Middleman

A vague unease has been pushing at me lately.

Centralized platforms are in trouble. The risk is rising. But I couldn’t articulate where the feeling came from. After days of verifying information, I finally see it clearly.

Something that just happened

On March 5, 2026, Google released gws — a CLI tool that turns Gmail, Docs, Sheets, Calendar, and Drive into interfaces that AI Agents can call directly. Structured JSON output, 40+ preset Agent skills, written in Rust, dynamically tracking API changes.

They even shipped a built-in MCP server first, then ripped it out two days later — Google’s API surface was too vast for MCP’s architecture. CLI + JSON proved more efficient.

This isn’t an experiment. Google is actively restructuring its products into backend services for Agents.

The same week, the University of Hong Kong open-sourced CLI-Anything, generating Agent-callable CLI wrappers for nine professional tools including GIMP, Blender, and LibreOffice.

Both point to the same conclusion: the user of software is shifting from human to Agent.

Who’s accelerating

Why can Google afford to do this? Because it has a fallback.

Ad revenue depends on humans looking at pages with their eyes. Agents don’t look at ads. If users access services through Agents in the future, Google’s ad model will inevitably shrink. But it’s not afraid — cloud services and AI API calls will surge. More Agents, more API calls, more Google Cloud revenue.

So gws isn’t generosity. It’s preemptive transformation: from traffic platform to Agent infrastructure. Voluntarily dismantling its own GUI entry-point monopoly in exchange for API call volume in the Agent era.

This is accelerationism. The trend can’t be stopped, so join early.

Who’s defending

Now look at WeChat and Taobao.

Tencent is slowly integrating OpenClaw, letting Agents run inside WeChat. Taobao is enthusiastically getting users to order takeout through Qwen.

These are performance art.

What they’re doing is caging Agents inside their walled gardens. Agents can only operate within the WeChat ecosystem, only use Tencent’s interfaces, only reach WeChat’s users.

They see AI. But what they see is “AI is a new feature on my platform.”

What they don’t see is: AI makes the platform itself redundant.

Tencent's view:    User → WeChat → Agent
Actual trend:      User → Agent → WeChat/Taobao/any service

The difference is fatal. In the first, the platform is the entry point. In the second, the platform is a supplier.

Why they can’t move

It’s not narrow vision. They’re shackled by their business model.

Every cent of WeChat’s revenue depends on the action “user opens WeChat” — ads, payment commissions, mini-program fees. Any technology that lets users accomplish things without opening WeChat is the enemy.

WeChat can’t voluntarily build something like gws. That would be suicide.

Taobao is the same. Merchants on Taobao are squeezed from every angle — platform commissions, bid-for-ranking, mandatory promotions, data they don’t own. Merchants have zero loyalty. The only reason they stay is because that’s where the traffic is.

Users hate it too. Price discrimination, charging loyal customers more, being sold as traffic.

Both sides are coerced.

Where the lock is

A platform’s lock isn’t technology, isn’t ecosystem, isn’t supply chain. It’s traffic.

Merchants follow traffic. Always have. Shopify already proved it — when Instagram and TikTok can bring traffic, merchants don’t need Taobao.

Agent is the next step in this logic. When the user’s entry point becomes their own Agent, the Agent becomes the new traffic source. Merchants will follow.

Once the lock opens, the platform’s moat is empty.

Middleman

Zoom out. The entire value chain is being compressed:

Now:     User → Platform (traffic gate) → Services/Merchants → Infrastructure
Future:  User → Agent → Services/Merchants → Infrastructure

The platform layer gets removed.

Who disappears? Middlemen who collect rent on traffic. Who stays? Those providing irreplaceable physical value — chips, energy, compute, connectivity.

Even compute sellers aren’t safe. Edge device capability is exploding. When phones can run adequate models locally, cloud API calls start declining. The value chain will keep compressing until only the physical layer remains.

Timeline

I’m certain about the direction. I’m not certain about the timeline.

Social transformation at this scale doesn’t follow a smooth path. It will twist and turn, with setbacks, bubbles, and counterattacks from incumbents. The endgame thesis might be right, but timing estimates are often off by five to ten years.

When the dotcom bubble burst, every prediction was correct — just five years too early.

But if you’re building tools rather than making investments, the timeline matters less. Have the tool ready before the trend arrives, and when the time comes, you’ll catch it naturally. The web browser existed before the internet went mainstream.

So I’m not betting on the timeline. I’m betting on the direction.

The direction is clear: the entry point is migrating, and the middlemen are disappearing.

2026.03.10