The most convincing agent demo this week was not a coding benchmark. It was a television.

On October 20, @Azadux started a thread of TV hacks he had seen in the previous 48 hours. People used Claude Code and Opus 5.5 to replace the screen everyone complains about. The clips are short, the results are visual, and you can judge them in seconds.

That is a better test than another leaderboard.

What people actually built

Five projects carry the thread.

1. A TV OS with a 3D apartment instead of ads. @srepdor says he vibe-coded his own TV OS with Claude Code. He replaced the ads with a live 3D model of his apartment. The lights and ceiling fan are wired in and update in the 3D scene in real time, by his description. The TV inside the 3D model shows the live TV feed.

Read that again. A screen inside a screen, driven by an agent-written interface.

2. A Game Boy Color home screen. @robbietilton turned his TV into a Game Boy Color with a Pokémon town as the home screen. To open an app, you walk into a building. To change a setting, you beat the gym leader first. His verdict on the model: "Opus 5.5 is the goat."

3. A family TV, rebuilt. @JoshDaws used Opus 5.5 to change how his family uses the TV. The post is a 3:50 video, with no build details. That is fine. The interesting part is that the user is a family, not a developer chasing a demo.

4. A minimal version in Rive. @z00med0ut kept it small and built the interface in Rive, a design and animation tool. Not every TV hack needs a full app stack.

5. The polished final cut. @chiziaruhoma posted a 40-second version and called it final. The iteration loop is visible in the thread itself.

The replies add the useful edge cases:

  • @iBuildWithAI built a remote on a MacBook that controls an Xbox and overlays a screen-shared window with transparency.
  • @tokifyi told Claude Code to "turn on the tv. open netflix on toki and play ai is gone wild." It did.
  • @ALL_Jonah is collecting the projects into one site so people can "liberate their TVs."

Why this demo works

Three things make a TV hack a better agent demo than a dashboard.

The result is visual. You do not need to read code to judge a home screen. You see the ads disappear. You see the Pokémon town. You see the 3D room update when a light turns on.

The problem is universal. Everyone has a smart TV. Almost everyone hates the home screen, the input switcher, and the ads. A fix is easy to want.

The software touches the room. The 3D apartment is not just pixels. The lights and fan are part of the scene, according to the author. That connection between code and a physical space is what makes the clip feel different from another chat UI.

This is also the first wave of agent demos that normal people can evaluate without a benchmark. That matters for adoption.

What the clips do not show

The videos show the happy path. The hard parts are off camera.

The install path. Replacing a TV home screen depends on the device. Some Android TV boxes allow a custom launcher. Many consumer TVs do not. Locked bootloaders, signed system apps, and vendor firmware stand in the way. None of the five posts explains how the interface gets onto the TV.

The input model. A custom UI still needs to work with a remote. Focus order, back behavior, wake from sleep, HDMI switching, and volume keys are the boring parts that decide whether a family uses it twice.

The failure mode. What happens when the agent-written interface crashes at 8pm with the family watching? Can you get back to the stock screen without a factory reset? Is there a physical button that always exits?

Security. A TV app with network access is a new device on your network. If it can talk to lights and fans, it can also talk to other things. Do not connect mains devices to code you have not reviewed. Use local APIs with tokens, a separate network, and a manual switch.

Maintenance. The agent wrote it, so the agent can fix it. That only helps if you can describe the bug and run the tool on the spot. A TV that needs a laptop and a 20-minute prompt to recover is not a daily driver.

The business problem

TV makers sell ads on the home screen. Replacing that screen removes revenue. So the long-term version of this trend will run into locked platforms, not just technical limits.

That is why the current wave lives on spare TVs, dev boxes, and Android-based hardware. It is also why the demos are so appealing. They show the product the vendor will not ship.

Expect two paths. Open hardware gets custom launchers. Closed TVs get cast-like workarounds and external boxes. The agent does not change that economics.

What I would do

  1. Use a spare TV first. Not the main screen. Not the one a partner or kid depends on.
  2. Pick one job. An ad-free launcher. An input switcher. A family dashboard. One job, one week.
  3. Keep the stock remote working. Map a physical exit before you add anything clever.
  4. Write the rollback down. Factory reset steps, the stock launcher name, and the firmware version.
  5. Isolate the device. Guest network or VLAN. No mains devices on unreviewed code.
  6. Measure family acceptance. If nobody uses it after a week, the demo failed, no matter how good the clip looked.

Verdict

The TV is the next app platform because it is the last screen most people do not control. Agents make a custom interface cheap enough to try. That is the real shift.

But a 10-second clip is not a product. The winners will be the people who solve recovery, input, and trust, not the ones with the best 3D room.

Try it on the spare TV. Watch this space. Do not hack the family's main screen yet.

More in AI Tools.

Sources

#claude code tv hacks#ai tv os#ad-free smart tv#smart tv app platform#claude code projects#opus 5.5#agentic coding
Raja CRN

Raja CRN

Tech enthusiast and developer.