# Every Harness Will Become A Claw — Sam Bhagwat, Mastra

## Executive summary

The evolution of AI agents is moving from localized 'Harnesses'—tools used for coding and task execution—to persistent, always-on services called 'Claws.' This transition involves imbuing agents with initiative, external connectivity (like a heartbeat), and continual learning capabilities. The speaker proposes Steinberger's law: every harness will expand until it becomes a Claw, driven by the desire for powerful, integrated developer experiences.

## Key takeaways

- Harnesses are evolving into Claws: The next generation of agents moves beyond local execution to become always-on services that listen to external events (e.g., Slack, mobile apps) and maintain a persistent 'heartbeat.'
- Agentic Spectrum Advancement: Agents are advancing through stages: Agent $ ightarrow$ Harness $ ightarrow$ Claw. Key technical advancements include durability, doggedness, planning mode, and parallel subagents.
- Cloud vs. Local Architecture: The shift from local harnesses to cloud harnesses provides greater parallelism and resources but necessitates a different distributed system architecture.

## Technical details

- Agent Definition & Loop: An agent differs from an LLM via the 'agent loop,' which includes tool calls, memory management, the ability to retry failed tasks, and context engineering. (7:35)
- Harness Capabilities: Advanced harnesses feature planning mode, parallel subagents for concurrent task execution, skills for dynamic agent creation, background bash tasks, and session-long tool approval. (8:02)
- Cloud Harness Architecture: Moving to cloud environments allows for more parallelism and resources than local machines but requires managing a distributed system architecture. Code output shifts from local worktrees to PRs pushed to GitHub. (10:35)
- Claw Functionality: A Claw is characterized by initiative, persistent memory in accessible locations, and the ability to perform continual learning based on generated traces. (12:48)

## Practical implications

- When building agents, ensure they possess the capabilities required by users, as the field is changing rapidly.
- Be aware of potential market 'shakeouts' in the future, requiring continuous adaptation and staying current with technological shifts.
- Focus on developing systems that can maintain persistence and initiative to meet the demands of a 'Claw'-level experience.

## Topics

AI Agents, Harnesses, Claws, Context Engineering, Agent Loop, Distributed Systems, Mastra, Principles of Building AI agents, GitHub

Source: https://www.youtube.com/watch?v=8qWIPUia2O8