Stop Building AI Agents Like Coding Agents: A 6-Step Framework for Knowledge Work
MaryamMiradi · x · 2026-09-19
Researcher Maryam Miradi argues most AI agent architectures today copy coding agents, but code is a special kind of knowledge with filenames, functions and explicit dependencies — legal, medical and financial knowledge work starts from open-ended intent instead. She proposes designing knowledge agents differently, starting with: decompose the intent (turn open requests into explicit research questions, identify entities, constraints, jurisdictions and missing facts rather than blindly searching the prompt) and plan the search (decide which evidence matters before retrieving). The core claim: knowledge agents need intent decomposition and evidence planning, not code-retrieval-style workflows.
Related event: Knowledge Agents Need Their Own Architecture, Researcher Argues(2 posts)→
More from coding & agent
- Devin secures more compute, makes SWE-2 free for Pro, Max & Teams until Oct 8 — DevinAI · 2026-09-19
- "RIP CLAUDE.md": developer signals a shift away from the Claude Code memory file — kieranklaassen · 2026-09-19
- WetRobo: coding agents as the interface between foundation models and lab instruments — sherryyangML · 2026-09-19
- WetRobo ships robot + demos + AGENTS.md, letting a coding agent adapt tools on site — sherryyangML · 2026-09-19
- Agent Portability Gets Real: Centralizing Skills and Keys Across Agent Platforms — ivanhzhao · 2026-09-19
- Hamel Husain: Data scientists have never had more alpha than right now — HamelHusain · 2026-09-19