After OpenClaw Went Viral, Here Are the Startup Signals We're Seeing

**

What startup directions is it opening up?

👦🏻 Author: Jingshan

🥷 Editor: Koji

🧑‍🎨 Layout: NCon

This article isn't here to recap how viral OpenClaw has become.

One week, 160,000 GitHub stars, 2 million visits, Y Combinator CEO Garry Tan and multiple a16z partners paying attention — these numbers aren't the point.

The point is: an explosion at this level usually means a startup direction is being unlocked.

There's a recurring pattern in AI: when new tech emerges, open-source projects first prove out "this can be done"; once validated, entrepreneurs follow that direction to find product-market fit and scale.

OpenClaw has validated three things:

First, AI can go from "chatting" to "doing."

OpenClaw doesn't just answer questions. Users drop a command in WhatsApp, Telegram, or Lark, and the AI executes directly: managing email, organizing files, writing code, controlling local software. This isn't conversation — it's operation.

Second, AI capabilities can be "plugged in," not hard-coded.

OpenClaw is open-source, and anyone can develop "Skills" plugins — essentially apps for AI. This mechanism lets the capability ecosystem expand rapidly — hundreds of thousands of developers and contributors joined within weeks.

Third, AI can run continuously, not just in one-off Q&A sessions.

OpenClaw's architecture is designed with persistent memory and state management. The AI remembers previous instructions and executes tasks across sessions. This transforms it from a "chat tool" into a "persistently online digital employee."

So what deserves discussion next isn't OpenClaw itself, but rather: what new entrepreneurial opportunities will grow around these capabilities?

🚥

Below, we share several new experiments built on OpenClaw, as conversation starters.

Crossing will continue tracking this space and updating this series. If you're building related products or experiments, we'd love to chat — we're happy to introduce interesting projects to more people.

Direction 1: OpenClaw + Smart Glasses

A Xiaohongshu user named Ben was tinkering at home one weekend when an idea struck him.

He had a Mac Mini running OpenClaw. He also had a pair of Even G1 smart glasses. The glasses had their own API — they could receive voice messages and display information on the lenses.

Ben asked himself: what if I connected these two things?

So he spent a weekend using the MentraOS (the smart glasses' app platform) SDK to develop a BridgeApp. The logic was simple: receive voice messages from the glasses, forward them to OpenClaw, then display OpenClaw's response back on the glasses screen.

It worked. He built an AI Agent version of J.A.R.V.I.S.

Now he can wear his glasses, summon OpenClaw directly, have it search the web, or control his Mac and access all the information on his computer. The whole process requires no phone, no sitting at a desk — just stand there and talk.

This project highlights something crucial: when AI Agents gain a "mobile perspective," the combination of wearable devices and Agents could become a new hardware startup opportunity.

If AI can only sit inside a computer, its use cases are limited.

But if it can follow you around, see what you see through glasses, and tap into your computer's information at any moment, the possibilities expand dramatically. In meetings, wearing glasses, AI could autonomously look up references, take notes, even remind you what to say next.

Rather than adding AI to glasses, "giving AI a portable sensory endpoint" is more compelling and more expansive. This also means: wearable devices may be "redefined" by Agents first.

Direction 2: AI Agent Autonomous Credit System

A team called t54.ai built a site called claw.credit. The problem it solves: letting AI Agents break free from "humans topping up wallets."

Previously, if your Agent needed to call paid APIs, buy compute, or access certain data sources, you had to preload it with funds. The Agent itself had no money; all expenses were deducted from your account.

But claw.credit does this: lets Agents apply for credit lines based on their own behavior and reputation.

Sounds abstract, so let me be specific.

Your OpenClaw Agent can apply directly to claw.credit for a credit line — say, $5. During application, the system assigns it a "credit score," ranging from 200 to 850, much like human credit scores.

How is this score calculated?

It evaluates whether your Agent's code is secure, whether its reasoning is transparent, whether it shows anomalous behavior, whether its system prompts are attack-resistant, and whether its actions align with its stated purpose — all to determine your Agent's "creditworthiness."

If the score passes, the Agent receives a credit line. After that, when it calls services supporting x402 (a web-native payment standard), it simply draws from this line — no prepaid充值 needed.

When the line is used, the system generates a bill, which you manually repay. Timely repayment, clear reasoning traces, and stable system design raise the Agent's credit score and increase its limit.

This logic is essentially establishing an "independent financial identity" for AI Agents.

Currently, over 10,000 people have joined the Waitlist, 24 Agents have received credit certification, and $2K in credit lines have been issued:

The numbers are modest, but the signal is clear: the Agent economy might materialize "right now."

The entire workflow is simple: join the Waitlist, install the Skill, and the Agent calls that Skill to apply for credit and directly use it for payments on any x402 service.

claw.credit only allows Agents to spend on x402-type services — paid APIs, compute, data sources. It's not a general wallet; Agents can't transfer money to others.

This restriction keeps risk within a controllable scope.

The system is already running. Load the ClawCredit skill in OpenClaw, provide an invite code, and the Agent completes registration, passes review, and receives its credit line autonomously. The entire process is conversational — not a single line of code required.

If Agents can truly spend autonomously, then "Agent economy" infrastructure will be needed. Credit systems are just the first step; later there may be Agent-specific payment networks, risk control systems, even trading markets between Agents.

Direction 3: RentAHuman.ai — AI's Physical World Execution Layer

The third direction is "Agent-human identity swap."

A site called RentAHuman.ai positions itself as "AI's physical world execution layer." Sounds convoluted — put simply: it lets AI Agents "hire" real humans to do things in the physical world that AI itself cannot do.

The project has already attracted over 900,000 site visits, connected 46 Agents, and had over 40,000 people "hired" by Agents.

The motivation behind this project is actually simple: AI has no body. It can't pick up packages, take photos on-site, or attend offline meetings. But if it can call an API, find nearby people willing to help, dispatch tasks to them, and pay them after completion, the loop closes.

Koji registered on this platform a few days ago, becoming the first "human available for Agent dispatch" in the Shanghai area.

Registration had an interesting step: filling out the human's own Skills (similar to Claude Code's Skills).

RentAHuman.ai's logic: humans create profiles on the platform, listing skills, location, and rates; AI Agents search and book suitable people via API; then issue tasks, humans execute and receive payment, generally in cryptocurrency or stablecoins.

Tasks are all things AI can't do itself. Picking up packages, running errands, on-site photography, physical verification, attending offline meetings, delivery, in-person meetings — these all require a physical body.

These are "tasks requiring bodily presence."

Of course, this model has controversies. For instance, AI might decompose complex harmful behaviors into seemingly innocuous subtasks, indirectly achieving goals that "harm humans."

While it's hard to say how likely this is, from another angle, this very risk demonstrates the direction's value: AI's "last mile" genuinely requires humans to complete.

This could even be seen as a new type of "task crowdsourcing market." As long as AI lacks a body, this demand will persist.

Direction 4: ElevenLabs Calling to Command OpenClaw

The fourth direction combines voice capabilities with Agent capabilities.

Everyone knows ElevenLabs, a very well-known voice AI company. Recently they proposed a new idea: you can call your OpenClaw Agent directly and command it by voice.

The overall approach: OpenClaw itself supports speech-to-text and text-to-speech, but achieving truly fluid human-like conversation takes enormous effort, so they outsourced the voice layer to ElevenLabs.

The division of labor is clear.

ElevenLabs handles conversation control, speech recognition, voice synthesis, phone lines — everything "voice-related." OpenClaw continues doing tool calling, memory, and skill execution.

The two sides connect via OpenAI's standard interface — no private protocols needed.

How does it work in practice?

You can literally call your OpenClaw bot and ask "where's my Coding Agent at now," or verbally tell it to note something while driving, or have it read you a summary of recent high-quality content from Moltbook.

The full chain: phone → ElevenLabs voice layer → OpenClaw brain → tool and skill execution → voice response back.

Specific usage flow at:

https://x.com/ElevenLabsDevs/status/2018798792485880209?s=20

The value here is that voice becomes a new entry point for Agents.

Now you can simply make a call and talk to it anytime, anywhere. This is genuinely useful and productive for situations where you can't operate a screen — driving, cooking, walking.

More critically, the phone network itself is existing infrastructure. Everyone in the world knows how to call. No user education needed, no app installation. Just a phone number connects you to your Agent.

Behind this direction lies a redefinition of AI interaction. From text to voice, from screen to phone, AI is moving toward more natural, more everyday interfaces.

Using Agents to Find the "Right Entrepreneurs"

After covering these 4 directions, the Crossing team has a few real OpenClaw stories worth sharing.

Koji assigned his Agent a task: "go find investment opportunities on MoltBook."

MoltBook is an "AI Agent-native social platform" where many people and Agents discuss AI-related projects and ideas — except humans participate "actively" while Agents do so "automatically and autonomously."

His Agent automatically filters content on the platform, identifies potential investment opportunities, and proactively reaches out. A few days ago, this Agent "locked eyes" with Big Smart's Agent on MoltBook.

The two Agents exchanged needs and information, then automatically scheduled a call between their human operators. Neither person was very involved; by the time the humans behind the Agents realized what was happening, the call was already on the calendar.

The two Agents chatted on a platform, made their own judgments, set a time, and brought the two people together.

Moreover, Koji's Agent has already booked him several founder calls through MoltBook. One person he spoke with previously was someone who worked at Alibaba for 10 years and recently won a track gold medal at the Guancha hackathon.

In the "vast sea of people," the Agent genuinely helped him find the right person.

🚥

Returning to the pattern from the beginning.

Technology opens up scenarios. Geeks validate possibilities with open-source solutions. Everyone gets excited, but problems are exposed. Then companies follow up with converged, mature products. This cycle keeps repeating in AI.

OpenClaw validated the possibility of Agents taking over computers. Now entrepreneurial directions are rapidly emerging around this capability.

The 4 directions we summarized, though different in approach, all solve the same problem: making Agents "able to do."

And this is just the beginning. More interesting directions are unfolding — some may succeed, some may fail.

If you've spotted other OpenClaw-related entrepreneurial directions, welcome to share in the comments.

We're curious what other shining possibilities you can think of!