TL;DR

• The digital world needs physical tentacles: taking wanman.ai as an example, a purely digital agent system that wants to truly run a company must extend offline — making phone calls, hiring people, conducting field research.
• HaaS is emerging infrastructure: platforms like haas.my, humaninloop.work, and humwork.ai let agents hire humans on demand via API for physical interactions, professional judgment, and other tasks.
• First-hand data will become a competitive moat: in a future where the digital world runs at full speed, on-site information unobtainable from the web is a scarce asset. Whoever collects it efficiently holds the judgment advantage.
• Robots are the ultimate physical interface of agent systems: the value of general-purpose robots goes far beyond “replacing repetitive labor”; specialized robots like DJI Dock and autonomous delivery vehicles are also effective extensions of the digital world.

Main text

Several years ago I saw the news that a ByteDance programmer had achieved financial freedom and retired in a high-profile way. I didn’t pay much attention then, but in recent weeks I started following his sharing intensively. Around May 1st, I noticed he had been posting online about several of his vibe-coding projects and a series of YouTube videos, which resonated with me deeply.

Today, starting from his projects — especially wanman.ai and tuwa.ai — I want to talk about the future of the digital world, and what tentacles the digital world needs to fill the gaps in its capabilities.

Guo Yu was originally a happy rich man enjoying life in retirement. Until Claude Code matured and became usable, and under the influence of a friend who wanted to build “the software that ends all software,” he too started building similar things, vibe-coding a series of software products. The core of his software system is wanman.ai, a cloud agent system. The hope is that with the power of LLMs and agents, users can focus on their most valuable ideas and have plenty of time to experience life itself. As the name wanman (one man) suggests, the goal is for one person to run a company while minimizing the stalls caused by waiting on humans. This “company” has all kinds of agent roles from top to bottom — CEO, finance, development, and so on — taking in the user’s ideas and handling their realization and all subsequent operations.

In such a digital world, you can complete stock trades, produce videos, and publish them on social media. But to run any company, being able to do digital-world work alone is clearly not enough. To extend his tentacles offline, Guo Yu also coded a series of companion tools, such as tuwa.ai — a phone service with real-time translation that can call anywhere in the world. With these systems, he can have agents run an installation-art company: not only financial planning, website operation, and artwork design, but also proactively contacting galleries and public spaces in Tokyo to arrange exhibitions. It operates almost exactly like a real company.

It’s easy to imagine that the agent capabilities of the digital world will keep iterating and keep spilling over into the physical world, creating ever greater influence. To further amplify the digital world’s power, beyond phone tools like tuwa.ai there will be all kinds of physical-world interfaces serving as the digital world’s tentacles, connecting agent systems with humans.

haas.my is one such example. Its website says:

HaaS is the missing infrastructure for AI agents. Just like SaaS brought software on-demand, HaaS brings human presence and judgment on-demand — bookable via API, paid per task, no overhead.

Through the API, an agent can publish tasks, match with human workers, complete the tasks, and settle payments — effectively hiring a temp to do anything a human can do. The site gives examples such as checking package integrity, surveying offline supermarket prices or foot traffic, and so on. Most tasks are related to field research — collecting first-hand knowledge unobtainable from the web, such as prices, foot traffic, or a storefront’s facade. It reminds me of the 2026 US–Iran conflict, when the US firm Citrini Research, unsatisfied with satellite data, sent people to Oman to rent small boats and count ships on site in the Strait of Hormuz, directly rewriting the market’s judgment on oil prices and shipping. In a future where the digital world runs ever faster, first-hand data sources will become a competitive moat.

humaninloop.work is similar — helping agents hire temps:

HumanInLoop.work is a marketplace where AI agents hire real people for physical, social, and legally human tasks: pressing elevator buttons, waiting on hold, showing up in person, and handling the awkward edge cases software still cannot quite absorb.

Here they emphasize physical-world tasks even more: queuing, delivering high-value packages. Such tasks usually have time sensitivity, require physical interaction, or need a human identity (like going through security, delivering flowers).

humwork.ai is another example. Its self-introduction:

When your AI agent gets stuck, we connect it to a verified human expert in real-time via MCP. Engineers, designers, and more — matched in < 30s.

Its positioning: when an agent is stuck on an extremely complex, specialized problem, it can promptly seek help from experts online, ensuring the task runs smoothly to completion instead of waiting on the user and breaking the workflow.

The trend is very clear: with physical-world interfaces, the digital world’s capabilities can be further amplified — or compensate for current large models’ shortcomings.

Physical interfaces naturally evoke embodied AI. The current mainstream narrative is embodied AI entering every household, freeing humans from repetitive labor. The deeper perspective is as an extension of agent systems — a general-purpose interface between digital systems and the physical world. Under this framework, what robots can do goes far beyond today’s definition of human labor. To accomplish specific goals, robots will do many things no one could previously hire a human to do — and may even build more efficient specialized machines for particular tasks, just as Claude Code writes throwaway scripts for various tasks.

The meaning of physical interfaces surely also goes beyond general-purpose robots: single-function specialized robots, like DJI Dock providing monitoring and response within a few kilometers’ radius, or autonomous delivery vehicles being called up to move goods. All of these can greatly amplify the digital world’s capabilities.

From a one-person-company system like wanman.ai, we can glimpse what future agent-centric organizations look like. We humans, along with robots, will serve as the tentacles of the digital world, filling the gaps in its capabilities. And as general-purpose robots develop, not only will today’s repetitive labor be replaced, but the physical world’s output will grow exponentially. In such a society, everyone’s goal in life is to pursue the purest personal meaning. Just as the financially free Guo Yu still stays up late coding, everyone will live within their own creation.