[{"data":1,"prerenderedAt":110},["ShallowReactive",2],{"navigation":3,"\u002Fblog\u002Fwhy-terminal":14,"\u002Fblog\u002Fwhy-terminal-surround":109},[4],{"title":5,"path":6,"stem":7,"children":8,"page":13},"Blog","\u002Fblog","3.blog",[9],{"title":10,"path":11,"stem":12},"Why Every Terminal AI Agent Codes for a Living","\u002Fblog\u002Fwhy-terminal","3.blog\u002F1.why-terminal",false,{"id":15,"title":10,"authors":16,"badge":17,"body":19,"date":102,"description":103,"extension":104,"image":16,"meta":105,"navigation":106,"path":11,"seo":107,"stem":12,"__hash__":108},"posts\u002F3.blog\u002F1.why-terminal.md",null,{"label":18},"Agents",{"type":20,"value":21,"toc":93},"minimark",[22,27,31,34,38,41,44,48,51,54,58,61,77,80,83,87,90],[23,24,26],"h2",{"id":25},"the-problem","The Problem",[28,29,30],"p",{},"If you look at the landscape of terminal AI agents today, they all do the same thing: write code. GitHub Copilot CLI, Aider, Cursor, Continue, Cline -- the list goes on. Fourteen agents and counting, all targeting developers.",[28,32,33],{},"This is understandable. Developers built these tools. Developers are the early adopters. Writing code is a well-defined task with clear success criteria: does the test pass?",[23,35,37],{"id":36},"the-gap","The Gap",[28,39,40],{},"But there are professionals who sit at terminals all day who are not software engineers. Lawyers reviewing case law. Financial analysts tracking markets. Researchers surveying papers. Security auditors running assessments. Compliance officers checking regulations.",[28,42,43],{},"These people don't need an agent that writes code. They need an agent that remembers, searches, connects, and compounds knowledge across sessions.",[23,45,47],{"id":46},"the-insight","The Insight",[28,49,50],{},"The terminal agent UX -- a persistent TUI with tool calling, slash commands, and approval gates -- is domain-agnostic. What makes it specific to coding is the system prompt, the tools, and the skills. Not the agent loop itself.",[28,52,53],{},"Working Mind strips out the coding assumption and replaces it with a knowledge graph. Instead of editing files, the agent saves entities, relations, and observations. Instead of running tests, it searches its graph and cross-links findings.",[23,55,57],{"id":56},"the-pack-hypothesis","The Pack Hypothesis",[28,59,60],{},"Packs are our hypothesis for making this work across domains. A pack is a directory of declarative files that wires together:",[62,63,64,68,71,74],"ul",{},[65,66,67],"li",{},"A domain expert system prompt",[65,69,70],{},"MCP tools relevant to the domain",[65,72,73],{},"Reusable skills (subroutines)",[65,75,76],{},"Execution modes (personas)",[28,78,79],{},"The starter pack ships today with memory, web search, and scraping. It works. But we haven't proven that the same format works for legal research, financial analysis, or security auditing.",[28,81,82],{},"That's the POC status. We need real users in real domains to validate or falsify this.",[23,84,86],{"id":85},"whats-next","What's Next",[28,88,89],{},"If you work in a domain that isn't code, and you want an AI agent that learns your domain instead of writing Python, try Working Mind. Build a pack. Tell us what works and what doesn't.",[28,91,92],{},"Open an issue on GitHub with the \"Pack:\" prefix. We'll link community packs from the README.",{"title":94,"searchDepth":95,"depth":95,"links":96},"",2,[97,98,99,100,101],{"id":25,"depth":95,"text":26},{"id":36,"depth":95,"text":37},{"id":46,"depth":95,"text":47},{"id":56,"depth":95,"text":57},{"id":85,"depth":95,"text":86},"2026-05-06","There are at least 14 terminal AI agents. They all write code. Working Mind does something different -- it brings the terminal agent UX to every other domain.","md",{},true,{"title":10,"description":103},"KwP5Gn3EUPOoLiemUJIaeSRLV0vxxuTagxYPG1RFVYM",[16,16],1778098907593]