Vini’s List

Agents are much more useful once your code, issues, PRs and actions all live in one place.

A very good second pair of eyes on every PR. They found that code written by one model and reviewed by a different one comes out better than when the same model does both. Also, greploops are very cool!

My default for understanding what people actually do in a product. Analytics, replay and experiments, and it has an MCP as well! Also, the founders are hilarious, and so is their marketing.

Before paying for an API, check if an open model already does the job.

Drop in everything about a topic and start asking questions. One of the best self-guided exploration tools out there.

This was from my old lab at Berkeley! Really nice for building intuition on how different models stack up, rated by people instead of benchmarks.

Our designs live here. Lots of cool work being done with Figma Weave.

Our interactive agent characters are built with it.

Very, very good transcription, and very fast inside products. When specific agents narrate, we are using Whisper.

Everyone seems to be excited about this one. I still have not gotten into it fully, but I see the appeal.

Lovely interactive presentations, and quite high quality for ten minutes of work. Worth exploring generating HTML slides directly with Claude or Codex too.

My pick if you want control: run it locally, then use it through MCP from basically any model.

Very underrated if you already live in Google Workspace. Great for turning inboxes, docs and recurring admin into little agents.

Pretty neat if you have a lot of emails.

My go-to for 0 to mobile app creation.

Barrier to entry for web development has never been lower! I sometimes use Lovable, but personally prefer going straight to coding agents. Great for folks who don’t want to work in the code.

The new search? I find it much better than the other models for helping me write well-sourced speaker bios or intros.

Use this when integrations matter most. It connects to basically everything.

The easiest of these to understand. Where I would tell a beginner to start.

The best generated speech I have used. Good enough that it is in our product: we use these in our scenario features!

The answer to the "which model has taste" problem: two designs, no labels, pick the better one.

Sharper than a general model on business French and German, which matters when a native speaker is going to read it.

Swap models without rewriting anything. This is how we manage stochasticity by having fallback models.

Once an AI system matters, you need to know whether it is actually getting better.