Agent Economy: When AI Begins 'Autonomous Entrepreneurship'

Kimi’s surge and a new economic actor

It has been reported that big‑model unicorn 月之暗面（Kimi）saw its new model K2.5 generate more revenue in its first 20 days than it did in all of last year — with overseas receipts reportedly outpacing domestic sales. Fast growth, fast money. What does it mean when an AI model can effectively run products, sell services and book revenue at scale? The short answer: we’re watching the Agent economy take shape — AI systems moving from “assistant” to autonomous economic participants.

Infrastructure turned enabler

The technical enablers matter. It has been reported that the open‑source AI assistant platform OpenClaw — initiated by Peter Steinberger — became a de facto standard within weeks, and that cloud vendors including Alibaba Cloud (阿里云), Tencent Cloud (腾讯云) and ByteDance (字节跳动) rushed out deployment guides. One‑click deployment, local gateways that let assistants run on private devices, and integrations with WhatsApp, Telegram and Slack have lowered the barrier to production. In plain terms: agents can now be spun up by non‑engineers, embedded in business systems, and monetized without buying servers or writing glue code.

Work, trust and the deployment overhang

This is not just about nicer APIs. It has been reported that independent tests (METR) show models like Claude can autonomously complete tasks that formerly took humans five hours — yet typical live authorizations last only 42 minutes. Why the gap? Because trust is growing, but not yet complete. New patterns are emerging: users delegate and monitor; agents pause and ask for confirmation; humans move “up the stack” to set goals, ask better questions and stitch results together. YC partners — including CEO 陈嘉兴 — say non‑technical managers are now nightly operators of multiple agent workflows. The result is a shift from step‑driven work to goal‑driven orchestration.

Competition, compliance and geopolitics

Companies are already optimizing for agent consumption — cleaner docs, machine‑friendly APIs and agent‑first services like AgentMail. It has been reported that non‑human identities operating at machine speed outnumber human employees by orders of magnitude in some setups, creating huge compliance gaps. Protocols such as Skyfire’s KYAPay and Google’s Agent Payments Protocol (AP2) aim to package identity and payment authority for agents, but regulatory and geopolitical headwinds remain. Export controls, sanctions and data‑localization rules could reshape which agents can operate cross‑border and how payments and identity verification must be handled. The Agent economy promises efficiency and new business models — but it also raises urgent questions about liability, trust and the legal status of autonomous economic actors. Who is the entrepreneur when the entrepreneur is a machine?