![[images/The Weird Future Of User Interfaces.webp]] **Creator:** Enrico Tartarotti · **Published:** 2026-06-23 · **Length:** 15:08 · [Watch on YouTube](https://www.youtube.com/watch?v=f32W5BEzWN0) > *Every tech company on Earth is trying to kill the UI you grew up with, and almost all of them have failed, badly.* (from the video description) ## 1) Detailed outline ### [0:00] Cold open: the Starbucks race - Side by side, the regular **Starbucks app** and the **Starbucks plugin for ChatGPT**, the "future" where agents do things for you. The old app finishes the order while "the future of computing" is still working on the drink (both clocks shown on screen). - "The most advanced technology ever built just lost the race to a few buttons on an app that's been around for 15 years." - Everywhere, buttons, menus and screens are being replaced by a gray box asking "What do you want to do?" Enrico says that isn't really the future. After a month down the rabbit hole and talking to people building at the cutting edge, he sees something else forming: neither the 40-year-old GUI nor chatbots taking over every app. - To get there, first understand why every attempt to kill user interfaces has failed. ### [0:58] Part 1: Every attempt to kill the UI has flopped - **Devices:** Microsoft just announced a new AI device he calls bad-looking. The **Humane Ai Pin** raised **$230 million** to replace screens and became one of the biggest tech fails ever; the **Rabbit R1** and other gadgets went the same way. - **Voice assistants:** Siri, Alexa and Google Home were pitched as the future of computing. His Google Home is now "a glorified light switch" for smart lights and a pasta timer. - **Forced AI buttons:** a **Meta AI** button now sits in the most prominent spot in WhatsApp, and Microsoft requires every new Windows PC to ship with a **Copilot key**. But great products spread organically: "No one had to put a Gmail key in your keyboard." - Science fiction taught us that asking tech to do things is the future, and these companies had billions and the best teams. The problem was simpler. Replaying the Rabbit demo ("Can you ride from my office to home now?"), they pushed a new interaction model onto tasks the GUI already does brilliantly: booking an Uber, ordering coffee, sending a message. - **Experiment:** close your eyes and unlock your phone. You probably know where every home-screen app is without looking. That muscle memory is what these devices were competing with. ### [2:40] Part 2: Meanwhile, people are abandoning GUIs by choice - The other end of the spectrum is "the biggest gold rush in the last 20 years of tech that no one had predicted." Coding agents like **Claude Code** and **Codex**, meant for software engineers, are spreading among regular people to edit videos, build automations, organize files and do taxes, with tutorials everywhere and no big AI button required. - Enrico's own setup: - He writes the video in **VS Code** using Claude Code in a terminal as a writing assistant, with a system of **skills** for research and searching X. - He did his **taxes** in a coding tool connected to his **Gmail** accounts: it pulled receipts and photos from Google, matched everything and saved it to **Notion** for his accountant, "without clicking a single button." - He now gets annoyed when a tool doesn't connect to his coding agent, or when an agent that ran autonomously for an hour stops to ask him to log in (five seconds of effort). - **Products are adapting:** **Linear** redesigned around this shift; **PostHog** (the analytics tool he uses for Flask) did too; **Salesforce** released a **headless** version with no UI because it's meant for agents; **Notion** and **Google** released **CLIs**, terminal commands meant for agents to create calendar events, read email or pull a block from a Notion page. Companies that spent years crafting UIs are effectively saying "you don't want to use that anymore." - So devices trying to kill the UI fail, while "techie" tools (terminals, CLIs, **MCPs**) see organic adoption for boring, repetitive, time-consuming work. - He made an on-screen animation entirely with Claude Code, specialized skills and **Remotion**, which would have been painful by hand in **After Effects**. He loves **Blender**, but hunting for one feature in its sea of menus "makes me sick"; he wants his agent to do it. - Tease: so is it all agents and no UI? No. First you need the one idea that made GUIs work, which is also AI's biggest problem: **direct manipulation**. ### [5:38] Sponsor: Lovable - For people who want to build a real product without touching terminals. Enrico builds a long-held idea from his middle-school list: a **stock market for your time**, where people bid on one future hour of someone's time (book your friend before they become the next Elon Musk) and trade it. - He dictates the concept (payments, stock-market feel, time-slot picker). After a few iterations it's a real app with user accounts, a database and payments with real money; **Lovable** can also buy a domain and host it. - Stats quoted: **40 million projects** built, **200,000 new ones** launching every day. ### [6:46] Part 3: Direct manipulation, the idea that makes GUIs work - Pinch to zoom a photo: you didn't type "zoom to 200%." You grabbed the photo and manipulated it, or dragged a file into the trash. That's **direct manipulation**: elements (album covers, list items, icons) that you tap, scroll, drag, zoom and pinch. - Its superpower: everything you can do is visible in front of you. You don't memorize commands; you open a menu and see them all. Perfect as long as the things fit on a screen, like the handful of options in the Starbucks app. - Add complexity and it breaks down: menus, then submenus, then panels, then customizable panels chosen from yet another menu. That's how we got today's daunting professional tools. - Contrast: the **Instagram Stories** editor feels simple (move, draw, tap to edit) because it sits within what direct manipulation handles well. **Adobe After Effects** feels intimidating because it's past that limit. - The pattern: AI attempts failed on simple tasks because GUIs are simply better there. Past a complexity threshold, GUIs turn from an asset into a liability (menus, dashboards, panels you have to learn), and you'd rather a robot do it. Chat is free-form, not limited to buttons a designer built: **indirect manipulation**. - The tempting conclusion: simple things use UI, complex things go agentic via chat. ### [9:16] Part 4: The middle of the bell curve and generative UI - What he left out: it's not a straight gradient but a **normal distribution**. Most daily tasks aren't setting a timer or analyzing huge databases. They're in the middle: browsing the web, writing a presentation, booking a flight, checking email. That middle is the missing link. - **Examples:** a normal-looking browser where the "slow loading" page was actually **generated on the spot** from what you typed (a plant-care page, "Watering Your Cheese Plant," shown on screen). Asking **Claude** to explain relativity produces an **interactive visualization** (a spacetime-curvature grid) instead of a wall of text. - The difference from AI app builders like Lovable: nobody built these apps. A machine decided a GUI was the best interface for what the user was doing and built it on the spot. This is **generative UI**, and it has **three levels**. - **Level 1:** whole interactive views generated in the moment, styled to the host's brand (Claude's colors). **Google** announced Search will generate little apps to visualize complex topics. They're slow now, but like web pages that once took a minute to load, generation will come to feel instant. - **Level 2:** regular UI with adaptive generative pieces. Flight booking's date pickers are faster than typing and more accurate than voice. But planning a trip for himself and producer **Bartek** to San Francisco (to shoot the Figma behind-the-scenes video), from Italy and Poland with a New York stop and uncertain dates, moved the task toward the complex end. **Google Flights**' AI mode accepted the prompt but couldn't show results, because the output underneath was still a fixed grid of flights. - His **mock-up** keeps the familiar Google Flights UI but lets you add any odd request, and returns UI no Google designer drew directly: a date **slider** with color-coded prices (cheapest to most expensive; shown as a €569 Poznań to SFO option on screen). - **Brian Chesky** said on X that Airbnb tried chat-based booking and scrapped it because it wasn't the right interface, and has launched a new AI lab building models purpose-made for generative UI. - At level 2 the craft changes: designers and engineers stop placing buttons on screens and instead design **rules**: the UI kit ("LEGO blocks") in your visual language and taste, plus rules for using them (when to show a map vs. a list, preferred layouts). - **Level 3:** the entire product is generative UI. "The rules that tell the product what to show are the product." ### [13:28] Conclusion: all three at once - The future isn't generative UI alone, all UI, or all chatbots. It's traditional direct-manipulation UI where that fits (usually simple jobs), powerful agents for complex jobs, and all three levels of generative UI where they solve the problem best. That's the biggest shift most products face in the coming years. - He's applying it to **Flask**: regular UI for recording on-screen drawings and feedback on videos (familiar and visual); generative UI for specific breakdowns of lots of people's feedback, once the tech is good enough; chat and agents for chores like organizing folders and managing team seats. - Callback to history: just as **Xerox PARC** invented the GUI that shaped computing 40 years ago (a 1984 **MacWrite** screen appears on screen, echoing the description's "design principle hiding in plain sight since 1984"), someone in a café right now is doing it again. - Outro tease: behind every product is a set of decisions that are neither design nor engineering, "the least talked about and most important part" of your tech, covered in the next linked video. ## 2) Things mentioned ### AI devices and assistants (the failures) | Item | Role in the video | |---|---| | **ChatGPT Starbucks plugin** | Agent ordering that loses a race to the Starbucks app. | | **Starbucks app** | The "few buttons" that won. | | **Humane Ai Pin** | Raised $230M to replace screens; one of tech's biggest flops. | | **Rabbit R1** | Another failed AI gadget; its "book a ride" demo is replayed. | | **Microsoft AI device** | Newly announced, unnamed in the captions; Enrico calls it bad-looking. | | **Siri, Alexa, Google Home** | Once "the future of computing," now light switches and timers. | | **Meta AI in WhatsApp** | A prominently placed AI button nobody asked for. | | **Microsoft Copilot key** | Mandated on Windows keyboards; contrasted with "no Gmail key." | | **Uber** | The kind of simple, habitual task GUIs already nail. | ### Agent and developer tools people are adopting | Item | Role in the video | |---|---| | **Claude Code** | Terminal coding agent Enrico uses to write videos, do taxes and make animations. | | **OpenAI Codex** | Another coding agent spreading to non-engineers. | | **VS Code** | Where he writes scripts with Claude Code in a terminal. | | **Skills, CLIs, MCPs** | How agents get abilities and connect to tools. | | **Gmail, Google Photos, Notion** | Sources and destination in his agent-driven tax workflow. | | **Remotion** | Code-based video package used for an on-screen animation. | | **Adobe After Effects** | The painful manual way to animate; also the example of an intimidating pro UI. | | **Blender** | Powerful free 3D tool whose menus he'd rather an agent navigate. | ### Products adapting to agents | Item | Role in the video | |---|---| | **Linear** | Issue tracker redesigned around the agent shift; on-screen Claude prompt asks to show Linear tickets and hits an expired login. | | **PostHog** | Analytics for Flask; redesigned. | | **Salesforce (headless)** | Released a version with no UI, meant for agents. | | **Notion CLI, Google CLI** | Terminal commands for agents to create events, read email, pull Notion blocks. | ### Generative UI examples | Item | Role in the video | |---|---| | **Generated website** | Browser page built on the spot from a search ("Watering Your Cheese Plant"). | | **Claude interactive visualizations** | Relativity explained with an interactive spacetime model (level 1). | | **Google Search generative UI** | Announced mini-apps for complex queries; called the biggest Search change in 25 years in the description. | | **Google Flights (AI mode)** | Prompt box over a fixed grid, so it couldn't answer his multi-city query. | | **Enrico's Google Flights mock-up** | Level 2 example: a color-coded date slider. | | **Airbnb** | Chat-based booking tried and scrapped; new AI lab for generative-UI models. | ### Sponsor and Enrico's projects | Item | Role in the video | |---|---| | **Lovable** (lovable.dev) | Sponsor; AI app builder used to make a "stock market for your time" app with accounts, database, payments, domain and hosting. 40M projects, 200K new daily. | | **Flask** (flask.do) | Enrico's video collaboration tool, being redesigned to mix regular UI, generative UI and agents. | | **Figma behind-the-scenes video** | The reason for the San Francisco trip; linked in the description. | ### Design concepts and history - **Direct manipulation** vs. **indirect manipulation** (asking an agent). - The **complexity threshold**: menus, submenus, panels, customizable panels. - **Instagram Stories editor** (simple) vs. **After Effects** (complex). - The **normal distribution** of everyday tasks; the important middle. - **Generative UI**, levels 1 to 3; designing **UI kits and rules** instead of screens. - **Xerox PARC** and the origin of the GUI; the **1984 Macintosh / MacWrite** era shown on screen. - **Bloomberg Terminal** and **Felix Haas** appear only in the description's keyword list, not in the captions. ## 3) Biographies ### Enrico Tartarotti (host, writer) Italian-born tech creator and designer behind the YouTube channel **Enrico Tartarotti**, where he goes "behind the scenes of the tech products you use every day" through designed video essays on UX and product design. He builds **Flask** (flask.do), a video collaboration and review tool, and takes sponsorships through **Rakugo Media**. Credits for this video: written by Enrico Tartarotti, produced by Enrico Tartarotti and **Bartek Malinowski**, edited by **Seequence**. He's on X as @EnriTarta and on Instagram as @enritarta. ### People referenced - **Brian Chesky**: co-founder and CEO of Airbnb. In the video he's cited for saying on X that Airbnb scrapped chat-based booking and for launching an AI lab focused on generative-UI models. - **Elon Musk**: CEO of Tesla and SpaceX, used as the "bet on your friend becoming the next..." example in the Lovable segment. - **Bartek Malinowski**: Enrico's producer, who traveled with him from Poland to San Francisco for the Figma shoot. - **Felix Haas**: designer named in the description's keywords; the captions don't say who he is or what he contributed, so his role here isn't confirmed. - **Seequence**: the editing studio credited for the video. ### Organizations referenced - **Xerox PARC**: Xerox's Palo Alto Research Center, where researchers developed the graphical user interface, the mouse-driven desktop and related ideas in the 1970s, which later shaped the Apple Macintosh (1984) and Microsoft Windows. - **Humane**: startup founded by former Apple employees Imran Chaudhri and Bethany Bongiorno; its Ai Pin launched in 2024 to poor reviews, and the company's assets were sold to HP in early 2025. - **Rabbit**: startup behind the Rabbit R1 handheld AI device (2024), criticized for doing little that a phone app couldn't.