Over the last 10+ years, I’ve helped take smart vehicles, consumer electronics, and AI products from 0 to $10M+ globally.
Now I’m sharing the unfiltered GTM playbook for AI & smart hardware—and building my own AI hardware MVP from scratch.
Here’s what I’ve learned about turning complex tech into global revenue 🧵👇
@thinkercar Firefly passing the ES9 in this table makes me curious about the showroom. Does the city-car buyer get a test drive built around parking and tight streets, or the same route as the ES9?
@tphuang The 1,024-card specs are useful. Is there a public comparison for serving the same DeepSeek model at the same response-time target? That’s a much easier number to buy against.
@thinkercar The partnership will feel real to an Opel buyer when they need a replacement part. Can their usual dealer order it, and who handles the warranty?
@DJiafei@RoboColosseum What still trips Astra up most often on the SO-101? That tells a team more about what to work on next than another move up the leaderboard.
@perplexity_ai A one-line plan before it runs would help. If it’s about to pull the price from the wrong tab, let me catch that before it builds the whole comparison.
@Yuchenj_UW For a meeting assistant, getting the action item right is only half the job. Put the wrong person's name next to it and the polished recap makes the mistake easier to miss.
@tryvexoai@realharismalik@Kianif98 Which part changed after someone wore the first build all day? That's the bit I'd love to see between the seven-day prototype and the production version.
@DJiafei@RoboColosseum The hotel version of this test: throw in two shirts stuck together and see if the robot can sort it out without calling someone over.
@chris_j_paxton A stronger base model might mean fewer clean demos. A slipping part or a fixture that got bumped is still worth recording, especially what the robot does next.
@OfficialLoganK@vishalsingh2972 The done notification is the one I'd use. Can it take me straight to the preview instead of making me find the build again?
@AravSrinivas Which Teammate task got noticeably better on Neon High? For a market monitor, missing one important change matters more than shaving a few seconds off the report.
@rasbt Open weights let a small team try a model on its own data before committing to a vendor. That's a big deal when the benchmark doesn't look anything like the work.
@simonw I'd run both settings ten times on the same messy task and count how many outputs I'd actually ship. One good result can hide a lot of retries.
If I were setting up a Plaud Note Pro demo in Vietnam, I'd start on Android. Record, stop, open the app, find the file. Background transfer is iPhone-only. Can a shopper get to their first transcript without the salesperson touching the phone?
#AIHardware