whereas knowledge work continuously morphs. | |
| Software as Content: Distribution Strategy for Solo Builders | 3 | 4 | 1 | 1 | Dan provides strategic distribution advice for solo developers, framing software as cheap content to be published continuously across communities. |
Statements from this episode (19)
Dan Shipper: Every generates $1.3M ARR across nearly 7,000 paid subscribers
“I can tell you we have about almost 7000 paid subscribers across all the apps in the media. And altogether we do about like 1.3 million in ARR. Growing really fast. We grew ARR 45% this quarter.”
Shipper: Every raised $2M 'sip seed' from Reid Hoffman, Starting Line
“We raised 600 K in 2020, which is when we started the business. And then we just raised another round From Reed Hoffman and starting line DC. And that was a two million dollar round, but we raised it as what I call a sip seed round. So they have committed up t…”
Shipper: Every probably will not raise a standard VC Series A
“And I think the investors, this, in this round also are aware that we're probably not going to go raise, you know, ten million ten million dollar series A from Sequoia or whatever. I don't even know if ten million, maybe ten million is too small now for a seri…”
Shipper: Every requires an AI-first approach across all roles
“We're all using AI to do everything from coding to writing to editing to design to marketing, all that stuff. And if you're not doing it with AI first, you're kind of weird which I think is the reverse of how things are in pretty much every other company.”
Dan Shipper: Monologue's initial version was built over a single weekend
“Naveen, who's the GM of Monologue, just was like, I don't know. I like, I want an invitation app. The ones that I, the ones that I've been using, like aren't quite right for me. And he just like built it over a weekend and then demoed it and everyone was like,…”
Dan Shipper: OpenAI awarded Every a plaque for processing 10 billion tokens
“Like yesterday I was at Dev Day, OpenAI Dev Day, and I got this like cool plaque that was like, you've processed ten billion tokens through, through OpenAI, which is, it was so cool.”
Shipper: ChatGPT excels at translating dense philosophical texts into plain English
“And I think Chad DBT does a really good job of translating, again, the different forms of English, translating from Kierkegaard's he calls it dialectical algebra, Kierkegaard's dialectical algebra to, like, normal, normal people speak, and that helps me get mu…”
Shipper: Prompting LLMs on internal meeting transcripts generates honest company handbooks
“I have a recipe on there that I made that I really liked it, that was really cool, which is, Go through all of my internal meetings and write our company handbook from what you see. And I want you to capture all of the unspoken ways that every works as a compa…”
Shipper: Claude Code represents the future of programming by eliminating code editors
“Over the summer, I started using cloud code and I was like, wow, this is the future of programming. And they did something really special here, which is they just eliminated the text editor. Like you don't need that. You don't need the code editor. You just ne…”
Shipper: Effective ghostwriters interview subjects instead of instantly generating output
“Good ghost writers. When you say, Hey, I want you to tweet about this or blog post about this ghost, ghost writers. Don't go ahead and say, great. I did it. Here it is. What they do is they say, okay, well, what was interesting to you about this article? What …”
Shipper: Current AI writing tools fail to interview users first
“And I think no AI tool really is doing that because they're very focused on just give me an answer right now.”
Dan Shipper: Big opportunity will come from unbundling ChatGPT into purpose-built apps
“And I think there's something very similar happening and going to happen with AI is people have to use ChatGPT and like realize that this new sort of workflow is possible. And then there will be a big opportunity to unbundle those workflows into purpose-built …”
Shipper: Every uses Claude Projects for developmental and copy editing
“Basically we have a cloud project with all of this stuff in it. That can do copy editing, can do developmental editing. So like helping you figure out like which ideas are interesting and why, or where's the hook for the article, that kind of stuff.”
Shipper: Claude is near capable of dynamic agentic copy editing via Notion
“Claude is very, very good right now at, you know, longer horizon agentic tasks. And so I would, And it now also has file use. So I would bet that we're getting close to, if not currently at a place where if you pointed it at a notion page and said, read the no…”
Shipper: Every employs 15 to 20 full-time staff across six business lines
“We are more than 10 people. We are about 15, between 15 and twenty-ish, it's a little hard to tell, full-time. We have six business lines now, so yes, a lot.”
Dan Shipper: GPT-4 and Claude Opus beat human experts in blindly judged benchmarks
“They found that GPT-V is better than experts, like blindly judged better than experts at completing these tasks like 40% of the time, and I think Claude Opus is at 40, at 49%”
Shipper: AI evaluations smuggle in human intelligence through task setup
“And to do all of the things you need to do in order to set up the eval is a job. And so the eval is what I'm saying is smuggling a lot of human intelligence into it because in order to take a like complex and dynamic activity, like doing a job and turn it into…”
Shipper: Driving is easier to automate than knowledge work because roads are static
“The difference is that the roads, even though things are complex and dynamic and all that kind of stuff on the road, the roads are the same. Like that's the same roads. You're getting from point A to point B and the way that that works, like you, there's a zil…”