![[images/Why Everyone is Suddenly OBSESSED With Claude Code.webp]] **Creator:** Enrico Tartarotti · **Published:** 2026-04-14 · **Length:** 12:10 · [Watch on YouTube](https://www.youtube.com/watch?v=gzt52Trk9w0) > *Claude Code looks simple, right? A black-and-white terminal tool for coders, no flashy UI, no notifications, no infinite scroll. And yet people are losing sleep over it, grinding at 3AM like it's Minecraft or Civilization.* (from the video description) A standalone essay on the design and psychology of **Claude Code**, Anthropic's command-line AI coding agent. Sponsored by **Vanta**. Written and produced by Enrico Tartarotti and Bartek Malinowski. *Summary written from the auto-generated captions and sampled frames. The captions spell the sponsor as "Advanta" and the host as "Eurico"; the correct names are used below.* ## 1) Detailed outline ### [0:00] Cold open: an unlikely obsession - Claude Code is a black-and-white, text-only coding tool that runs in an "intimidating terminal," yet clips show people saying they need Claude Code skills and that it became their second hand. - Enrico says he's never seen software get this many people this obsessed this fast, and **most of them aren't developers**. - Many products obsess over crafted interactions, great design and ease of use and still fail. This one is hard to install, terminal-only, never built for retention, and it won. - He promises three psychology and design principles that explain it, and notes that **on the day he recorded, Claude Code's entire source code leaked**, so he'll also show how it works behind the scenes. ### [1:01] Reason one: it accidentally follows game design - Games like **Civilization**-style strategy titles are addictive in the "one more turn" way rather than the social-media way: you look up and it's 3 a.m. - Claude Code feels the same. People lose sleep building one last feature, and "just another prompt" feels like "just another turn." - Enrico uses **The Art of Game Design** (Jesse Schell), "basically the Bible for anyone building games," with **113** principles (the book calls them lenses; the on-screen pages read "Lens #55" and "Lens #36") for what makes a game impossible to stop playing. - **Lens 63, short feedback loops:** great games are many small loops of action and reward. In Claude Code you say what you want, watch reasoning, actions and file edits happen, get a result, then tweak and prompt again. What makes it addictive isn't how smart the model is. - **Lens 64, juiciness:** at a 2012 game design conference in Sweden, a talk took a plain game of bouncing rectangles and added color, sound, screen shake and wobbly eyes one step at a time. The rules never changed, but the juicy version felt playful. Claude models are friendlier and more conversational than ChatGPT, and Claude Code shows fun, unnecessary status text while it works. The leaked source shows there are exactly **187** of these messages. A little personality goes a long way. - **Lens 55, visible progress:** games show XP bars and revealed maps. In Claude Code the progress bar is your whole project: an empty folder fills with files, features light up, the file tree grows in real time. It's an **open-map game** with no natural stopping point, like a strategy game where a technology is two turns away and a building is ready in four. - **Lens 36, chance:** unpredictable rewards are more addictive than guaranteed ones, which is why slot machines work. Anthropic tries to reduce mistakes, but imperfection helps engagement: sometimes the result is brilliant, sometimes okay, and about 10% of the time it almost works. That near miss is the slot-machine moment that triggers "just another prompt." - **The labor illusion:** in games like **Factorio**, **Satisfactory** and **Cookie Clicker**, watching the machine work is itself the reward. Harvard researchers call this the labor illusion: seeing work done for you makes you value the result more. A consumer AI tool would hide the work behind a spinner; Claude Code shows it reading files, writing code, running tests, hitting errors and fixing them. ### [5:12] Reason two: Tesler's law and perceived simplicity - Nobody at Anthropic set out to make it addictive. Enrico shows a tweet listing AI products built to be beautiful, friendly and approachable, all "either dead or dying," while **OpenClaw** and Claude Code, which hide nothing about terminal commands or Bash and have no buttons or cute onboarding, became huge. - **Copy and paste** is "brain-dead simple," two key combos that work everywhere. Underneath, the computer manages clipboard formats, converts rich text, handles images and passes data between apps, and Apple even makes it work across devices. That's **perceived simplicity**, not simplicity. - **Tesler's law:** complexity can't be destroyed, only moved. Larry Tesler also invented copy and paste. - **The power vs. simplicity chart:** many new products are simple and well designed but not powerful, like a sleek gadget that can only book an Uber (the frames show a rabbit-like AI gadget among the "simple" examples). Tools like **After Effects** and **Blender** are powerful but hard to learn. Claude Code started as a powerful coding tool and moved itself toward simple. - **Self-fixing as simplicity:** Enrico writes his videos with **Notion** plus Claude Code and custom skills. When he asked it to sync the script from Notion, it looked for an **MCP server**, found it wasn't working, switched to the Notion API, found the docs online and made it work. A complicated job felt simple. The "near Nirvana level" is asking Claude Code to debug Claude Code. He didn't need to read docs or know what MCP is, and if he's stuck he asks it to unstick him. ### [7:51] Sponsor: Vanta - Builders who get traction get asked by big customers whether they're **SOC 2** or **ISO 27001** compliant. - **Vanta** automates security processes and compliance across more than 35 security and privacy frameworks, including SOC 2, ISO 27001 and HIPAA. Offer: $1,000 off at vanta.com/enrico. ### [8:53] Reason three: most tasks are secretly code - Most Claude Code users aren't developers. They use it to book airline tickets, file taxes, monitor tomato plants and comb through music archives, choosing a coding tool over purpose-built apps. - Moving a folder is a terminal command. Renaming 500 photos and organizing folders is a script Claude can write. Analyzing a spreadsheet doesn't need **Excel's** UI; Claude reads the data directly and writes **Python** you never have to read. - Most daily tasks are some form of code with an interface wrapped around them for humans. - On **Skyscanner** you can't ask for flights with a one-day layover in Singapore that then connect to Taiwan and maybe on to the US, because the UI is fixed. Claude Code can figure out which APIs to call, gather the travel data and present a solution, all through code. - It isn't teaching people to code; it removed the barrier hiding the code that was already there. ### [10:17] What the source-code leak shows - **The agent loop:** it reads your project and files, works out what's going on, picks a plan and acts, choosing among **more than 30 actions** (read files, write code, run commands, search the web). When something breaks it reads the error, figures out what went wrong and tries another approach, in a loop that runs until the job is done. - **Very complex:** around **five different strategies** for compacting the conversation to retain context, and **fake tools** added so competitors have a harder time distilling how it works. - **Also quite dumb:** prompts that literally beg the model to please follow instructions, and logging of when users curse at it, used as a benchmark for how well it's working. ### [10:57] It worked too well - The top plan is **$200 a month**, which Enrico pays and would once have called insane. He routinely runs three or four instances in parallel. - One user burned about **$27,000 worth of compute** on a single $200 subscription, so Anthropic is losing a lot of money daily. The product works so well that it works too well. ### [11:27] Outro: tech running at two speeds - AI chats are replacing jobs and companies spend billions on products nobody uses, yet **AirDrop** still doesn't work and printers are horrible. Enrico says our tech runs at two speeds because of "one of the weirdest laws" deciding how tech gets built, and points to another of his videos on it (the end-screen video isn't named in the captions). ## 2) Things mentioned ### AI tools and products | Item | Role in the video | |---|---| | **Claude Code** (Anthropic) | The subject: a terminal-based agentic coding tool used far beyond coding. | | **Claude** (Anthropic) | Its models are friendlier and more conversational in tone than ChatGPT's. | | **ChatGPT** (OpenAI) | Comparison point for tone and for generic chat AI. | | **OpenClaw** | Another unpolished, terminal-oriented AI tool that succeeded where friendly AI products didn't. | | **Claude Code skills** | Custom skills Enrico built for his scriptwriting workflow. | | **MCP (Model Context Protocol) server** | What Claude Code first tried for Notion before falling back to the API. | | **Claude Code $200/month plan** | The highest tier, which Enrico pays; one user ran about $27,000 of compute on it. | | **Claude Code source-code leak** | Revealed 187 status messages, 30-plus actions, five context-compaction strategies, decoy tools, pleading prompts and frustration logging. | | **Single-purpose AI gadgets** | "Sleek" hardware that can only book an Uber; examples of simple but not powerful. | ### Games and game design | Item | Role in the video | |---|---| | **The Art of Game Design** (Jesse Schell) | Source of the lenses: 63 short feedback loops, 64 juiciness, 55 visible progress, 36 chance. | | **"Juice it or lose it" talk (Sweden, 2012)** | The bouncing-rectangles demo of how polish and feedback make the same game feel fun. | | **Strategy games (Civilization-style)** | The "one more turn" kind of addiction. | | **Minecraft and Civilization** | Named in the description as the comparison for 3 a.m. grinding. | | **Factorio, Satisfactory, Cookie Clicker** | Games where watching the machine work is the reward. | | **Slot machines** | The model for variable, unpredictable rewards. | ### Design and psychology concepts - **Short feedback loops** of action and reward. - **Juiciness**: small bits of personality and feedback that don't change the rules. - **Visible progress** and **open-map** design with no natural stopping point. - **Variable rewards** and the near-miss "almost works" moment. - **Labor illusion** (Harvard research): seeing work being done raises perceived value. - **Tesler's law** (conservation of complexity): complexity can only be moved, not removed. - **Perceived simplicity** vs. real simplicity, illustrated by copy and paste. - **Power vs. simplicity chart** for placing products. ### Software, apps and services | Item | Role in the video | |---|---| | **Terminal and Bash** | Claude Code's environment; things it doesn't hide from users. | | **Copy and paste / clipboard** | Example of hidden complexity, including Apple's cross-device clipboard. | | **Notion** | Where Enrico organizes scripts; Claude Code synced from it via its API. | | **Adobe After Effects, Blender** | Powerful but hard-to-learn tools. | | **Microsoft Excel** | A UI Claude doesn't need to analyze a spreadsheet. | | **Python** | Code Claude writes behind the scenes for data tasks. | | **Skyscanner** | Flight search whose fixed UI can't handle complex multi-leg requests. | | **Uber** | The one thing the sleek gadget can do. | | **AirDrop** (Apple) and printers | Everyday tech that still doesn't work well. | | **Flask** (flask.do) | Enrico's video collaboration tool, plugged in the description. | ### Companies and compliance | Item | Role in the video | |---|---| | **Anthropic** | Maker of Claude and Claude Code. | | **Apple** | Cited for making copy and paste work across devices. | | **Vanta** (vanta.com/enrico) | Sponsor; automates security and compliance, $1,000 off. | | **SOC 2, ISO 27001, HIPAA** | Compliance frameworks big customers ask about; Vanta covers 35-plus frameworks. | ## 3) Biographies ### Enrico Tartarotti (host, writer, producer) Italian-born tech creator and designer behind the YouTube channel **Enrico Tartarotti**, where he goes behind the scenes of the design, psychology and stories of everyday tech. He's building **Flask** (flask.do), a video collaboration tool, offers one-hour calls through enricotartarotti.com, and handles sponsorships through **Rakugo Media**. In this video he describes writing his scripts with Notion and Claude Code with custom skills, paying for the $200-a-month plan and running three or four instances at once. He's on X as @EnriTarta and on Instagram as @enritarta. ### Crew - **Bartek Malinowski**: co-writer and co-producer of this video. - **Seequence**: the editing studio credited for the video. ### People referenced - **Jesse Schell**: American game designer, CEO of Schell Games and professor at Carnegie Mellon's Entertainment Technology Center, author of *The Art of Game Design: A Book of Lenses*, the book whose lenses Enrico uses. The video refers to the book rather than naming him. - **Larry Tesler** (1945–2020): American computer scientist who worked at Xerox PARC, Apple, Amazon and Yahoo. He's credited with cut, copy and paste and modeless editing, and with the law of conservation of complexity that bears his name. - **Martin Jonasson and Petri Purho**: indie game developers who gave the 2012 "Juice it or lose it" talk at a game conference in Sweden, the bouncing-rectangles demo shown in the video. The video doesn't name them. - **Harvard researchers on the labor illusion**: the concept comes from work by Ryan Buell and Michael Norton at Harvard Business School showing people value services more when they can see the work being done. The video credits "researchers at Harvard" without naming them. ### Organizations referenced - **Anthropic**: AI safety and research company that develops the Claude models and Claude Code. - **OpenAI**: maker of ChatGPT, used as the comparison for tone. - **Vanta**: security and compliance automation company; sponsor of this episode.