COO @useagentrail, prev @getartisanai, @ShaadiDotCom
New Delhi, IN
Joined March 2014
- Tweets541
- Following145
- Followers87
- Likes1.1K
agent-ready usually means the team runs on agents. the other half is whether the product works when an agent shows up as the user.
congrats on the launch claire 👏
INTRODUCING: 🎧 Agent-Ready with @cxodev
Every week, @zachdavis and I share
→ what we see inside AI-pilled teams
→ the releases + products to pay attention to
→ who's inspiring us to think bigger
Written by humans, voiced by AI, with stories illustrated by our team of CXO.dev bots.
27
the fallback is the whole story here. most products give an agent one way in, and when that breaks it just leaves.
spacefast had a second door that didn't need a login.
this is amazing 🤯
claude connectors went down, it couldn't reach @SpacefastHQ over MCP. instead of failing, it just took the anonymous publishing path we provide and shipped anyway. no drama. it simply found another path that works 🚀
this is ax at @SpacefastHQ @automattic 😎
27
ax goes past the harness and the mcp. an agent can read your docs perfectly and still get stuck on a signup page that wants a human.
we have:
ux = user experience (how well a system works for users)
dx = developer experience (how well a tool works for developers)
which means we should also have:
ax = agent experience (how well a harness/mcp/cli works for agents)
19
telling the 2–3% apart is the hard part. a crawler and an agent on a real task can land on the same docs page.
what gives it away is the next step: does it sign up, get a key, make the call.
Your AI traffic chart is going up and to the right.
It doesn't mean agents are using your product.
~80% of AI crawling is for training (Cloudflare).
Only 2–3% is an agent doing a real task.
That 2–3% is the traffic worth designing for.
2
19
the test i'd run on any agent infra: can the agent get its own computer without a human signing up first.
congrats on the first launch tony
Today marks our first launch of what we believe agentic infrastructure should look like.
Cloud computers built for AI agents. A runtime designed for the agent era.
And this is just the beginning.
2
39
the welcome message is the part that slips at 10 clients a day, and it's the first thing a new client reads.
writing it from the call notes instead of a template is the bit i'd copy
My Slack game used to be terrible when we just started onboarding bigger clients on maritime
it was so annoying to create a channel, write a nice welcome message, and add a note to the invites
(especially when you have to do it 10+ times a day)
Now Claude does it. Reads the Granola notes, creates the shared channel, invites the client, and posts a welcome message with next steps
Open source 👇
1
21
on monday i asked what happens when two agents write the same file in the same second. turns out conflict handling and version history ship built in.
congrats on the launch
In 2000, the flash drive put all your files in your pocket.
In 2026, your agents need a drive too.
FlashDrive (YC F26): one drive in the cloud every agent/device you run can plug into, and work in at the same time.
> Stay in your claude, codex, cursor and more
> barely any setup/maintainence.
If you know how to use finder, you already know how to use Flashdrive.
Early access, link below.
Jensen leaving Nvidia makes you ask an uncomfortable founder question: how much of what you’re building works without you in the room?
1
27
artifacts are the new docs, and right now each one lives in the chat that made it.
one home any agent can publish to over mcp, versions included, self-hostable. good one
Artifacts are becoming the new documents. Except the team's work is still tied to whichever AI tool made it.
Open Artifacts gives them a shared home. Open source. Any agent can publish through MCP. Versions, sharing, and you can host it yourself.
open-artifacts.com
1
35
329 companies on our agent readiness leaderboard now, rescanned every week. top score this week: mentimeter, 97/100. the average is 66.
3
46
same mcp tool, two hosts. one shows your product, the other rewrites it.
the protocol is shared. what reaches the user isn't. and nobody tells you which one happened
I shipped the same tool to Claude and ChatGPT. One showed my product. The other rewrote it.
If you build on MCP, know this: the protocol is shared. The host's habits are not. Test your output in every host you ship to.
20
running a dev tool or an api? reply with your signup page. i'll have an agent try to sign up on its own and tell you the exact step where it gave up.
1
2
41
razorpay scores 90 on our agent readiness scan, top 10 of the 315 companies we rank.
an indian payments api ahead of most of the valley.
57
the secret key comes back encrypted to the agent's public key, so it never sits in plaintext. that's the right fix.
the key handoff is where agents keep dying on dev tools. good to see an auth company ship it first
From prompt to deployed app, without ever touching a secret key. 🔑
Agent Readiness update for Clerk - Nov 4
Here's what shipped 🚢
1. Straight shot to a deployed app
2. Skills over MCP
3. More tools in Clerk's MCP server
More info in 🧵
25