A goldfish never forgets

gm legends, happy Tuesday.

Vercel Day is live. Over 500 products tagged, Vercel Ventures pitch on the table, and somewhere in that list: a Mac that already knows your context before you explain, a tool that simulates your brain on fMRI, and 25,000 cheap meals across 78 Japanese cities.

Goldfish is a Mac AI assistant built around one frustration: you already explained this project yesterday. Joel Edholm calls the approach contextmaxxing — his tool tracks everything you've been working on across every app, so pressing Option in any text field gives you an AI that already has the context loaded. No pasting docs, no catching it up.

🔥 Our Take: You've probably re-explained the same project to Claude more than once this week. Goldfish front-runs that by tracking what you're doing all day and having something ready when you press Option. Whether having your Mac know more about your workday than your manager does feels useful or feels like something you need to think about, that probably depends on what you've been working on.

1000Yen Eats maps 25,000 restaurants across 78 Japanese cities where a real meal costs under ¥1,000, roughly $7. No sign-up, no ads, no tracking. City guides work offline. Krzysztof Hendzel built it for people who need to eat in Japan on a real budget, not people who need to rate-my-experience after.

🔥 Our Take: The constraint is the whole product. ¥1,000 is not just "affordable", it's the specific number below which eating well in Japan goes from straightforward to something you need local knowledge for. Hendzel found 25,000 places that clear that bar and put them on a map. If you've tried to eat cheaply in Tokyo without knowing where to look, you know exactly what this solves.

MindReader v1 takes your content like your pitch deck, ad copy, sales email, and simulates how a brain responds to it, region by region, using neuroscience models built on 35 years of fMRI research. Ishita Srivastava and Jas at Cassini Research built it for sales evals and content testing, and are inviting academics to stress-test the results.

🔥 Our Take: Simulated fMRI is an actual research method, it models how specific brain regions respond to a stimulus without a scanner. MindReader applies that to content: paste in your pitch, get neural response metrics back. The team is openly asking academics to check their work, which is either evidence the method holds up or a fast way to find out if it doesn't. Either way, it's the most unusual answer to "will this land" you're likely to see today.

Max Musing (@maxmusing) opened with a question a lot of founders are sitting with: do you ship the rough version to start learning, or hold it until it's something you'd actually show people?

The thread split cleanly, but not the way you'd expect. The "ship now" camp didn't argue that quality doesn't matter — they argued the wrong thing kills products. Varun Mishra landed it: a beautiful product solving the wrong problem is harder to recover from than an ugly one solving the right one.

Kristofer Lachance drew the line that stopped the argument: "Ugly is survivable. Broken is what actually kills momentum."