Aakash Thumaty

I (unexpectedly) won the @a16z x @elevenlabsio x @OpenAI consumer AI hackathon a couple months ago, here’s what I learned: Build more games, it’s super fun. DM me if you want any of the code or a tutorial I wrote for the @OpenAI realtime API.

It’s super frustrating when a company argues about timelines like 4 months or 6 months for shipping some change (like a front end redesign). Obviously they are not aware of or utilizing LLM powered IDEs. If they were, timelines on that scale are just unfathomable. Witnessing…

Liquid glass is analogous to bringing typography to the Macintosh, it’s beautiful and original and it will take so long for any other device/design language to catch up or implement something similarly exciting

“he built Clay ($8B co) with a code block that can fit in a tweet!” Was my gut reaction I didn’t (and probably still don’t fully) appreciate how much of creating something immensely valuable is packaging tons of small cool things like this into a product that’s 1. just…

The amnesia in LLMs is not just a technical gap, it’s amplified by the (currently popular) architecture of app-centric AI. Most software tools now claim to be AI-enhanced, but the approach is beet narrow: a language model paired with a vector DB, confined to a walled garden……

Way too riled up

Claude got a little riled up 😤 (We had a banger of a Memorial Day weekend)

It’s really exciting to build for cool people. It’s even more exciting to build something that makes your users cool. As in they are cooler for having used your product. This can either be: 1. the output of using your product is shareable/promotable/accoladable 2. the act of…

If you want to build something legendary it has to be iterative and compounding - can’t be one shot use cases. User has to get sucked in. Side effect is the larger the free tier the better because the more sucked in a user gets and the more they WANT to update to paid to keep…

The execution difficulty delta between services businesses AI companies (eg Harvey) and replace-a-function AI companies (eg Decagon) is so much wider than people realize

Applovin growing ad revenue at 80% year over year is the mid stage AI developer effect of “more builders but same attention/user pool” hence -> paid ads. Just imagine what would’ve happened if tiktok actually got banned, $app would’ve made $pltr look cheap

“Cracked edge” is a poor phrasing choice, this isn’t some metaphor involving eggs. “Cracked” in GenZ means “GOATed” in genAlpha both of which mean “super strong advantage” in Millenial.

I think if @BillAckman lived in SF and had to regularly rely on ride share (not just to give feedback to @dkhos), that $uber slide would’ve looked really different (also that $googl slide 👀) (But he’s rich and I’m not so don’t listen to me.)

Best product isn’t a moat. Having the best product isn’t a perennial quality, it’s a statement about a point in time. It can change quickly. I wonder what the right way to think about moats are. Was talking to a really smart person who pointed out that a company reaching 100M…

I learned about a few companies with cracked edges (as in advantages) recently. I have until now misunderstood what a real edge is. A real edge makes you never want to even mildly overlap with someone let alone directly compete with them. If you think you have an edge, I think…