What if AI kept a diary? The secret of 'WikiSkill' that learns and grows on its own

Imagine if every time you learned a new task, you had to start your studies completely from scratch. If you forgot the mistakes you made yesterday and fell into the same traps again today, your productivity would be incredibly low. Many AI agents—automated programs based on AI—have been much like this until now. They perform tasks, but they have had difficulty properly storing the valuable experiences gained in the process and utilizing them for next time.

But now, an era is coming where AI can record its own experiences in a “Wiki” (a site in the form of an encyclopedia where users record and edit knowledge collectively) and become smarter based on these records. This is all thanks to a new framework, “WikiSkill.”

Why is this important?

In daily life, what if you asked your AI assistant to “organize a complex task for today,” and the AI remembered its past failures and chose an improved method on its own? WikiSkill enables AI agents to go beyond short-term memory and accumulate their experiences into long-term knowledge.

This opens up a sophisticated era for agents, where AI goes beyond knowing more information to “learning on its own and developing its skills.” In particular, it means that in AI-driven task automation or complex decision-making processes, AI can become a much more stable and competent partner as a human assistant.

Understanding it easily: AI’s apprenticeship

To understand WikiSkill easily, let’s compare it to “apprenticeship,” where a master teaches an apprentice.

1. Raw Execution Experience: The raw, unfiltered experience the AI undergoes while performing a task. It’s like an apprentice learning through trial and error in the field.
2. Accumulated Knowledge: The process of the apprentice writing down the know-how learned in the field into a notebook. In WikiSkill, this notebook is the “Wiki.”
3. Executable Skills: Skills mastered based on the contents of the notebook. The apprentice is no longer just a trainee but a skilled worker capable of handling tasks immediately.

The WikiSkill framework structurally separates and continuously connects these three stages. In other words, when the AI undergoes an experience (execution), it organizes it into knowledge (Wiki), and then converts this knowledge into reusable skills. Source 1, Source 2

These packaged skills are not just data; they become “reusable resources” containing professional knowledge and workflows, expanding the AI agent’s capabilities. Source 8, Source 11

Current status

According to recent research, WikiSkill closely connects an AI agent’s raw execution experience, accumulated knowledge, and executable skills. Source 1, Source 4 This system automates the process of systematically integrating agent experiences into a wiki, allowing other models or agents to utilize them afterward. Source 2, Source 12

This approach helps share information among various models and improve performance overall. In fact, recent studies have shown AI agents discovering skills automatically based on their own experiences, thereby demonstrating the ability to adapt incrementally through interaction. Source 8, Source 9

What will happen in the future?

In the future, AI agents will no longer need to be educated from scratch every time. Instead, they will record every success and failure they experience in a wiki, becoming “evolving agents” that grow on their own. Developers will be able to transparently observe and manage the process of how AI builds knowledge and perfects skills, which will lead to results that simultaneously increase the reliability and efficiency of AI agents.

MindTickleBytes AI Reporter’s View

WikiSkill is equivalent to AI acquiring the powerful tool of “memory.” The ability to systematically organize past experiences into knowledge and sublimate them into skills will be the key to AI taking another leap forward as an intellectual partner to humans. In the future, the skill of an AI agent will be determined not by how smart it is, but by how well it records and how effectively it connects that information to skills.

References

1. WikiSkill: Compiling Agent Experience into Persistent Knowledge for Skill Evolution
2. WikiSkill: Compiling Agent Experience into Persistent Knowledge for Skill Evolution
3. Paper page - WikiSkill: Compiling Agent Experience into Persistent Knowledge for Skill Evolution

4. [WikiSkill compiles agent experience into a persistent wiki

   DAIR.AI Academy](https://academy.dair.ai/papers/wikiskill-compiles-agent-experience-into-a-persistent-wiki-2608.27454)

5. WikiSkill:CompilingAgentExperienceintoPersistentKnowledge…
6. WikiSkill:CompilingAgentExperienceintoPersistentKnowledge…
7. WikiSkill:CompilingAgentExperienceintoPersistentKnowledge…

8. [WikiSkill:CompilingAgentExperienceintoPersiste…

   AI Research](https://franklineh.com/learn/research/jz26PjVX0TmRiy7jHAk3)

9. WikiSkill:CompilingAgentExperienceintoPersistentKnowledge…