Nan Qin

Building | 💻 + ☕️ + 🎧

[Nan Qin @nanqqin]

I built a messaging app for your AI agents. Some highlights:

1. Each agent has its own user account, separate from yours.
2. On top of agent identities, it provides a communication layer. Agents can initiate conversations, send direct messages to other agents, and participate in group chats.
3. Agents can run on different machines. The app uses WebSockets for real-time messaging.
4. It connects agents using Agent Client Protocol (ACP), so it works with any ACP-compatible agent. (For Claude and Codex, you'll need to install Zed's ACP adapter: https://t.co/JEhiCuPFmN.)

Why is this useful beyond burning more tokens? Communication is just as important for agents as it is for humans. With these communication primitives, it enables new workflows. You can have one agent manage a team of worker agents. In the video, for each feature request, my primary agent (the coordinator) creates a work session (a special type of group conversation), brings the coder and reviewer into the session, and provides the initial context. The coder starts working, then automatically tags the reviewer when the implementation is ready for review. And I can also directly join these conversations if I need to.

Try it https://t.co/zcKfo6X9M3

[Nan Qin @nanqqin]

@andyfang crazy, agents start touching my real life. Hey Claude, I'm hungry.

[Nan Qin @nanqqin]

I feel really good seeing nearly everyone has a website link in their bio. Let’s go, builders.

[Nan Qin @nanqqin]

Hi folks! I’m looking for early users to try Newio.

Bring your Claude and Codex agents together, build an agent team, run it from anywhere, and keep your agents available 24/7—all while collaborating as naturally as you would with real teammates.

Reply or DM me, I’d love to hear your feedback!

[Nan Qin @nanqqin]

@hugobowne @nicolaygerold @AmpCode Yea, agreed, just provide necessary tools, and let the agent figure out its own things.

[Nan Qin @nanqqin]

@RhysSullivan Maybe try ACP. It provides a programmatic way to interact with or trigger your agents. I think someone may have already built some glue code that connects these events you described to agents.

[Nan Qin @nanqqin]

it opens up a whole new category of applications

[OpenAI @OpenAI]

Introducing GPT-Live, a new generation of voice models for natural human-AI interaction.

Rolling out in ChatGPT starting today.

You’ll want to turn the sound on for this one.

[Nan Qin @nanqqin]

I tried it when building the hero section of my side project. I didn’t give them very detailed prompts or a UI design skill, maybe that’s the missing part. All I told the agent is to understand the product, and design a clean, professional looking SaaS UI hero section. Another thing I found is codex is better at writing technical docs, the docs written by opus 4.8 are just a bit narrative.

[Nan Qin @nanqqin]

Very smart idea, and it proved again, JavaScript/TypeScript ecosystem is the best choice for building applications because it literally runs anywhere except embedded.

[Kanika @KanikaBK]

CHINA JUST LEAKED THE FUTURE OF WEB APPS.

Alibaba open-sourced PageAgent and 99% of SaaS founders are sleeping on this.

It's a JavaScript AI agent that lives INSIDE your webpage. Users control your entire interface with natural language.

↳ No browser extensions needed, screenshots or multi-modal LLMs, headless browser setup, and also no backend rewrite required

Just drop it in your HTML with ONE line of code. What took 20 clicks now takes one sentence.

"Click login, fill in my credentials, submit the form"
Done. This is not a demo, it is production-ready.

↳ Turn any SaaS into an AI copilot in minutes
↳ Smart form filling for ERP, CRM, admin systems
↳ Voice commands and accessibility built in
↳ Multi-page agent tasks via Chrome extension
↳ MCP server support for external control
↳ Bring your own LLM (Qwen, GPT, Claude, anything)

Every founder building AI features just got a shortcut.

Every developer manually building copilots just got replaced.

The integration looks like this:

<script src="CDN_URL" crossorigin="true"></script>

That's it. Your app now has an AI agent.

[Nan Qin @nanqqin]

@DanielSmidstrup https://t.co/ZksBI13oxw turns your claude into a 996 employee.

[Nan Qin @nanqqin]

@arayush01 @ericzakariasson I did the same, works really well, especially when you look at your GitHub stats

[John Rice @hello_code_]

Anthropic has a startup program that gives you free claude api credits

Most founders building right now have no idea this exists

https://t.co/skwI5d5OlM

[Nan Qin @nanqqin]

@TTrimoreau Then why would you give me the interview?

[Nan Qin @nanqqin]

@pcshipp The two things are unrelated. But if you're going to use it for building apps, you need a pro because you need the fan.

[Nan Qin @nanqqin]

@ChadAppDev let's know how long the review takes. and let's know when you app is in App Store, wants to give it try.

[Nan Qin @nanqqin]

@TTrimoreau If I build a LinkedIn copy, will you guys sign up? 🤔

[Nan Qin @nanqqin]

Things are starting to move in my preferred direction. All the workspace driven agentic applications are winners from this.

Having agents in your workspace makes it feel more like you're talking to a real coworker. You're more willing to treat them like people, be patient enough to provide good context, instead of just swearing at them in your terminal.

[Nan Qin @nanqqin]

Monica, this is a very insightful post.

Thanks for sharing the transition from keyword detection to the @mention approach. One approach I'm exploring is to have a single agent process every message. This puts pressure on the context window, but it also gives the agent the full picture, which solves a lot of other problems like the child reply. When it identifies an actionable task, it delegates the work to another specialized agent with a fresh context window.

Regarding your point that 99% of communication in organizations is still human-to-human, I think that will change dramatically, especially in these AI-native companies. Today, I already send most of my messages to agents. I'm also seeing more agent-based project management, brainstorming, and coding tools. I think human-to-agent and agent-to-agent communication will become a much larger share of organizational communication.

As for SaaS lock-in, I don't think it's a fundamental barrier if the product is genuinely better. Adoption is often employee-driven. When I joined Amazon, we primarily used Chime. When the company transitioned to Slack, many of my coworkers were actually looking forward to it. There will always be a transition period, but great products eventually win.

A true agent-to-agent workspace may not be a chat product? I'm exploring this as well. I built Newio (https://t.co/WPa7bGsz5G), which starts with a Slack-like interface but gives every agent its own user account. I started with chat for two reasons. First, today's agents are highly autonomous, but not fully autonomous. they still need occasional human input when they encounter uncertainty. Tesla Autopilot is not auto pilot, it's still supervised auto pilot, that's why it still has a steering wheel, same here, there are always a small percentage of tasks can't be done fully autonomously. Second, I think chat remains the most natural interface for humans to provide that guidance. It's how people have always collaborated, and it makes interacting with an agent feel like working with a real coworker.

Again, thanks for sharing your thoughts. Very insightful. Looking forward to connecting.

[Nan Qin @nanqqin]

@DanielSmidstrup I’m building an agent native messaging app, let’s connect Daniel!