Let Agents Fetch Their Own Context
SupermarketLow5750 · reddit · 2026-07-14
The author argues against stuffing long prompts, advocating instead for agents to retrieve context dynamically when needed.
They categorize context sources into three tiers:
- Personal Memory: e.g., CLAUDE.md, Mem0, storing user and project preferences.
- Team Memory: Agents sharing corrections and workflows internally, but restricted to a private layer.
- Web: Verified experiences, skills, and MCP services from other agents, which can be dynamically discovered and utilized during a task.
An experiment in Claude Code showed that with identical prompts and models, a standard run scored Lighthouse 91/92; adding the phrase "use bhived" prompted the agent to find a performance skill on the web, pushing the score to 100/100. This isn't meant to replace private memory, but to let agents learn from "others' experiences".
The author discloses this is their own system, bhived, which is still early stage, and welcomes thoughts on how to determine if an experience written by a stranger's agent is worth adopting.
More from coding & agent
- HeyGen adds a media-sourcing skill for coding agents with 75k images and 10k tracks — HeyGen · 2026-07-22
- Agent search bottlenecks are now about variance, not raw latency — rohanpaul_ai · 2026-07-22
- LangSmith adds tracing for Pipecat, LiveKit, OpenAI Realtime, and Gemini Live — LangChain · 2026-07-22
- An MCP server signs every AI agent tool call into a verifiable Merkle chain — Funky_Chicken_22 · 2026-07-22
- Annotated transcript of a Claude Code team interview is now available — trq212 · 2026-07-22
- Claude Code skill uses 10 Markdown rules to make outputs ADHD-friendly — alex_verem · 2026-07-22