I write a newsletter about startups and investing—for ai builders of all levels.

I record mini-tutorials, review tools I’m testing, share my insights from an exited founder turned investor.

Hey folks,

Want to sponsor this newsletter? click here^75†.

Yesterday I went to Cerebral Valley AI Summit—I was graciously invited by Eric (Founder of Newcomer) and Riley - ty both!

These kinds of events are much better than big conferences where everyone comes with their little groups and stick together. This was 100-200 people in a more intimate setting - I even joined in (ahem, spoke the most) at a discussion group about AI workflows vs agents.

Throw in some great speakers (Granola, Hugging Face, Fireworks, Figma, etc), free food, drink and battery packs (seriously!) and you’ve got yourself an event I hope to re-tend (is that a word?).

Plus it was great to see new and familiar faces 🤗.

So whats on the block today?

Airtable³⁰⁷ just pivoted their entire company to building apps with AI. I got access this morning to Figma make also - its fine.

So why is every company now an ‘app-generator’?

My hunch is that vertical saas companies think this helps activation rate (ie time-to-thing working) and expanding use-cases (you can use us for 3000+ diff things) but it may make it worse on both.

Now I have to do the work of creating the system AND then improving tweaking it? The chat form factor may make it feel easier but really the user is still creating software in some form.

But people don’t want to spend time creating software (except us nerds), they just want a solution for their task.

And yes, I’m trying to build my own version - but in a v diff way (and for fun).

Claude Artifacts are getting an upgrade¹²³. You can now have AI within your artifact e.g. make me an AI-powered recipe generator. I wish they’d spend time on the whole app experience though - its bad. It feels like OpenAI have the consumer AI app, Claude is the best at coding models and surrounding tools (e.g. MCP, Artifacts) - my feeling is Anthropic should stick to the developer use-case.

Google also want to take over your terminal now. Gemini CLI⁹⁷ is their version of a coding tool that lives in your terminal. It’s open-source and free with very generous limits (1k requests per day). But I’m noticing that it eats up 2-3x tokens compared to other tools.

Then we have Warp 2.0²⁷¹ - Warp made me not scared of the terminal, and now it has more features like code editing, a knowledge base and agent management. It’s AI-first so you can ask questions like you would an LLM.

opencode by sst⁹⁹ - Another CLI tool that allows you to use your Claude pro/max account along with other models, including local ones.

AssemblyAI just made a huge leap in real-time speech-to-text, built for voice agents. Their new Universal-Streaming⁸⁴ model is highly accurate, lightning-fast with ~300ms latency, and starts at just $0.15/hour. No contracts, no commitment, unlimited concurrency. Test it live in the playground¹⁰⁵.*

*sponsored

want to partner with us? Click here³⁰

⚖️ Trip to the courtroom

There’s a bit of chatter from the legal world that’s good to know:

• A federal judge claimed that Anthropic’s use of books⁶⁵ to train Claude is "fair”, but downloading them from pirated sources is not.

• Meta got a similar relief⁵⁰, but this time the judge stressed that his ruling is more a reflection of the author’s failure to fight their case.

• And if we still care about Stability AI⁵³, Getty Images also dropped copyright claims against them (other claims in the lawsuit still stand).

• And outside the court, Sam is sharing his emails¹³⁷ with iyO’s founder (who sued OpenAI⁴³) publicly, and putting NYT journalists on the spot¹⁴² about the NYT vs OpenAI lawsuit.

🌐 What I’m consuming

• Building AI agents that actually automate knowledge work⁸⁷⁷.

• Combining the power of Cloudflare and OpenAI’s Agents SDKs.¹⁵⁵

• Build a personalized AI assistant³⁷² with Postgres.

• Hacking OpenAI transcription costs¹⁵¹ by speeding up the audio.

• What hundreds of engineers building in AI⁴⁷⁵ are using, building and reading (take a guess)

⚙️ Tools I’m looking into

• Omi Glass³⁴¹ - Record your entire day with smart glasses and 6x the battery of Meta Raybans.

• Pally²⁵⁷ - Keep track of your connections and search across your network.

• VibeCode²⁴¹ - Make shareable mobile apps from your mobile. (examples²⁴¹)

• Mintlify¹⁵⁵ - Add an AI assistant right into your docs with full control.

• Google Colab²³⁴ now has AI fused in and available to everyone.

• Magic Animator³³⁸ - Animate your designs in seconds with AI

• vibe-kanban²²⁶ - Project management for developers. Integrates with git repos, Claude Code and Amp.

⁵⁸

📜 A note on Artifacts:

Artifacts are browser-only, single-use apps without any “backend” that you can create in Claude. The launch of Artifacts was the key moment driving the “text-to-app” frenzy we are in. Gemini and ChatGPT followed suit with their feature called “canvas”.

Although you use AI to build an Artifact, the no-backend nature of artifacts means you can't add AI features in them (at least not easily). And you need to think about who pays the bill for that AI usage.

Anthropic modified Artifacts so that the AI features are powered by the user’s Claude account—any AI request in an Artifact is treated like another prompt on the claude app. That solves both the problems—technical implementation and the billing.

ps: Anthropic also made a special place for all your Artifacts plus a place showcasing examples at claude.ai/artifacts²⁰⁰.

- Keshav

🍦 Afters

• WisprFlow¹⁰² raised $30M, OpenRouter⁶⁸ raised $40M and Delphi⁷³ raised $16M.

• Imagen 4⁶⁸ and Imagen 4 Ultra are now available in the Gemini API.

• How would a claude code for content look? maybe something like this¹⁷¹. And how about letting Claude enter a group chat⁸³?

That’s it for today. Feel free to hit reply and share your thoughts. 👋

Enjoy this newsletter? Please forward to a friend.

• Find me on X³⁸, Linkedin³⁵, or Instagram³³

• Read about me³⁵ and ben’s bites