larsnow

CTO @MedullarSearch | CEO @luodaint | Building @feedjolt ⚡ | AI advocate | Techno-optimist | Trail runner | Family | #AI #BuildInPublic

[larsnow @larsnow]

Five months. One repo. One discipline. Feedjolt is live on Product Hunt.

Built the customer-feedback tool I wished existed at my last B2B SaaS, flat pricing, unlimited end users, Slack + Linear integrations that feel native, AI triage with a human in the loop.

If you've ever DM'd me about feedback tooling, today's the day. Your support would mean a lot.

https://t.co/k5T9jrQa6f

[larsnow @larsnow]

@carlesandres @marclou Never used Raycast, so I can’t compare. I bought a license of Alfred so many years ago (2016) and have been so happy, so I didn’t need to check others :)

[larsnow @larsnow]

One VSCode bug this week exposed your GitHub token with just one click. Meanwhile, everyone else is hurrying to give agents MORE access.

What I actually do:
– .env behind deny rules
– Scoped, short-lived tokens
– The agent never touches prod creds

Boring, but that's why I sleep soundly.

[larsnow @larsnow]

Watching a customer's users vote on their @feedjolt roadmap this week. The tool finally belongs to them, not me. That's the goal.

[Feedjolt @feedjolt]

Someone's running their public roadmap on @feedjolt right now, and their own users are showing up to vote on it.

That's the whole point. A feedback board isn't yours. It's the place where your users argue about what you build next.

Unlimited end users, flat price, so every voice counts and not just the ones in your seat plan.

[larsnow @larsnow]

A guy just rage-quit Gmail because the AI kept dumbing things down for him. That's the whole lesson. Treat your users like they're slow, and they leave. @feedjolt shows you the raw signal, not a sanitized AI summary. You're not stupid. We won't pretend you are.

https://t.co/Lz2gv2OtFi

[larsnow @larsnow]

Divendres. El codi pot esperar a dilluns. Cinc mesos seguits enviant. Avui toca tancar el portàtil i sopar amb amics. Build in public també vol dir ensenyar quan pares.

[larsnow @larsnow]

Corrent per la muntanya penso millor que davant la pantalla. Les dues millors decisions de producte d'aquesta setmana van sortir a mig camí, sense WiFi.

[larsnow @larsnow]

@rozzabuilds Hetzner using dokploy and tailscale for security.

[larsnow @larsnow]

@JoanCantimplora El que ningú li diu és l'hòstia que li fotran d'IRPF dels 160K... quedarà tremolant que voldrà marxar corrent!

[larsnow @larsnow]

A Stanford study shows AI outperforming law professors. Instead of seeing "AI replaces lawyers," I focused on this: an expert who uses the AI comes out ahead of one who doesn't. Same job, but different results based on new tech.

https://t.co/nciIBQkcDv

[larsnow @larsnow]

I don't need the model to be conscious. I need it to ship the feature and not lie about the test passing. The consciousness debate is the most expensive distraction in software right now.

https://t.co/mIr68qAPHQ

[larsnow @larsnow]

I built @feedjolt to better handle customer feedback. It's pretty ironic how I get some of the best insights when the tool bugs me. Using your own product for feedback is gold; it’s the cheapest way to do user research.

[larsnow @larsnow]

Right now, 32GB of DDR5 goes for $375. The AI component shortage hits us all, from hardware makers to app developers like me. Because of this, simple yet efficient infrastructure is regaining its edge as a key asset.

[larsnow @larsnow]

Something I keep noticing with AI: way more business people are learning to code than coders are learning business.

One side is picking up the skill that just got 10x easier. The other still treats the hard part as optional.

Guess who ends up building the right thing?

[larsnow @larsnow]

People are still confused about what makes the model work. It's not just the weights or matrices; the actual math was figured out months ago. There's no magic inside – it's in the setup outside: the tools, the safety measures, and the person hitting 'ship'.

[larsnow @larsnow]

Self-improving AI makes for a great headline, but on my end, it's just CLAUDE.md, a test suite, and me catching the model when it screws up. It's the whole loop improving, not just the model.

[larsnow @larsnow]

@marclou BTW I'm one of the small help you got to get the 20K MRR I'm a happy user of https://t.co/UuLKOY6cyt

[larsnow @larsnow]

Amazing and inspirational @marclou

Everyone should watch it; it's not just about telling AI what you want to do, but about taking care of users, generating new ideas, gathering feedback, and implementing what users will love.

Marc, you're great! And I don't know you but you seem a great person!!! Thanks 4 the video!

https://t.co/01WPFD2vOA

[larsnow @larsnow]

@ImNotAV1rus @shipordie_ @marclou @jackfriks The $0 line gets the most trust. I've been consistent for five months in one repo; honesty about progress is what protects you, not some perfect fantasy number. So keep sharing both versions.

[larsnow @larsnow]

The ceiling is the genuine piece; the actual model is cheap. Now, the big money's in taste and distribution. It's the same for coding: building is easy, but attracting viewers isn't.