Play-First AI Fluency
Build real AI intuition by playing — invent fun side projects, use everything, and share the artifact not the doc.
- Difficulty
- Easy
- Time to result
- ~weeks to results
- Steps
- 4
- Confidence
- 90%
The fastest way for a leader or team to develop genuine AI intuition is not to read about capabilities but to play with them experientially. Leaders should invent small, fun side projects that force real hands-on use of many AI products (including competitors'), lead by example by sharing the actual artifacts they build, and give teams explicit permission — even a full week off meetings — to explore. Playing (in the psychological, curious sense) teaches both what the models can do and what product form factors they enable.
Origin
Howie Liu's practice, reinforced on the show by two ideas he cites: Dan Shipper's (Every) signal that AI-adopting companies have CEOs who use ChatGPT/Claude daily, and Nick Turley's (Head of ChatGPT) point that you don't know what AI can do until it's out and people try it.
Core principles
- 01Reading a TechCrunch writeup or a tweet about a capability is not enough — AI is something you have to play with to understand.
- 02There's a difference between 'checking the box' to get a job done and coming in with curiosity and exploring; play is both more fun and teaches more.
- 03Understanding the raw model is different from understanding the product form factors (tool-calling, agentic workflows) the model can be placed into.
- 04In AI there are major new releases weekly, not yearly — staying abreast requires continuous experiential exposure.
How to run it
- 1
Use as many AI products as you can, including non-yours
Deliberately try new releases across the ecosystem — Runway's world engine, Sesame's voice demo, a new GPT model the day it ships — even when they aren't core to your product, purely to get a feel for what's out there.
Pro tip Study competitors' new releases the way SaaS founders once studied Salesforce's annual releases — except the cadence is weekly, not yearly.
Watch out Twitch-clicking a demo for ten seconds doesn't count; you have to actually run it on real inputs.
- 2
Invent a fun side project as a forcing function
Give yourself a concrete, enjoyable reason to use the tools end-to-end. Liu builds things like a comedic short: deep-research a topic in ChatGPT, generate a script, turn it into a HeyGen avatar video, download and play it — an ~hour project done purely for fun.
Pro tip Pick a project that's useful or fun to you personally — a real problem you want solved makes you push the tools far past golden-path demos.
- 3
Actually build, don't just observe
Studying other products is necessary but insufficient — like learning industrial design by examining great chairs, you must then build your own, and another, and another. Build a Yelp-style app with a synced map+list view; discover which parts are hard and what affordances solve them.
Pro tip Daytime job projects count too: prototype several directions of a real feature to see how each feels rather than debating in a doc.
Watch out Only looking at others' work, never building, leaves you unable to judge what's technically one-shottable versus what needs a human review loop.
- 4
Lead by example and grant permission
Share the actual artifacts you make — a Replit landing page link, a deep-research report, a prompted-into-existence memo — instead of writing a doc, so the team sees exactly how you're using AI. Explicitly tell people they can cancel meetings for a day or a week to go play with every relevant AI product.
Pro tip Give people 'the ultimate excuse' — permission to say the CEO told them to block out a week for exploration — to remove the guilt of stepping away from delivery.
Watch out If leaders only tell teams to 'use AI' without modeling it and sharing real outputs, it stays an abstract mandate nobody internalizes.
In the wild
To force real use of multiple AI products, Liu runs deep research on a topic in ChatGPT, has it generate a comical script, turns the script into a HeyGen avatar, downloads and plays the video — a roughly one-hour project done for fun, not for a YouTube business.
→ The exercise gave him hands-on understanding of both the underlying models and the product form factors they can be placed into, which he then shares with his team as links and screenshots to model the behavior.
Common mistakes
Reading about capabilities instead of trying them
You can't get the real feel of an AI capability from a tweet or a TechCrunch article; without hands-on play you misjudge what's possible and design the wrong product.
Mandating AI use without modeling it
Telling a team to 'play with AI' while hiding your own usage and shipping traditional docs gives them no concrete pattern to copy, so the mandate never becomes practice.
Is it for you?
Best for
A founder, product leader, or team member trying to rapidly build genuine AI intuition and product sense in a fast-moving landscape.
Not ideal for
Teams in a pure delivery crunch with a locked, well-understood scope where open-ended exploration would just be a distraction.
From the transcript
“If you want to cancel all your meetings for like a day or for an entire week and just go play around with every AI…”
“coming up with like these different like fun weekend projects is a really useful construct to like force myself to actually try these products in…”
“you have to try it to really get a sense of like what it is”
From the episode
How we restructured Airtable’s entire org for AI
Howie Liu (co-founder and CEO)