I recently got asked to join a well known crypto company trying to grow their L2. It was a typical, fly to every conference try to sign partners, be present at hacker houses type thing. 

I told them they were wasting their time. The only thing that matters is their AI tooling (they have none). That’s the only thing that will bring growth and everything else is just a Press Release.

They were not ready to accept the truth.

Base’s CDP llms.txt. Celo’s MCP Server. Your MCP Server. Various doc sites that have an LLM built in. I’ve seen good miniapp scaffolding/templates but not good ones made specifically for working with Agents. Documentation teaching others how to use it all, beyond the basic X streams, but those are good too.

I like baking /goodbread and building apps on web3 protocols. DevRel @ ENS Labs

i've never used an mcp server before building my own the other day, and i generally don't like vibe coding full features or apps. i just tab in cursor and ask the chat sidebar questions

the value of llms.txt is clear to me and great. the value of context7 (mcp server that returns data about libraries or something) seems cool as well and i've installed it. not used yet

i don't really understand what a standalone mcp server for docs would look like outside of that. can you explain?

what's an example of ai tooling for a chain?

🏗️ crypto builder  |  
💼 @gigbot.eth  |  
💸 paymagic.xyz  |  
🌐 nf.td/corbin

That's a tough one because there are so many tools and you have to figure out what's write. Definitely tools for your developer or to effectively increase the size of your dev team. Claude Code is my favorite but a lot of people are still using Cursor. Then I think you can actually get a ton done with one of the Claude or ChatGPT plans, using things like Projects. I have a whole product with Scout Game info and any time I need to create a partner doc, a cast, a marketing thing, it is so fast and has all the context. People are doing crazy motion graphics ads with Veo.

What AI Agents / Tools are most important for crypto startups?

Support, dev rel, marketing, and maybe BDR?

What would you recommend?

Man the piñata llms.txt was a lifesaver when I was doing some work with it a few weeks back.

DX Engineer | DevRel at OpenZeppelin | Building orbiter.host & bhvr.dev | opinions are my own

I would add the importance of AI discoverable content. Similar tactics of SEO where you want to pump out as much written and video content and make it easy for bots to scrape it. We took this approach with Pinata and wrote hundreds of blog posts and tutorials, and when people asked ChatGPT how to do xyz with IPFS we came up all the time. Just like legacy SEO it does take time to populate that space but still super important imo since agents are also googling and using search results.

they do have... https://docs.cdp.coinbase.com/llms.txt and https://docs.cdp.coinbase.com/llms-full.txt as a start

What other tooling do you wish they had, but don't? I agree, this is the best way forward - as more developers rely on their AI toolchain, it is ai that will help pick a winner

Applied AI, databases, data, vibe coding (sic.), open source, community/devrel, remote. 

itinerant businessperson. globalist.

Kuala Lumpur, Federal Territory of Kuala Lumpur, Malaysia

haha, oops. yup - i was thinking base was doing pretty well :) 

i guess i have no idea whom you're talking about, but if you wanna share, feel free to dm.

this whole event driven strategy suggests its marketing metric driven. devrel should be in the eng org

It wasn’t Base. I agree they are doing better than most and as a result a lot of people have success building on Base. An MCP server would be next. A prompt library would be good.

brand strategist & founder http://Unknownventures.io ▪️ web3 since "20 ▪️ resident ns.com

how has the worked out in crypto over the last 5 years, where there is still very few developers or users?

Disagree - human connections will always matter - Al traditional marketing tactics will die because of AI.

I do think there’s room for a fast chain with good agent sdk to make a splash

its hard to bring founders out of their delusions, they would rather waste 2 years to learn the lesson

That's the issue we're gonna be facing when web2 founders/dev started coming to web3.

You didn’t reject the offer, you dodged a bullet. Chasing mindshare without tech is just noise.