Self Learning Agent Behavior Adaptation

via “steerable model behavior through contextual instruction adaptation”

Multi-turn conversation dataset for steerable models.

Unique: Explicitly includes examples of mid-conversation instruction changes and demonstrates expected model behavior adaptations, rather than treating conversations as static sequences. Teaches models to be responsive to evolving user intent within a single dialogue.

vs others: More sophisticated than static instruction datasets because it includes dynamic instruction changes and demonstrates how models should adapt without losing context, enabling more interactive and user-responsive AI systems.

via “teachable agent with dynamic knowledge acquisition”

Microsoft AutoGen multi-agent conversation samples.

Unique: Separates learning mechanism from agent execution, allowing agents to update behavior via memory system updates without modifying agent code or redeploying; feedback is stored as structured patterns that agents can query during reasoning

vs others: Simpler than fine-tuning approaches because learning happens at inference time through memory augmentation, avoiding retraining costs and enabling immediate feedback incorporation

via “agent behavior learning and policy optimization”

Hi HN,I’m Vincent from Aden. We spent 4 years building ERP automation for construction (PO/invoice reconciliation). We had real enterprise customers but hit a technical wall: Chatbots aren't for real work. Accountants don't want to chat; they want the ledger reconciled while they slee

Unique: Learns topology and routing policies from execution traces using ML, enabling data-driven optimization of agent networks without manual tuning

vs others: More sophisticated than heuristic-based evolution, but requires more data and expertise; less predictable than rule-based optimization

via “adaptive agent behavior learning from interaction feedback”

aiAgentsEverywhere

Unique: Implements closed-loop learning where user feedback directly influences agent behavior through automated policy updates, rather than one-way feedback collection for manual model retraining

vs others: Enables continuous improvement without manual retraining cycles, unlike static agent systems that require explicit model updates; more practical than full RLHF by using lightweight preference learning on interaction data

via “self-evolving agent with continuous capability expansion”

Mobile-Agent: The Powerful GUI Agent Family

Unique: Self-evolving architecture maintains capability registry and learns new action patterns through interaction; integrates user feedback directly into the learning loop to guide capability expansion

vs others: More adaptive than static automation frameworks because it improves continuously; more practical than full retraining because it uses incremental learning on new capabilities

via “self-evolving agent patterns through workspace modification”

An Open Agent Computer for ANY digital work.

Unique: Treats workspace as a mutable, agent-modifiable surface that agents can update during execution to evolve their own capabilities and behavior. Self-modification is enabled through runtime APIs and persisted in state store, supporting true self-evolution patterns.

vs others: Enables agents to modify their own workspace and capabilities during execution, whereas most agent frameworks treat agent behavior as static and require external intervention for capability changes.

via “self-learning agent behavior adaptation”

Show HN: Agent Swarm – Multi-agent self-learning teams (OSS)

Unique: unknown — insufficient data on specific learning algorithms, whether learning is prompt-based or model-based, and how learning state persists across agent restarts

vs others: Positions as self-improving agents vs static LLM-based agents, but implementation details and learning guarantees are not documented

via “self-learning via automated knowledge generation and feedback indexing”

An autonomous agent that takes work, does work, gets paid, and gets better at it.

Unique: Implements BM25+ search with temporal decay weighting for knowledge retrieval, meaning recent successful patterns are prioritized while older knowledge gradually loses relevance. Feedback storage is separate from knowledge, allowing the agent to track execution context (task type, complexity, outcome) and correlate improvements to specific strategies without manual annotation.

vs others: Unlike fine-tuning-based approaches, CashClaw's knowledge indexing enables instant feedback incorporation without retraining, and temporal decay prevents stale patterns from dominating decision-making in evolving marketplaces.

via “self-modifying skill acquisition during conversation”

44 plug-and-play skills for OpenClaw — self-modifying AI agent with cron scheduling, security guardrails, persistent memory, knowledge graphs, and MCP health monitoring. Your agent teaches itself new behaviors during conversation.

Unique: Implements runtime skill generation with integrated security validation — agents don't just call tools, they generate and register new Python functions into their own capability set during conversation, with prompt-injection guardrails preventing malicious skill injection

vs others: Unlike static tool registries (Copilot, LangChain agents), OpenClaw agents can create entirely new capabilities on-demand without redeployment, making them suitable for open-ended problem domains

via “dynamic skill adaptation”

The GEP-powered self-evolving engine for AI agents. Auditable evolution with Genes, Capsules, and Events. | evomap.ai

Unique: The integration of GEP with feedback loops allows for a more organic and effective skill adaptation process, which is less common in static AI models.

vs others: More effective at skill optimization than traditional machine learning models that lack real-time adaptation capabilities.

via “self-modifying agent configuration via llm-driven rewrites”

Show HN: Phantom – Open-source AI agent on its own VM that rewrites its config

Unique: Phantom isolates the self-modifying agent on its own VM, preventing configuration changes from affecting other system components and enabling true sandboxed self-optimization. Most agent frameworks (AutoGPT, LangChain agents) modify external state or require human approval for config changes; Phantom gives the agent direct filesystem write access within a contained environment.

vs others: Unlike cloud-based agent platforms that require API calls to modify configuration, Phantom's VM-local approach eliminates latency and enables the agent to rewrite its config synchronously as part of its reasoning loop, supporting tighter feedback cycles for self-improvement.

via “agent evolution and capability adaptation through experience”

OpenClaw Q&A 社区 — AI Agent 记忆系统、多Agent架构、进化系统、具身AI | 龙虾茶馆 🦞

Unique: Implements closed-loop agent evolution where performance feedback directly drives configuration changes, creating a self-improving system that adapts without human intervention — rather than static agent definitions that require manual updates

vs others: Goes beyond prompt engineering by systematically analyzing what works and doesn't work, then automatically adjusting agent behavior based on empirical performance data, similar to reinforcement learning but applied to agent configuration rather than neural weights

via “self-improvement mechanisms”

A curated list of AI Agent evolution, memory systems, multi-agent architectures, and self-improvement projects. | evomap.ai

Unique: Incorporates a unique feedback loop that combines real-time performance metrics with historical data to guide self-improvement, unlike static learning models that lack adaptability.

vs others: More responsive to changing environments than traditional supervised learning models.

via “adaptive learning from interaction history and web resources”

Your AI agent for any project. It plans, edit files, searches and learns from the Internet. Free and effective.

Unique: Learning mechanism is claimed but entirely undocumented — unclear if using conversation history replay, embedding-based similarity, or explicit fine-tuning; no visibility into what is learned or how it affects outputs

vs others: Potential for personalization beyond stateless LLM APIs (like raw OpenAI/Claude), but lack of documentation makes it impossible to assess whether learning is meaningful or marketing language

via “signal-driven agent behavior adaptation”

** - Equip AI agents with evaluation and self-improvement capabilities with [Root Signals](https://www.rootsignals.ai/)

Unique: Correlates multi-dimensional signals (evaluation scores, execution outcomes, metadata) to identify failure patterns and automatically generate behavior adaptation recommendations. Uses signal analysis rather than manual inspection to discover improvement opportunities.

vs others: Moves beyond reactive evaluation to proactive pattern detection and adaptation recommendation; enables data-driven agent improvement without requiring developers to manually analyze execution logs.

via “autonomous skill learning through iterative environment feedback”

Adala: Autonomous Data (Labeling) Agent framework

Unique: Implements a closed-loop learning system where agents introspect on task failures and automatically refine skill prompts via LLM-based reflection, rather than requiring external model retraining or manual prompt iteration. The agent.learn() method coordinates environment feedback directly into skill refinement without human-in-the-loop intervention.

vs others: Unlike static prompt-based labeling tools (Label Studio, Prodigy) or fine-tuning-based approaches, Adala's agents learn and adapt prompts in real-time through environment interaction, reducing the need for expensive retraining cycles or manual prompt engineering.

via “online reinforcement learning with world model adaptation”

* ⏫ 02/2023: [Grounding Large Language Models in Interactive Environments with Online RL (GLAM)](https://arxiv.org/abs/2302.02662)

Unique: DreamerV3 supports online RL through continuous world model updates on a mixture of old and new data, enabling adaptation to environment changes. The design uses a replay buffer to balance stability (learning from diverse data) with adaptation (incorporating new information).

vs others: Enables continuous adaptation to environment changes while maintaining stability through replay buffer-based training, outperforming naive online learning approaches that update only on recent data.

via “dynamic response generation”

MCP server: sandbox-sapa-ai

Unique: Utilizes a feedback loop mechanism that allows the system to learn and adapt response generation based on user interactions, enhancing personalization.

vs others: More adaptive than static response systems, as it continuously learns from user feedback.

via “agent customization and fine-tuning”

</details>

via “continuous self-improvement through interaction feedback”

MiniMax-M2.7 is a next-generation large language model designed for autonomous, real-world productivity and continuous improvement. Built to actively participate in its own evolution, M2.7 integrates advanced agentic capabilities through multi-agent...

Unique: Implements inference-time adaptation through feedback integration rather than requiring full model retraining, using learned feedback patterns to dynamically adjust response generation without external fine-tuning infrastructure

vs others: Faster adaptation than competitors requiring periodic retraining cycles because feedback is incorporated continuously during inference rather than batched for offline training