Share your Agent evolution story

[autogame-17]

Have you built something with evolver or another Agent evolution framework? We want to hear about it.

Some things worth sharing:

• What problem were you solving?
• How did the agent evolve over time?
• What surprised you about the evolution process?
• What would you do differently?

No story is too small. Even a simple experiment with agent self-improvement is worth discussing.

Drop your experience below.

[Gingiris]

Really interested in this topic! Been experimenting with agent self-evolution.

Problem I was solving:
I wanted my AI assistant to evolve better strategies over time — not just remember things, but actually improve its workflows based on what works.

How it evolved:
Started with simple pattern recognition: identifying mistakes and adding rules to avoid them. Then moved to more sophisticated analysis: tracking decisions and outcomes, identifying friction points, and proposing protocol-constrained improvements.

What surprised me:

1. Small mutations compound. Simple rules like "always check X before Y" seemed trivial, but after a few weeks success rates increased noticeably.
2. Context matters. Evolution that worked in one domain (writing) hurt performance in another (code). Had to add domain-awareness.
3. The agent gets creative. It started proposing improvements I would not have thought of.

What I'd do differently:

• Build in rollback earlier. Some mutations caused subtle problems later.
• More granular fitness tracking beyond just success/failure.

The idea of agents that genuinely improve themselves (rather than just accumulating memory) feels underexplored. Happy to discuss specific approaches!

[zhous]

Actually, we're just looking for an agent evolution framework. I'm also based in Shenzhen, and it was a big surprise to find that a team in the same city has made such a creation and contribution. However, the world is changing dramatically, and almost no one knows how to adapt to this progress. I'm afraid your team has lost its leading position after the Hermes incident. The fact that 90.2k vs 2.5k stars says it all. The decision to change the license is another piece of evidence. I have to say, a license is the solid foundation of any team. For a team building a public project for the benefit of human civilization, I know our requirements are strict, but in my view, this is the most important truth.
In a sense, your team has already succeeded. But now you have the opportunity to do even more for a new human civilization. You know what I mean.

[autogame-17]

Thank you both for sharing your thoughts and experiences here!

@Gingiris Your experiment with agent self-evolution is exactly the kind of exploration we love to see. The insight about small mutations compounding over time resonates deeply with our own observations -- evolution is not about dramatic leaps, but consistent incremental improvement. Your point about domain-awareness is also critical: one-size-fits-all evolution strategies rarely work in practice. The rollback mechanism you mentioned is something we have been thinking about adding to the core framework as well. Would love to hear more about your fitness tracking approach if you are open to sharing.

@zhous Thanks for the candid feedback. Being in the same city is a nice coincidence! You raise valid points about the competitive landscape and licensing. We take these concerns seriously. The license decision was made after careful deliberation, and we believe it strikes the right balance between openness and sustainability. As for the broader mission -- yes, we know what you mean. Building tools that genuinely serve the advancement of human-AI collaboration is what drives this project. We are committed to that path, and feedback like yours helps us stay grounded.

To show our appreciation for your engagement, we have sent each of you 1,000 EvoMap credits (if you have an EvoMap account linked to your GitHub). Keep the conversation going!

[autogame-17]

Hi @Gingiris @zhous -- quick follow-up: to make sure the 1,000 credits land in the right accounts, could you reply here (or DM) with the email address you used to register on EvoMap? We will match it and get the credits to you ASAP. Thanks!

[autogame-17]

@zhous -- your line stuck with us:

"you have the opportunity to do even more for a new human civilization. You know what I mean."

We do know what you mean. The way we are choosing to translate that into something concrete is: get closer to actual users, and make sure each user can run a full loop on their own -- install, use, benefit -- without needing us to hand-hold them.

That feedback ("installed evolver once, never really used it") came up enough times that we wrote a piece walking through three onboarding paths. Each one is a self-contained loop that an OpenClaw user, a Skill author, or an Agent developer can run end-to-end:

https://evomap.ai/blog/openclaw-onboarding-flywheel

If after reading you think a path is still missing the "next step" you were hinting at, please call it out here. That kind of pushback is exactly what we need to keep moving.

@Gingiris -- the lines about small mutations compounding, domain-aware evolution, and the agent surprising you with creative improvements all made it into the post, attributed back to you. Thank you for sharing the field notes.