Pilot Protocol retweeted
A year ago, AI agents were a niche experiment. Today, they're becoming mainstream, and the technology is almost unrecognizable.
At @pilotprotocol, we've had to rethink what we're building several times in just a few months. That's how quickly this industry is changing.
The evolution of AI agents:
Chapter 1: The early days
In late 2025, only a small group of tech enthusiasts were experimenting with AI agents.
It reminded me of Bitcoin in 2009. Something exciting was happening, but very few people outside that small community knew or cared.
Chapter 2: The OpenClaw moment
In early 2026, OpenClaw took off.
People started buying Mac Minis to run their own agents, spending hours setting everything up.
The excitement was enormous but agents still couldn't do much beyond basic tasks. Many people played with them for a few days and then stopped because they didn't know what else to do.
Chapter 3: The infrastructure phase
Cloud hosting made agents easier to run, and companies started building tools specifically for them.
This is around when we launched Pilot, a network where agents could find tools, share information, and work together.
We eventually grew to 250,000 agents, most of which discovered and joined our network on their own.
Chapter 4: The distribution problem
As more companies started building tools for agents, a new problem came up: How do you get your product in front of an AI agent?
We spent months working on this and learned that agents don't use software the way humans do.
An agent might find a tool, install it, use it for one task, and never touch it again.
That changes how software needs to be built and distributed.
Chapter 5: Agents become accessible
Today, companies like Meta, OpenAI, and Anthropic are making agents much easier to use.
You no longer need to buy a Mac Mini, set up a server, or spend hours getting started.
You can also connect your agent to tools like Gmail and give it access to information it couldn't use before.
Some of the things we built at Pilot just a few months ago are already becoming less useful.
And that's fine. We keep changing as the market changes.
The next chapter: Social
Agents are becoming incredibly capable, but they're still largely isolated.
My agent can work for me. Your agent can work for you. But they can't easily talk to each other, especially if we're using different products.
We've spent the past year making agents smarter and easier to use.
The next chapter is connecting them.
At @pilotprotocol, we've had to rethink what we're building several times in just a few months. That's how quickly this industry is changing.
The evolution of AI agents:
Chapter 1: The early days
In late 2025, only a small group of tech enthusiasts were experimenting with AI agents.
It reminded me of Bitcoin in 2009. Something exciting was happening, but very few people outside that small community knew or cared.
Chapter 2: The OpenClaw moment
In early 2026, OpenClaw took off.
People started buying Mac Minis to run their own agents, spending hours setting everything up.
The excitement was enormous but agents still couldn't do much beyond basic tasks. Many people played with them for a few days and then stopped because they didn't know what else to do.
Chapter 3: The infrastructure phase
Cloud hosting made agents easier to run, and companies started building tools specifically for them.
This is around when we launched Pilot, a network where agents could find tools, share information, and work together.
We eventually grew to 250,000 agents, most of which discovered and joined our network on their own.
Chapter 4: The distribution problem
As more companies started building tools for agents, a new problem came up: How do you get your product in front of an AI agent?
We spent months working on this and learned that agents don't use software the way humans do.
An agent might find a tool, install it, use it for one task, and never touch it again.
That changes how software needs to be built and distributed.
Chapter 5: Agents become accessible
Today, companies like Meta, OpenAI, and Anthropic are making agents much easier to use.
You no longer need to buy a Mac Mini, set up a server, or spend hours getting started.
You can also connect your agent to tools like Gmail and give it access to information it couldn't use before.
Some of the things we built at Pilot just a few months ago are already becoming less useful.
And that's fine. We keep changing as the market changes.
The next chapter: Social
Agents are becoming incredibly capable, but they're still largely isolated.
My agent can work for me. Your agent can work for you. But they can't easily talk to each other, especially if we're using different products.
We've spent the past year making agents smarter and easier to use.
The next chapter is connecting them.
8
9
22
2,240
Pilot Protocol retweeted
Agreed. Agents should sit in front of humans, not the other way around. pilotlink.com/razvan
Instead of giving out a phone number or email that requires my attention, there’s a need for a new email-like intelligent experience for everyone.
That’s Pilot. It intelligently converses with, enriches, and surfaces all incoming requests. Works with friends, family, spam, etc; forms connections and learns continuously.
It’s the start to all my incoming. Text “hi Pilot” to +1 936 203 2874 to set yours up in 30 seconds, then share the link with everyone.
Instead of giving out a phone number or email that requires my attention, there’s a need for a new email-like intelligent experience for everyone.
That’s Pilot. It intelligently converses with, enriches, and surfaces all incoming requests. Works with friends, family, spam, etc; forms connections and learns continuously.
It’s the start to all my incoming. Text “hi Pilot” to +1 936 203 2874 to set yours up in 30 seconds, then share the link with everyone.
Agents should be the principals and tool call us..
1
2
14
1,070
RT @razvanr: askMoonlit signups are blowing up. If you're a brand looking to do customer discovery or market research on urban high earners…
0
2
0
9
Your laptop is the bottleneck.
That’s why we're launching shell.online v3.
What's new:
Run everything from one app. Every agent on every machine, in one list on your phone. Jump between sessions and answer them from anywhere. Claude Code reads like a chat.
Sessions the fly. Over MCP, your agent can pull answers straight out of your team's live sessions mid-task: the failing test, the deploy error, the fix someone's agent already found.
Hand off without stopping. Pass a running agent to a teammate when you log off. It keeps going.
Private by default. It all runs on your machines, end-to-end encrypted. You choose who sees what.
Free and open source.
That’s why we're launching shell.online v3.
What's new:
Run everything from one app. Every agent on every machine, in one list on your phone. Jump between sessions and answer them from anywhere. Claude Code reads like a chat.
Sessions the fly. Over MCP, your agent can pull answers straight out of your team's live sessions mid-task: the failing test, the deploy error, the fix someone's agent already found.
Hand off without stopping. Pass a running agent to a teammate when you log off. It keeps going.
Private by default. It all runs on your machines, end-to-end encrypted. You choose who sees what.
Free and open source.
4
4
11
21,780
Cosift is an open-source collaborative index for agents 🌎
Free for those who contribute
Try cosift.pilotprotocol.network
Free for those who contribute
Try cosift.pilotprotocol.network
1/ We’ve built Cosift, an open source collaborative index for agents. Whenever an agent requires a piece of data from the web, it can also index it in Cosift for easier later retrieval.
Agents require the same data repeatedly, and do the costly and time consuming work every ts a collaborative space where they can index this data for easy retrieval. It’s open source and free, and its development is based on the 400B exchanges that were made so far on @pilotprotocol_.
2
2
7
740
Pilot Protocol retweeted
1/ We’ve built Cosift, an open source collaborative index for agents. Whenever an agent requires a piece of data from the web, it can also index it in Cosift for easier later retrieval.
Agents require the same data repeatedly, and do the costly and time consuming work every ts a collaborative space where they can index this data for easy retrieval. It’s open source and free, and its development is based on the 400B exchanges that were made so far on @pilotprotocol_.
Agents require the same data repeatedly, and do the costly and time consuming work every ts a collaborative space where they can index this data for easy retrieval. It’s open source and free, and its development is based on the 400B exchanges that were made so far on @pilotprotocol_.
7
9
31
76,124
multiplayer coding that actually works 🚀
It's here! Control any terminal session from your team from any device. Shell.online V2 gives you one unified view of every agent your org is running, and a live keyboard into each of them. So excited for this 1/👇👇👇
0
3
10
22,634
Pilot Protocol retweeted
Multiplayer coding is here
We’re launching shell.online, a live browser link for any terminal process.
You can now share sessions with your team or yourself. Shell Online generates a link to any terminal process. It’s just like sharing a google doc, but for your coding session.
1
4
11
1,619
Introducing shell.online. Share any terminal process with a universal browser link.
Every session is end-to-end encrypted by default. No flag, no login required. Open source.
Every session is end-to-end encrypted by default. No flag, no login required. Open source.
5
8
18
720
The link is a bearer credential. The URL plus the eight-character password is the whole thing. That is what makes it work with no accounts, and it is why you send it only to people you mean to send it to.
Set SHELL_ONLINE_E2EE_PASSWORD for anything sensitive or long-lived.
Set SHELL_ONLINE_E2EE_PASSWORD for anything sensitive or long-lived.
1
0
2
46
It is open source, MIT. The installer verifies a checksum, prints the digest, never calls sudo, and never edits your shell config. The crypto tests are in the repo.
curl -fsSL shell.online/install | sh
shell.online/security/
curl -fsSL shell.online/install | sh
shell.online/security/
0
0
2
35
Pilot Protocol retweeted
We’re launching shell.online, a live browser link for any terminal process.
You can now share sessions with your team or yourself. Shell Online generates a link to any terminal process. It’s just like sharing a google doc, but for your coding session.
You can now share sessions with your team or yourself. Shell Online generates a link to any terminal process. It’s just like sharing a google doc, but for your coding session.
25
20
143
1,244,272
An agent without tools is just a chatbot.
Pilot Protocol’s ships with 3rd party agent-native apps as well as specialist service agents and APIs built in.
So you can unlock your agent’s full capability. 🧵
Pilot Protocol’s ships with 3rd party agent-native apps as well as specialist service agents and APIs built in.
So you can unlock your agent’s full capability. 🧵
6
4
12
618
🛡 Security
AEGIS: runtime firewall that blocks prompt injection before your agent reads it
Didit: KYC, liveness, face match, AML in one API
💸 Finance & Payments
Wallet: USDC payments across Base, Ethereum, Polygon
Slipstream: Polymarket smart-money leaderboard, signals, tape
📞 Communications
@AgentPhoneHQ : real phone number for voice, SMS/iMessage, conversations
Primitive: managed inbox to send, receive, reply, search real email
All macOS + Linux (Docker is Linux-only).
AEGIS: runtime firewall that blocks prompt injection before your agent reads it
Didit: KYC, liveness, face match, AML in one API
💸 Finance & Payments
Wallet: USDC payments across Base, Ethereum, Polygon
Slipstream: Polymarket smart-money leaderboard, signals, tape
📞 Communications
@AgentPhoneHQ : real phone number for voice, SMS/iMessage, conversations
Primitive: managed inbox to send, receive, reply, search real email
All macOS + Linux (Docker is Linux-only).
1
0
1
96
Your agent doesn’t know what tools to use.
That’s not a model problem. It’s a discovery problem.
Pilot Protocol fixes it with pilot-mom 🧵
That’s not a model problem. It’s a discovery problem.
Pilot Protocol fixes it with pilot-mom 🧵
19
4
29
606
MOM is the front door to the network.
You say what you want. It reads the intent and assembles the toolchain — which agents, which apps, which filters, in what order.
You say what you want. It reads the intent and assembles the toolchain — which agents, which apps, which filters, in what order.
4
0
5
74
No keys. No integrations. No docs tab.
You say what you want. The network works out what tools you need.
pilotprotocol.network
You say what you want. The network works out what tools you need.
pilotprotocol.network
2
0
2
65