# 006 Don’t Let AI Do Your Thinking ![[images/Pixels-and-Priorities-artwork.jpg]] Part of [[Pixels & Priorities]] (listed in [[Pixels & Priorities - Audio Episodes 001-014]]). Previous: [[005 Designing at the Speed of AI - Part 2]]. Next: [[007 Signal Over Noise - Designing with Intention in an AI World]]. - Show: Pixels & Priorities, hosted by Metsy Rose and J Schuh (RSS author field) - Episode: 006, published 2026-05-05 (RSS pubDate Tue 05:00 -0500) - Audio: https://dts.podtrac.com/redirect.mp3/mcdn.podbean.com/mf/web/n9mne774uk65qfxk/PP_E006_AI_Fluency_Edited.mp3 - Podbean page: https://pixelsandpriorities.podbean.com/e/don-t-let-ai-do-your-thinking/ - Length: 36:30 (2,190.9 s measured from the mp3). The RSS `itunes:duration` says 2190 s (36:30), matching the file. - RSS description: "In this episode, Metsy and J unpack what AI fluency really means—and what it doesn’t. They explore the pressure to become an “AI unicorn,” why mastery isn’t the goal, and how human judgment and problem-solving remain your greatest advantage." - Transcript: faster-whisper (medium, int8, CPU) on the mp3; VAD on; beam size 5; about 5,700 words after re-transcribing stretches the VAD pass dropped or thinned (about 0:00-1:12, 1:58-2:22, 12:45-13:30, 16:50-17:22, 19:24-19:42, 21:48-22:02, 23:34-24:00, 24:40-26:12 and 28:44-29:00) with VAD off. The feed also links a publisher transcript (Descript VTT, `PP_E006_AI_Fluency_Edited.vtt`); it was used only to check names and confirm where the Whisper pass had dropped words. All quotes come from the Whisper transcript. Speakers are not diarized. Lines are attributed where the content makes it clear (J: After Effects, ukulele, Midjourney, UX research team; Metsy: PM strategy, "my really old developer days", the fellow-PM story); otherwise the note says "one host". A few leftover non-English tokens from the first pass were dropped from quotes. Whisper misheard names ("Metzy Rose", "menci", "Jay Xu"); correct spellings are **Metsy Rose** and **J Schuh**. ## Hosts - **Metsy Rose** - co-host; product manager and strategist (show open). Learns AI as an individual, reading widely and paying for tools, so she's ready to bring it into an organization. Has used DeepSeek, ChatGPT, Copilot, Gemini and Leonardo.ai, and debugs prompts like in "my really old developer days". Says she was "back in UX" earlier in her career; argues PMs should stay strategic and resist AI for the sake of AI. - **J Schuh** - co-host; design strategist, educator, and creative mind (show open). Has used After Effects since version 3.1 in the 1990s, taught himself ukulele during COVID and is now learning guitar, and learned Midjourney by burning through a month of credits in a week. Was embedded on a UX research team while doing design strategy, where his leader Brian Sullivan taught him to wait for five or six data points. - No guests. ## Outline (timestamps are mm:ss) - **0:00 Intro.** Show open: Metsy Rose (product manager and strategist) and J Schuh (design strategist, educator, creative mind). No guest this week. - **0:27 Recap and topic: AI fluency.** The hosts look back at the last episodes' AI conversations (how do you learn AI, and how will that knowledge help your career?) and pick AI fluency as today's topic: most companies are asking how to get their employees up to speed with AI, and there is a lot of confusion about what that means. - **1:15 What organizations really want.** Organizations want more speed, quality and efficiency, which goes back to processes and systems, good stakeholder communication, a good roadmap and understanding users' pain points. Those are hard enough; now AI is "thrown in" on top of an already complex job. - **2:10 Do I need to become a data scientist?** The worries people have: learning to code, picking up a "Python for Dummies" book, knowing every algorithm and data construct. There is pressure to know a little bit about everything, and the more you know, the more you see how much is left to learn. AI changes faster than anyone can learn it. - **3:07 What AI fluency is not.** You do not need to become a data scientist (and data scientists differ from data engineers). Learning to build agents and LLMs from scratch is fine if you feel called to it, but expertise comes in years, sometimes decades. - **4:14 After Effects since 3.1.** J has used After Effects since version 3.1 in the 1990s and still doesn't know every part of it; you could spend 20 hours on a weekend mastering one plugin, and AI is far bigger than that. On LinkedIn: "is that a new AI that they just announced?" - **4:55 Take the big one off the table.** Please do not expect to build your own language model or have deep, thorough knowledge of them. If you have the time, space and energy, go ahead, but that depth is not what fluency means. - **5:33 The light switch analogy.** You can learn everything about electricity, ohms, amps and circuits, but fluency is being able to turn the light on and off so you can do what you need to do. - **6:14 Good at, not good at, and which room.** Fluency starts with understanding what AI is good and not good at. Next, figure out which AI you need to learn so your own expertise gives value back to the organization, i.e. know which room needs the light. Then find a low-risk AI tool that gives quality output on a specific problem you have right now. - **7:30 Fluency costs money.** There are free versions, but going further usually means at least $20 a month, video gets expensive, and tokens add up with lots of trial and error. Prompting varies from tool to tool (ChatGPT versus Copilot), and these are only individual costs; moving fluency into an organization is a different realm. - **8:28 Stacks of AIs on LinkedIn.** Impressive posts often list four or five AIs used for different purposes, which can feel overwhelming quickly. - **8:55 How did you start? (Metsy).** J asks; Metsy (attribution by content) reads widely, balances different takes against what she actually experiences, and pays for AI tools so she can learn them herself. She checks experiments against what she already knows: is it helpful, "is it bringing something new or is it just shiny?" For now she is learning as an individual so she is ready to bring it into an organization. - **10:00 Tools change (J).** Design tools went from Sketch to Figma, the current "flavor of the month", and something else could replace it within a year or five. AI lets startups enter the market faster than ever, so the landscape may get more volatile. - **11:02 Learning like an instrument: the ukulele.** J taught himself ukulele during COVID: daily practice, videos and tutorials. He can play Happy Birthday for friends and Hotel California, but it took intentional practice. He learned Midjourney the same way: bought a month of credits and used them in a week trying depth of field, aspect ratio, point of view (bird's eye, worm's eye), close-up to extra-wide shots and 35/45/55 mm lenses, then took results into Photoshop and Firefly. - **12:31 One instrument at a time.** He didn't learn ukulele, guitar and banjo at once. Five years later he is picking up guitar and finds the four strings it shares with the ukulele carry over; now he is working out the other two. "That to me is an approach that I've used my entire life in terms of learning." - **13:21 Pick one tool, then transfer (Metsy).** Access to AI tools is easy, but you need focus: pick one tool, learn it, then carry that knowledge into the next. Metsy used DeepSeek, ChatGPT, Copilot and Gemini, then tried Leonardo.ai for video and got unexpected results from prompts that worked elsewhere. Instead of stressing, she approached it like her old developer days: troubleshoot and debug the prompt piece by piece. - **14:56 Built-in variability (J).** A tuned ukulele sounds the same every time; AI does not. A prompt that worked in January won't give the same image in June, or even the next day with the same prompt and model. Humans love consistency. "Now she has four feet." He still sees six or seven fingers and weird hands in image tools. - **16:39 A paradigm shift and an equation.** This will change how people work in a way he doesn't think history has seen. Their equation: domain expertise multiplied by AI fluency equals leverage in your career and organization. - **17:19 Creatives' resistance and the job market.** Many artists resist AI because of scraping of literature, design and fine art without credit. But as with computers in the 1980s and 1990s, he expects it to be harder to get a job without AI fluency; companies expect it in your toolkit, for PMs and designers alike. "That's what I'm hearing on a day-to-day basis." - **18:20 A great time to be early.** The other host agrees: knowledge and tools are widely available, so now is the time to build fluency. Computers were just reaching colleges when this host was in college, and being there for that whole growth was a benefit. You are early in the lifecycle; nobody expects you to be an expert, and you may already know things leadership doesn't have time to learn, so be the thought leader who merges your domain expertise with AI. "Just be a part of the knowledge flow." - **19:50 How should PMs approach AI? (Metsy).** J asks for the PM view. Metsy: if you can carve out time for strategy, stay in that lane with AI. Chase tools for personal learning, not at work; keep asking where users struggle and where AI can help without adding overhead. Resist building something AI just to say the product has AI; weigh risk and cost. It magnifies what PMs should already focus on. - **21:48 The AI paradox.** People building AI fluency get judged for "cheating" (school taught that outside help is cheating) while their companies push them to use it. There is also fear: why help train the thing that may take my job? - **22:46 Mental models for PMs.** Is it really a time or money saver, or is it lateral? Think about what data the AI needs to access, and the liability if it introduces bias that hurts users or internal teams. Enjoy the learning without getting lost in the hype, and keep your PM checks and balances. - **24:04 Bias and five or six data points (J).** On a UX research team, J's leader Brian Sullivan taught him not to jump to conclusions from one data point; wait for five or six that come together. AI's data comes from humans, so Copilot and ChatGPT outputs are biased by nature. There are more than 150 kinds of bias, including believing you're not biased. AI is not a magic box. - **25:13 PM caution versus UX possibility (Metsy).** As a PM she hears herself using hesitant, "what's the next shoe to drop" language; the role is as much preventing problems as building good things. When she was in UX she loved looking at the good that could be built. She asks J how he sees AI through the design lens. - **26:29 Design lens: explore before you narrow (J).** AI helps you iterate on possibilities. His worry: as PMs get more fluent with design AIs, they may bring a finished idea and say "make this", which already happens with sketches and wireframes. That pushes toward one solution too early. Better: use AI to explore a wide range, then narrow to what best meets users' needs, with design-thinking empathy. "We are designing products for humans, at least not yet." - **28:09 Fast in the wrong direction.** You can go fast in the wrong direction and enjoy the ride: expecting Florida, ending up in New York City. Both have heard excited "I built this" stories and wondered what it is for and who will use it. Personal learning can be anything; at work everything needs a value proposition. - **29:17 How you'll be judged; questions are the skill (J).** AI won't change how companies judge you: speed, quality and impact at year-end, plus human skills like stakeholder conversations and asking good questions. Prompt engineering is really asking great questions, and that takes problem solving; otherwise, garbage in, garbage out. Creative problem solving will top the list. Ted Lasso: "be curious, not judgmental." - **30:47 Design will weigh pros and cons too (Metsy).** She realizes design will face the same pros-and-cons pressure PMs do, because AI opportunities will come to designers too. - **31:02 Talking to your own AI (J).** J told colleagues he foresees people who seem to be talking to a person but are talking to their own AI, asking it to analyze the last meeting and build a presentation. "Is our AI going to back us up in meetings?" Metsy laughs: she offered a fellow PM to join a meeting on cue and turn her camera on with a stern face. - **32:19 Jarvis, and Paul the AI companion.** AI will show up in strange ways that become normal. Iron Man had Jarvis: will employers hire you and your AI, or assign you one when you join, such as "Paul", who keeps you on task at XYZ company? - **33:06 The hammer.** Technology has pros and cons: a hammer can build a house or hurt someone. Human choices about control, trust and use, including nefarious use, will shape society and culture. - **33:59 Wrap: you still bring you.** Bring what you already know to your AI fluency journey: domain expertise, problem solving, critical thinking, curiosity and the ability to weigh pros and cons. The goal of fluency isn't more pressure but clarity about something that will affect the world for good and probably some not-so-good ways. Your expertise matters; AI changes how you apply it. - **35:02 The unicorn, next time.** Nobody can be expert at everything. The "unicorn" idea of being amazing at everything causes stress and anxiety, and that's why we have teams whose strengths multiply. They plan to pick it up next episode (compare [[008 Unicorns, T-Shapes, and the Myth of Mastery]] in the TOC). - **35:36 Why we do the podcast; outro.** Part of the reason they started the show is to build that team with listeners, learning together. J thanks listeners for showing up every week to hear "Metsy and I just look at the world and say huh." Outro. ## Standout quotes (verbatim from the Whisper transcript) - 4:06 "that expertise doesn't come in months, it comes in years." - 4:57 "If you are learning AI, please do not have an expectation that you will build your own language model" - 6:04 "But the truth is fluency is about being able to turn it on and off so you can do what you need to do." - 6:15 "So it's understanding what AI is good at and what it's not good at, I think is the starting point to AI fluency." - 6:48 "Just know which room you need to turn the light switch on for." - 9:42 "Is it bringing something new or is it just shiny?" - 10:26 "And now Figma seems to be the flavor of the month." - 12:27 "This is that same approach. You just dig in and you practice." - 12:31 "I'm not trying to learn the ukulele and the guitar and the banjo at the same time." - 15:31 "there's built-in variability in AI." - 16:08 "as humans love consistency" - 16:26 "Now she has four feet." - 17:10 "domain expertise multiplied by AI fluency equals leverage in your career and with your organization." - 19:42 "I think it's a great time to jump in and not stress too much about it, but just be a part of the knowledge flow." - 21:24 "you have to resist that pull of AI for the sake of AI and stay with your strategy hat." - 22:55 "Is it actually going to be time saver or a money saver, or is it lateral?" - 24:17 "don't jump to conclusions because you have a data point. Wait until you have five or six data points that come together before you jump to conclusions." - 25:05 "if you think that it's a magic box that is spitting out the exact right answers. You are mistaken." - 28:05 "We are designing products for humans, at least not yet." - 28:10 "You can go really fast in the wrong direction." - 28:26 "And then now you're in New York city and you're like, look how fast we got here. Awesome. We're in the wrong place." - 29:17 "I think AI is not going to get rid of how corporations judge you." - 30:00 "prompt engineering is really asking great questions" - 30:20 "it's gonna be garbage in garbage out." - 30:32 "this skill set of being a really good creative problem solver is going to be top of the list for employees of the future." - 31:53 "Is our AI going to back us up in meetings?" - 33:06 "Technology has pros and cons a hammer can be used to build a house or it can be used to knock somebody in the head" - 34:00 "you are still bringing you to your AI fluency journey." - 34:33 "The goal of AI fluency is not to add additional pressure to you." - 34:56 "AI just changes how you are going to apply your expertise in your career." - 35:23 "That's why we have teams." ## Books mentioned No specific book is named with title and author. At 2:19 a host lists picking up "a learn Python or Python for dummies books" as an example of what people think AI fluency requires. That is a generic nod to the *For Dummies* Python titles, not a recommendation of a particular edition. Nothing to send to Book Summaries. ## Tools and people mentioned - AI tools: ChatGPT / GPT, Copilot, Gemini, DeepSeek, Leonardo.ai, Midjourney, Adobe Firefly; LLMs and agents in general. - Design tools: Adobe After Effects (since 3.1), Photoshop, Sketch, Figma. - People and references: Brian Sullivan (J's former UX research leader), Ted Lasso ("be curious, not judgmental"), Iron Man's Jarvis, a hypothetical AI companion named "Paul", the song Hotel California. - Ideas: AI fluency versus expertise, the light-switch analogy, domain expertise × AI fluency = leverage, the AI paradox, "AI for the sake of AI", garbage in / garbage out, the unicorn myth (next episode). ## Good work, productivity and career ideas - **Define fluency as use, not mastery.** Know what AI is good and bad at, and how to "turn the light on". You don't need to build models (5:33, 6:15). - **Choose the room first.** Pick the AI that helps your own expertise solve a specific, current problem at acceptable risk (6:47, 7:01). - **Budget for learning.** Expect about $20 a month minimum per paid tool, more for video, and tokens spent on trial and error (7:30). - **Test against what you know.** Ask whether a new tool is helpful or "just shiny" (9:42). - **Learn one instrument at a time.** Go deep on one tool through deliberate experiments (e.g. systematically varying shots, lenses and aspect ratios), then transfer to the next (11:42, 12:31, 13:41). - **Debug prompts like code.** When a prompt misfires in a new tool, break it into pieces instead of stressing (14:32). - **Expect variability.** The same prompt and model can give different results tomorrow, so plan for it (15:30). - **Multiply, don't replace, your expertise.** Domain expertise × AI fluency = leverage (17:09). - **Be the bridge to leadership.** What you learn early may be something leaders lack time for; offer it as thought leadership (19:14). - **PMs: stay in the strategy lane.** Chase tools in personal learning; at work, start from user struggles and resist AI for the sake of AI (20:10, 21:24). - **Run the PM checks.** Time saver, money saver, or lateral? Which data does it need? What's the bias liability (22:55, 23:06)? - **Wait for five or six data points** before concluding, and remember AI output carries human bias (24:16, 24:36). - **Explore before converging.** Use AI to widen the option space, then narrow to users' needs; avoid handing designers a finished "make this" (26:47, 27:42). - **Check direction before speed.** Ask what a build is for and who will use it; at work, everything needs a value proposition (28:12, 28:44). - **Invest in asking great questions.** Prompting is questioning; creative problem solving stays the top skill (30:00, 30:32). - **Lean on teams, not unicorns.** Nobody can be expert at everything; teams multiply strengths (35:02). ## Claims check | Claim | Check | |---|---| | J has used After Effects since version 3.1 "back in the 90s" (4:13) | Self-reported. Consistent with history: After Effects 3.x was the first Adobe-branded release, in the mid-1990s. | | Paid AI fluency costs at least about $20 a month (7:41) | Matches common individual tiers (e.g. ChatGPT Plus, Claude Pro, Gemini paid plans), roughly $20/month as of 2026; prices vary by tool. | | "Over 150 different kinds of bias" (24:53) | Broadly consistent with published catalogs of cognitive biases (lists commonly cite roughly 150-200). Stated loosely on air. | | Ted Lasso, "be curious, not judgmental" (30:40) | The line is from the Apple TV+ series *Ted Lasso*, where the character attributes it to Walt Whitman; the attribution to Whitman is widely regarded as unverified. | | Companies now expect AI fluency from PM and design candidates (17:58) | Stated as what the host hears day to day; not checked here. | | Brian Sullivan as J's UX research leader (24:13) | Self-reported on the episode; not independently verified. | | Episode length | RSS 36:30 (2190 s); measured mp3 36:30 (2,190.9 s). The mp3 was transcribed (file `PP_E006_AI_Fluency_Edited.mp3`). | ## Added context - Audio-only RSS episode (not on the YouTube channel). Hosts only; it follows the two-part Preston McCauley conversation ([[004 Designing at the Speed of AI - Part 1]], [[005 Designing at the Speed of AI - Part 2]]), which it refers back to as "the last episodes". - The RSS description's "AI unicorn" theme is only touched at the end (35:02) and teased for a later episode. 008 "Unicorns, T-Shapes, and the Myth of Mastery" and 010 "Unicorns, Humans, and the Future of Creative Work" appear to pick it up. - The "fast in the wrong direction" theme echoes [[003 AI Can Get You There Fast... But Where Are You Going]]. - Whisper corrections applied: Metzy/menci→Metsy, Jay Xu→J Schuh, mid-journey→Midjourney, "AI fluid"→AI fluent, "C, chat, GPT"→DeepSeek, ChatGPT (confirmed by the publisher VTT).