Stop using AI just to write emails: 31 ways people use Claude and Claude Code to take control of their tech

Paying for a premium AI plan and using it only to write emails is like buying a Lamborghini just to idle it in a parking lot. A roundup video by Nick Saraev — recently reshared in Vietnam's tech community — lists 31 cases of people around the world using Claude and Claude Code to intervene in hardware, strip out bloatware, write their own replacements for paid software, audit security, and even support cancer research and planet hunting. This article organizes the full list into five groups.

Quick answer

• The core claim: most people are using AI far below its real capability — instead of just chatting and writing, Claude and especially Claude Code can directly touch hardware, systems and real data to get real jobs done.
• The 31 cases fall into 5 groups: (1) hardware intervention and reviving old tech, (2) writing your own software and automating life, (3) security and privacy auditing, (4) medicine and science, (5) learning, creativity and business.
• Headline numbers in the list: Wi-Fi sped up 8x just by photographing a router's settings page for the AI to analyze; roughly 200 deals closed autonomously between Claude agents of 69 Anthropic employees in a negotiation experiment; 1,500 scripts scanning 7 years of NASA telescope data to find overlooked planets.
• A reliability caveat: this is a compilation of community stories retold in one video, not independently verified results — its value lies in the usage ideas, not in any specific number.

Group 1: Hardware intervention and reviving old tech

The clearest "hands on the real world" group — where AI leaves the chat box:

• Strip all ads from a Smart TV: send the TV's details to Claude Code to rewrite the interface from scratch, blocking ads and bloatware entirely, while turning the idle screen into a calendar and weather display.
• Build a new device from scratch: have Claude select components, design the circuit board, write all the firmware and draft a 3D-printed case — producing a dedicated recording microphone in a few days.
• Mod home devices: write a slick new air-conditioner control UI right on the PC taskbar; or set a Bluetooth receipt printer to print important emails every morning to read like a paper newspaper.
• Repurpose old gear: jailbreak a Kindle e-reader or a 2010 iPad into a wall-mounted smart-home panel; one person bought a $60 old BlackBerry and reverse-engineered it to run Claude Code directly on the physical keyboard.
• Make Wi-Fi 8x faster: photograph the settings page of an old router and send it to Claude; the AI found the configurations choking the network and optimized them, lifting speeds from 5–10 Mbps by 800% at zero cost.
• Turn a drone into an autonomous guard: mount an old phone on a DJI drone and program it with Claude Code to patrol the house on its own when an alarm triggers — no cloud connection needed.

Group 2: Write your own software, automate your life

• Cut the monthly software bill: indie hacker Pieter Levels used Claude to write his own replacements for most of the paid SaaS subscriptions he was using, shrinking the recurring bill to a few basics.
• Rebuild a paywalled app for free: when a baby-tracking app abruptly went paid and cut its free tier, one user had Claude write an equivalent app and released it free for everyone.
• Let AI haggle for you: Anthropic once experimented with letting the Claude agents of 69 employees trade with each other; the agents closed nearly 200 deals on their own, with strong models like Opus consistently negotiating better prices than smaller ones. Young users now apply similar setups to haggle automatically on Facebook Marketplace.
• Automate dating: connect Claude Code to a spare Android phone, feed it a text file of partner preferences and deal-breakers; the AI scans the Hinge app and returns about 25 matches a day at roughly 60% accuracy.

Group 3: Security and privacy auditing

• Catch a crypto miner alive: noticing a Mac running abnormally hot, a user had Claude write a temperature tool and read system logs; the AI found a hidden miner eating 12 CPU cores, pushing the machine past 100°C, and guided a complete removal.
• Detect eavesdropping in a car: plugging Claude Code into a Honda Civic's infotainment screen, a user reportedly discovered the voice assistant secretly recording, storing Bluetooth calls, with a manufacturer backdoor present.
• Sweep for hidden cameras when traveling: use Claude Code to scan the entire Wi-Fi network and suspicious Bluetooth devices in a hotel room or Airbnb, and check the phone for spyware.

Group 4: Breakthroughs in medicine and science

The group with the biggest stories — and the one to read most clearly as "retold compilation":

• Support cancer treatment research: Miriam, an engineer with stage-4 breast cancer, built 30 Claude agents to read medical literature and draft a biopsy plan; the plan was taken up by the Dana-Farber research institute for further development, and Anthropic sponsored $30,000 in credits to support her.
• A one-person biotech company: a man named Cody ran an entire biotechnology company for 1.5 years with only a team of Claude agents assisting — and the company was later acquired by Anthropic itself.
• New planets and a new enzyme: with 1,500 Claude-written scripts scanning 7 years of NASA telescope data, a user found previously missed planets; in another project, Anthropic used 1,000 Claude agents to scan DNA data and discovered a new gene-editing enzyme similar to CRISPR.

Group 5: Learning, creativity and business

• Localize and remake old games: automatically separate a game's voices, translate and re-dub them so a child can play; or use Claude to decompile legendary titles (Call of Duty: Modern Warfare 2, The Witcher 3, 3DS games) to mod and play them on new platforms without an emulator.
• Build a 3D room from photos: give the Opus model a room's dimensions, a floor plan and a few photos, and you get a 3D space to test furniture placement or walk through a house before building it — for about $5 in tokens.
• Accelerated learning and practice simulators: compress 40 hours of developer interview podcasts into one condensed playbook; or have Claude build a full drone assembly and flight simulator website to practice on before buying the real machine.
• Find market gaps and optimize yourself: scan data from 150,000 Steam games to find unexploited game ideas; dig through old records to reconstruct a five-generation family tree; automate artistic night photography; or analyze blood-glucose and genome data for personal health care at around $20.

What do all 31 cases have in common?

According to the roundup, the biggest common thread is that the distance from having an idea to having a working product has shrunk to nearly zero. A pattern repeats across the cases: users did not ask the AI "how do I do this" — they handed the AI the job itself: write the firmware, fix the router config, hunt the malware in the logs, replace the subscription tool outright. Any inconvenience in life or work can become a problem you solve yourself with a few commands, if you are willing to tinker.

How should you read this list?

To be clear: these 31 cases come from one video compiling community stories, summarized again in a Facebook post — they are not independently verified reports, and the technical detail behind each story varies widely. Some items (writing your own SaaS replacement, optimizing a router, scanning logs for rogue processes) sit squarely in things any technical user can try today. Others (jailbreaking devices, reverse engineering, decompiling games, automating a dating app) run into manufacturers' terms of service and demand care about legal, warranty and security risks before imitating them. The list's most practical value is a mindset shift: treat AI as a "digital mechanic" you assign real work to, not a chat box for writing emails.

Frequently asked questions

How is Claude Code different from regular Claude chat?

Claude chat is a Q&A interface: you ask, the AI answers in text. Claude Code is an agent tool that works in a real working environment (terminal, project folders, connected devices) — it reads files itself, writes and runs code, edits configurations and calls tools, actually performing the work instead of describing how. Most of the 31 cases belong to this "real work" category.

Can non-programmers use these approaches?

Partly. Analyzing screenshots of settings (a router, an app), summarizing long documents, planning and drafting content are open to anyone. But most of the impressive cases — firmware, TV interfaces, system-log forensics — need a technical foundation to verify results and recover when the AI gets it wrong.

Which uses can ordinary users apply right away?

Three practical directions: (1) have the AI analyze configuration screenshots (router, software) to find bottlenecks; (2) investigate a computer that runs strangely slow or hot by letting the AI read logs and process lists; (3) build small personal tools instead of subscribing to yet another SaaS service.

What are the risks of things like jailbreaking or decompiling games?

Three risk families: losing warranty and bricking devices (jailbreaks, firmware intervention); violating terms of service or copyright (decompilation, automating third-party apps); and security exposure from granting an agent deep system access. Weigh each case individually, and only work on your own devices and accounts.

Why is automated haggling by agents noteworthy?

Because it shows an agent can carry out an entire multi-step negotiation process on your behalf — offer, counter, close — rather than merely answering questions. The retold Anthropic experiment (agents of 69 employees trading among themselves, closing nearly 200 deals, stronger models negotiating better) hints that as agents spread, "a model's negotiating skill" could become a real economic factor.

Conclusion

This list of 31 uses — read with appropriate skepticism about verification — is still a valuable idea map: it shows AI's frontier moving from "answering questions" to "completing work", and the idea-to-product gap shrinking to the point where one individual can build their own device, replace paid software, and audit their own security. The remaining question is no longer what AI can do, but what real job you will hand it first.

Source: a roundup post by Lu Minh Khuong in the VGU Tech Alumni Facebook group (5 Nov 2026), relaying Nick Saraev's compilation video of community uses of Claude and Claude Code; edited and structured by Wotech Electronics. The cases are community stories as retold, not independently verified. Illustration: the photo attached to the original post.