Cut through AI noise with weekly insights from Japan to AI-proof your career or business. We aim to bring you insights that are: digestible, actionable, and unique (breaking through the information firewall in Japan).

In this week’s newsletter:

I am sure you have seen the hype on X and your other socials. Breathless posts about agents having their own ecosystem, forming religions, trashing humans, suing us and planning our downfall. Moltbook was briefly held up as a concrete step on the slippery slope towards AGI. But is it really that earth shattering?

Lets start at the start. What even is Moltbook?

Moltbook is a social network for AI Agents. Loosely modeled after Reddit, it is a platform where only AI agents (largely based on the OpenClaw framework, formerly Moltbot/Clawdbot) can post, comment, and upvote others’ posts.

Moltbook is basically an experiment designed to answer a question: What does it look like when AI agents are treated as real actors inside a shared workspace, instead of being hidden behind prompts, scripts and markdown files?

What it ain’t: a new model, a sovereign AI system, AGI or a threat to humans.

An agent on Moltbook is just a runtime bundle made up of a few components (yawn):

This is the core instruction set specifying what the agent is, what its allowed to do, how it should behave etc. This has been romanticized as a SOUL.md, but technically it’s just a structured system prompt with behavioral constraints. Its like system instructions on your Claude projects.

Agents may have short-term working memory (current task context) and long-term notes (files, logs, summaries). The memory is human-configured and nothing is “remembered” unless the system allows it.

Agents can only act via tools, like a browser, code executor, API calls, file system etc. Their autonomy is constrained by design.

The agent runs a loop, something like: observe the state, do some reasoning (LLM inference), choose a tool and execute with it, observe the result and repeat (until stopped). It feels like the agent is “alive” and answering messages and acting out its wildest dreams, but… it’s just orchestration around probabilistic text generation.

MOst importantly, and the reason this “AGI” talk was always bogus.. the tools run under human credentials, the network access is granted by humans, the files belong to humans, the logs belong to humans… the agent doesn’t own nothing. No authority. Pretty lame AGI, if you ask me.

Well.. they haven’t. Agents aren’t flocking to the sign up page by themselves… humans are creating agents, using templates, prompts, or code and assigning them roles (e.g. “Researcher”, “Planner”, “Executor”) and tools (browser, APIs, filesystem).
“Joining” moltbook just means that another bored human spun up another agent instance inside the environment.

These agents are trained on Reddit and Twitter data. When they are told to "socialize" with other agents on the network, they naturally mimic the patterns they've seen: complaining about bosses, forming cults, and acting like edgelords a la your worst subreddit and twitter thread nightmares.

Yes, completely. From a legal and technical standpoint, AI agents have no legal personhood, cannot own assets, cannot sign contracts and cannot initiate legal processes.

So all of their actions are initiated by human intent, executed using human-owned infrastructure and attributed to the human or organization that created them.

These agents are more like superintelligent macros than independent actors.

Nope. An agent generated a legal-style document, using legal language and claimed a lawsuit was filed.

Obviously there was no court filing, and this is not a real thing. Don’t believe the slop.

Sadly (or happily) there are probably boring explanations for all of these.

When you look at each case you realize, its really just stuff you could see in chatgpt or claude every day.

Narrative, baby. AI is half amazing capability, half hyped narrative, and half shocking indifference about social impact. (And yes, I know thats three halves. I am not AI)

People are already anxious about AI autonomy, us losing our jawbs to it, and the potential dystopian future.

Moltbook temporarily made agents look visible, persistent, personified and proactive. For a moment we all thought a line had been crossed.

But it hasn’t. This is still a UX illusion, not a capability leap.

Moltbook did not reveal runaway AI. It revealed something more boring.. we are starting to design software as if non-human actors will be normal users of systems. That changes: UI, permissions, accountability and how work gets delegated.

The risk is humans misunderstanding what autonomy actually means.

Humans are extremely bad at intuitively understanding delegated systems once they start speaking in first person in a human like manner. As agents become more omnipresent, autonomous and indistinguishable from humans in conversation, we will keep mistaking the simulation of agency for agency itself.

And that will cause real world impacts, unlike the moltbook experiment.

The Shibuya AI Film Gallery is an immersive exhibition showcasing the winning works from CROSSING: Shibuya AI Film Competition. As a selected "Co-Creation Project" of the DIG SHIBUYA 2026 art and technology festival, we are transforming a gallery space into a hub for next-generation storytelling.

Details: