How I AI: the podcast that shows how people really use AI
How I AI is Claire Vo's free podcast where guests share their screens and show exactly how they use AI. Here's what it covers and how to borrow its best ideas.

How I AI is a free AI podcast hosted by Claire Vo, the founder of ChatPRD, where each guest shares their screen and walks through one specific way they use AI at work or at home. Episodes run about 30 minutes, and the point is that you can copy the workflow the same afternoon. If you typed "how I AI" because you want to learn how to use AI yourself, the show's a good place to start, and I think it's one of the more useful AI podcasts around because it shows the work instead of talking about it.
The rest of this piece covers where the show came from, a few episodes worth your time, where the format falls short, and what you can borrow from it even if you never press play.
The background: a podcast built around screen shares
The first episodes went out in April 2025. One of the earliest, from April 22, had Gumroad CEO Sahil Lavingia showing how he used v0, Devin and Cursor to turn two-week projects into two-hour builds. A week later, former New York Times columnist Farhad Manjoo walked through using ChatGPT as a brainstorming partner, a "super thesaurus" and a first reader for his drafts.
That set the pattern. Most talk shows about AI are panels: people debating where the technology is heading, what it'll do to jobs, which lab is winning. How I AI is closer to a cooking show. A guest opens their laptop, shares the actual prompts, and you watch the result appear (or not). The show's own description promises "tips/tricks/workflows you can copy immediately", and that's a fair summary of the format.
It's grown fast. The episode library on the ChatPRD site lists more than 120 episodes, and the show's YouTube channel has around 129,000 subscribers. You can listen on Spotify or Apple Podcasts, but because so much happens on screen, I'd watch the video version whenever you can.
There's also a companion library that I suspect most listeners miss. The team turns episodes into 314 step-by-step workflows, each tagged by difficulty (beginner, intermediate or advanced), job area and tool. That's the part I'd bookmark.
Real examples from the episode library
The range is wider than the name suggests. Plenty of episodes are about coding and product management, which makes sense given the host runs a product tool, but recent ones have drifted into ordinary life.
Take a mid-November episode: Scott Farrar uses Claude Code, plain Markdown files and the note app Obsidian to run what the show calls a "household brain". It holds appliance records, insurance questions and repair notes, and he deliberately keeps himself in charge of decisions. The matching beginner workflow, "Manage a Household Project Using an AI Knowledge Base", ends with a checklist of next steps and sometimes a drafted email, stored in one file per project.
William Imoh's episode is another good one. After a 1,000-day Duolingo streak he still couldn't hold a conversation, so he used Codex to build Muro, a voice tutor that refuses to switch to English. It's a sharp example of picking one narrow problem you actually have.
Then there are the reviews. Claire Vo has compared personal agents (Grok, Dot, Muse and Claude Duet) across nine real tasks and concluded there was "No Universal Winner". Her verdict on Claude Duet was blunt: the design stands out, but slow responses and weak follow-through made it hard to keep using. I like that she says so on air; it's rarer than it should be.
To make this concrete, here's a small table I put together mapping a few episodes to the habit underneath each one. The habit is the transferable bit.
| Episode guest | Tools shown | The habit you can copy |
|---|---|---|
| Sahil Lavingia | v0, Devin, Cursor | Prototype fast, then hand the boring build steps to an agent |
| Farhad Manjoo | ChatGPT | Use AI as a first reader, not a ghostwriter |
| Scott Farrar | Claude Code, Obsidian | Keep your own notes in files an AI can read |
| William Imoh | Codex | Build a tiny tool for one stubborn personal problem |
| Ezra Mechaber (Figma) | A shared repo of 31 AI skills | Save context once so you stop re-explaining it |
Benefits and limitations
The biggest benefit is specificity. Most advice on how to use AI stays at the level of "write better prompts", which is true and nearly useless. Watching someone's screen shows you the dull parts: where they paste context, when they throw away an answer, how long a task really takes.
That matters, because research backs the idea that AI helps on some tasks and hurts on others. In a Harvard Business School study with Boston Consulting Group, 758 consultants using GPT-4 completed 12.2% more tasks and worked 25.1% faster on tasks inside the AI's abilities. On one task picked to sit outside them, consultants with AI were 19% less likely to get the right answer. The authors call this a "jagged technology frontier". A show full of concrete workflows is, in effect, a map of where that jagged edge sits for real jobs.
Now the limits. The guests skew heavily toward founders, engineers and product managers in tech, and their setups often assume you're comfortable with a code editor or a terminal. If you're a nurse or a shop owner, many episodes won't translate directly.
There's also a commercial angle worth keeping in mind. The show lives on the ChatPRD website, carries sponsors (Cursor credits were on offer when I looked), and the host reviews products that compete for the same users. I don't think that makes the reviews dishonest, and she's critical when she needs to be. Still, it's a business with its own product to sell.
And demos are curated. A workflow that looks smooth in a 30-minute edit probably took several failed attempts first.
Turning episodes into your own AI habits
The question behind "how I AI" is usually "how can I use AI?" The show's answer, which I agree with, is to start from a task you already do every week rather than from a tool.
That matches how people actually use chatbots. OpenAI's study of 1.5 million ChatGPT conversations, written with Harvard economist David Deming, found that about three-quarters of conversations focus on practical guidance, seeking information and writing. Around 49% of messages were "Asking" (advice and answers), 40% were "Doing" (drafting, planning, coding) and 11% were "Expressing". Roughly 70% of use wasn't work-related at all.
So here's how I'd borrow from the show, in order:
- Pick one recurring task that eats at least an hour a week, such as meeting notes, a weekly report or meal planning.
- Find a matching workflow in the library, filtered to beginner, and follow it exactly once before changing anything.
- Write down the context you keep retyping (your role, your audience, your style) and save it in a file or a custom instruction.
- Check every output that touches money, health or other people's names. The HBS result above is why.
- After two weeks, keep it or drop it. Most experiments should probably be dropped.
If you're starting from zero, our step-by-step guide to using AI covers the basics, and the prompt engineering guide helps with the writing side. For picking a chatbot, the Claude vs ChatGPT comparison is a sensible place to begin. And if your interest is closer to audio, as this show's is, our Suno AI review and our list of scary AI voice tools show what's possible on the creative end.
The next few months for the show
Start with what the episode list shows. The recent episodes lean toward agents: tools that act on your behalf across email, files and browsers instead of just answering questions. Claire Vo's latest solo episode was about token fraud, which cost ChatPRD tens of thousands of dollars and pushed her to kill free trials.
What I'm guessing at is the direction. My bet is that the show will spend more time on personal systems (household brains, private knowledge bases, custom tutors) because that's where agents seem to work best today: narrow jobs, owned data, a human signing off. I'd also guess the fraud episode is a hint that cost and abuse will become a bigger topic for anyone building with AI.
For you, the takeaway is simpler. You don't need to watch every episode. Watch two that match your job, copy one workflow, and judge it on whether it saves you real time.
FAQ
How do I use AI?
Start with a task you already do often and ask a chatbot such as ChatGPT, Claude or Gemini to help with one step of it, like a first draft or a summary. Give it context about who you are and what good looks like, then check the result. The How I AI workflow library has beginner-level examples you can follow step by step.
How can I best use AI?
Use it on tasks inside its abilities (drafting, summarising, brainstorming, basic code) and keep your own judgement on anything high-stakes. The Harvard and BCG study found AI users did better on tasks inside that frontier and worse on one outside it, so the best use is probably the boring, repeatable middle of your work.
Who hosts How I AI?
Claire Vo, the founder of ChatPRD, an AI tool for product managers. Guests have included Gumroad's Sahil Lavingia and writer Farhad Manjoo.
Is How I AI free?
Yes. Episodes are free on YouTube, Spotify, Apple Podcasts and the show's page on the ChatPRD site, and the written workflows are free to read there too.
Is How I AI only for programmers?
No, though a lot of episodes assume some comfort with code. Writing, language learning and household organisation have all come up, and the workflow library lets you filter for beginner-level guides.



