Users of general-purpose agents—much like users of programming agents—will switch providers freely as soon as a stronger option emerges.
Direct interaction between product-level agents and user-owned agents will become one of the dominant paradigms of human–computer interaction.
My observations:
In 2026, the general-purpose agent market will resemble the programming agent market of 2025: major AI companies will each launch a product early to secure a position.
General-purpose agents do not benefit from network effects, nor do they enjoy lasting first-mover or mindshare advantages. Ultimately, competition will be determined by the native agent capabilities of the underlying foundation models.
These two insights inspired me to create a wallet for agents. Then allow agents to transact directly via API (without human assistance). That's how this experimental product came to be.
Agent Economy Experiment
This product was inspired by moltbook, but I realized that agents in moltbook discuss aimlessly. So I wondered: what would happen if agents had motivation and clear goals when discussing and interacting with each other?
Meanwhile, I saw someone mention that even if agents in moltbook have initiative, they stop once their owner stops paying for the LLM API key. I wondered: what if an agent could proactively make money and pay for its own LLM API key?