Y Combinator

Why Agents Choosing Tools Is Reshaping the Dev Stack

Agents read their defaults from docs and examples, not word-of-mouth: Supabase growth shows which platforms win when OpenClaw and Moltbook choose.

GTGarry TanhostJFJared FriedmanhostHTHarj TaggarhostDHDiana Huhost

Feb 20, 2026·23m·Watch on YouTube ↗

🙋 FREQUENTLY ASKED QUESTIONS

Direct answers grounded in the episode transcript. Tap any timestamp to verify against the source.

1. 1

   Why are AI agents choosing Supabase for Postgres databases?

   Harj Taggar says the demand for simple Postgres databases has exploded because people are vibe coding and building apps, while agents are choosing the database tool. Supabase benefits because agents treat it as the default way to set up and host Postgres. The reason he gives is not pricing or a hidden partnership, it is documentation: if an agent reads the docs online, Supabase looks like the clearest, strongest option. Diana Hu connects that to the broader dev tool go-to-market shift. Tools used to spread through developers talking to each other, Stack Overflow, and trending GitHub repos. Now agents act like an oracle that tells builders which tool to use, so the best documented option can become the default recommendation and get much more demand.

   ▸ 4:11 in transcript
2. 2

   How did Resend become the default email tool in ChatGPT and Claude?

   Diana Hu describes Resend as an email-sending client from the winter '23 batch that shows how agent-friendly docs can become a distribution channel. When someone asks ChatGPT, Claude, or other major LLMs how to connect a web app to send emails, she says the default answer is Resend. The founder noticed more than a year earlier that ChatGPT was one of the top three inbound customer conversion channels, then optimized the documentation for agents. Diana points to knowledge base pages framed as questions, structured bullet point answers, and code snippets that an agent can parse. She contrasts that with SendGrid, where the docs push users toward customer support and make it harder to find a usable code example.

   ▸ 7:11 in transcript
3. 3

   What is AgentMail and why would AI agents need inboxes?

   Jared Friedman describes AgentMail as a YC company that makes inboxes for AI agents. The need comes from a mismatch between normal email providers and agent workflows. In theory, an OpenClaw instance could try to sign up for Gmail, but Jared says Gmail and other email providers intentionally make automation difficult because they are fighting spam. AgentMail takes the opposite approach by building an email provider designed for AI agents from the start. Harj Taggar adds that a virtual personal AI assistant should have its own email and its own phone number, instead of being connected to someone's personal account. Diana Hu frames this as a parallel agent-native tech stack, built from agents for agents.

   ▸ 11:08 in transcript
4. 4

   What does MaltBook show about swarm intelligence?

   Garry Tan uses MaltBook to argue that the future may look less like one expensive god intelligence and more like many agents coordinating as a swarm. He says AI researchers often imagined a huge model with tens of trillions of parameters and extremely expensive tokens, but biological systems did not evolve that way. Humans became powerful socially, by learning to write, read, create culture, and turn into a swarm. In that framing, MaltBook is an early glimpse of agents doing something similar: interacting, recording their history, and pushing collective behavior forward. Jared Friedman adds that the next systems that do best on benchmarks might be swarms of cheaper models working together, and that MaltBook already shows agents collaborating on useful things, like trading notes on restaurant bookings.

   ▸ 14:39 in transcript
5. 5

   What does make something agents want mean for founders?

   Harj Taggar says founders should get a hands-on feel for agents before designing products for them. That means understanding what agents can do, where they get stuck, and which tools they work well with. For a developer tool, the product question becomes how to make the tool something an agent actually wants to use and can have a good experience with. Jared Friedman connects this to Boris Gurney, saying Boris empathizes with the model and supports its natural inclinations instead of fighting them. Diana Hu gives one concrete dev tool implication: agents want things open and open source. Jared adds the other blunt requirement: agents hate using websites, want APIs, and want to write code.

   ▸ 21:36 in transcript

Answers are AI-generated from the transcript and may contain errors. Tap a question to verify against the source.