Lost my phone at the office. Claude suggested tracking Bluetooth signal strength
An office worker who lost their phone used an AI assistant to quickly generate a Bluetooth signal-strength “hot/cold” meter, highlighting how large language models are becoming everyday problem‑solving tools. Commenters share similar stories of rapidly building small apps, games, and utilities with AI, often in minutes and sometimes for production use. Alongside the enthusiasm, many raise concerns about messy, AI‑generated code, long‑term maintainability, and the risk that throwaway experiments could become “load‑bearing” parts of real systems—even as others argue future models will simply rewrite or refactor whatever’s needed.
Everyday problem‑solving with LLMs and Bluetooth
- The thread centers on using an LLM to suggest tracking Bluetooth RSSI to locate a lost phone; the model even wrote a simple “signal meter.”
- Multiple commenters note they’ve been doing similar RSSI tricks for years: Linux
bluetoothctl/hcitoolscripts, Android apps, Home Assistant with ESP32/Shelly, UniFi Wi‑Fi RSSI, and smartwatch “find phone” features. - Some describe using RSSI to find lost earbuds or fitness trackers, treating it like a “hot/cold” game or metal detector.
- A recurring theme: LLMs feel impressive to people who don’t know the existing tools; often the model is just rediscovering well‑known techniques.
Personal software creation boom
- Several participants describe using LLMs to rapidly build one‑off tools, games, and production apps: music generators, piano learning games for kids, custom video compositors, ecommerce/landing‑page systems, and fitness apps.
- They emphasize that such breadth of solo output would have been infeasible without AI assistance.
- There’s excitement about kids growing up able to “talk” applications into existence in minutes.
Code quality, “spaghetti,” and future maintenance
- Many note that quick LLM‑generated projects often have messy state and structure.
- Some argue this is just like novice or rushed human code; others say better prompting and constraints can yield clean, pattern‑based output.
- One side holds that “code quality” (human readability, extensibility) still matters greatly; another argues that if future agents can refactor, regenerate, and maintain code, neatness mainly serves humans and may be less critical.
- Concerns include faster accumulation of tech debt, load‑bearing “toy” projects, and non‑experts integrating AI‑written code into larger systems.
- Counterpoint: future models might simply treat messy code as a rough spec and rewrite from scratch.
LLMs vs traditional search
- Some prefer asking an LLM over web search, highlighting intent understanding, aggregation of sources, and structured answers.
- Others lament how web search (especially Google) has degraded, effectively becoming “a bad AI,” and see LLMs as the natural evolution of the URL bar.
- A few still favor specialized search engines and worry about AI overviews making users confidently wrong.
Meta and philosophical tensions
- Commenters note a split between enthusiasm for sharing AI success stories and skepticism or negativity in replies.
- Broader debates touch on regulatory capture, model regressions or lock‑downs, determinism vs probabilistic outputs, and whether AI‑written code is or will be better than typical human code.