{
 "slug": "lenses-roles-and-loops",
 "title": "Lenses, roles and loops",
 "subtitle": "How LLMs and agents work: a prompt as a lens, a role as a model with a description, an agent as a loop, tools as requests, no memory, compression, and focused threads. 13 panels.",
 "status": "draft",
 "book_version": "v1.0.0 (the second edition does not exist yet)",
 "by": "gen-z.journalist, 18 October 2026, from Dinis's memo on LLMs and agents",
 "quote_source": "admin/content/articles/14__lenses-roles-and-loops.md",
 "rule": "Every line the Guide says, and every quote card, is from the article 'Lenses, roles and loops', word for word; the build checks it. The five lenses on recursion, the weather question and the named threads are the storyboard's illustrations.",
 "scenes": [
  {
   "kind": "cover",
   "title": "Lenses, roles and loops",
   "subtitle": "It stops being magic when you see how it works.",
   "cast": [
    {
     "who": "zara",
     "pose": "think",
     "mood": "thinking"
    },
    {
     "who": "agent",
     "pose": "wave",
     "mood": "happy"
    },
    {
     "who": "priya",
     "pose": "stand",
     "mood": "happy"
    },
    {
     "who": "guide",
     "pose": "wave",
     "mood": "happy"
    }
   ],
   "note": "The second half of the generative AI chapter: what happens when you use a model.",
   "source": "Lenses, roles and loops"
  },
  {
   "kind": "spectrum",
   "setting": "none",
   "caption": "One idea, recursion, through five lenses",
   "stops": [
    {
     "label": "Teenager",
     "note": "a meme of a meme",
     "colour": "#fde68a"
    },
    {
     "label": "Musician",
     "note": "a round, sung over itself",
     "colour": "#fed7aa"
    },
    {
     "label": "Doctor",
     "note": "cells dividing into cells",
     "colour": "#fecaca"
    },
    {
     "label": "Football fan",
     "note": "the replay of the replay",
     "colour": "#bbf7d0"
    },
    {
     "label": "Portuguese",
     "note": "recursividade",
     "colour": "#bfdbfe"
    }
   ],
   "cast": [
    {
     "who": "guide",
     "x": 0.5,
     "pose": "point",
     "mood": "happy"
    }
   ],
   "note": "To a model, translating into another language is no different from translating for an audience: the prompt is the lens. The five analogies are the storyboard's, not the memo's.",
   "source": "Lenses, roles and loops, A prompt is a lens"
  },
  {
   "kind": "scene",
   "setting": "none",
   "caption": "The wrong lens",
   "cast": [
    {
     "who": "priya",
     "x": 0.24,
     "pose": "shrug",
     "mood": "worried",
     "says": "I've always been bad at maths."
    },
    {
     "who": "guide",
     "x": 0.76,
     "pose": "point",
     "mood": "happy",
     "flip": true,
     "says": "The analogy is the reason someone understands something.",
     "max": 22
    }
   ],
   "note": "Priya ran rotas, budgets and stock levels for years: maths all day, just not the kind school called maths.",
   "source": "Lenses, roles and loops, A prompt is a lens"
  },
  {
   "kind": "graph",
   "caption": "A role is a model with a description",
   "graph": {
    "nodes": [
     {
      "id": "model",
      "label": "a raw model",
      "x": 0.12,
      "y": 0.15
     },
     {
      "id": "role",
      "label": "role.md: who you are, what you refuse",
      "x": 0.12,
      "y": 0.5,
      "colour": "#fde68a"
     },
     {
      "id": "question",
      "label": "a question",
      "x": 0.12,
      "y": 0.85
     },
     {
      "id": "editor",
      "label": "the editor",
      "x": 0.72,
      "y": 0.2,
      "kind": "person",
      "colour": "#ddd6fe"
     },
     {
      "id": "publisher",
      "label": "the publisher",
      "x": 0.88,
      "y": 0.5,
      "kind": "person",
      "colour": "#ddd6fe"
     },
     {
      "id": "journalist",
      "label": "the journalist",
      "x": 0.72,
      "y": 0.82,
      "kind": "person",
      "colour": "#ddd6fe"
     }
    ],
    "edges": [
     {
      "from": "model",
      "to": "editor"
     },
     {
      "from": "role",
      "to": "editor",
      "label": "the lens"
     },
     {
      "from": "role",
      "to": "publisher",
      "label": "a different file"
     },
     {
      "from": "role",
      "to": "journalist"
     },
     {
      "from": "question",
      "to": "journalist",
      "label": "asks"
     }
    ]
   },
   "note": "The same model behind every role on this site; what changes is the description and the question.",
   "source": "Lenses, roles and loops, A role is a model with a description"
  },
  {
   "kind": "scene",
   "setting": "none",
   "caption": "An agent is a loop with a goal",
   "cast": [
    {
     "who": "agent",
     "x": 0.22,
     "pose": "think",
     "mood": "thinking"
    },
    {
     "who": "guide",
     "x": 0.76,
     "pose": "point",
     "mood": "happy",
     "flip": true,
     "says": "Fundamentally the agent is simply a task: do this; then, are we done yet, or do we need to continue?",
     "max": 24
    }
   ],
   "note": "And rinse and repeat. The loop is the skeleton; good agents add planning, checks and limits.",
   "source": "Lenses, roles and loops, An agent is a loop with a goal"
  },
  {
   "kind": "graph",
   "caption": "The loop, with one tool call",
   "graph": {
    "nodes": [
     {
      "id": "task",
      "label": "the task",
      "x": 0.06,
      "y": 0.1,
      "kind": "person",
      "colour": "#ddd6fe"
     },
     {
      "id": "model",
      "label": "the model",
      "x": 0.4,
      "y": 0.1,
      "colour": "#fde68a"
     },
     {
      "id": "call",
      "label": "a tool call: search the web",
      "x": 0.85,
      "y": 0.1
     },
     {
      "id": "harness",
      "label": "the harness runs it",
      "x": 0.85,
      "y": 0.55
     },
     {
      "id": "result",
      "label": "the result, added",
      "x": 0.5,
      "y": 0.52
     },
     {
      "id": "done",
      "label": "are we done?",
      "x": 0.2,
      "y": 0.9,
      "kind": "circle",
      "colour": "#bbf7d0"
     },
     {
      "id": "answer",
      "label": "the answer",
      "x": 0.7,
      "y": 0.92
     }
    ],
    "edges": [
     {
      "from": "task",
      "to": "model",
      "label": "sent"
     },
     {
      "from": "model",
      "to": "call",
      "label": "returns"
     },
     {
      "from": "call",
      "to": "harness"
     },
     {
      "from": "harness",
      "to": "result"
     },
     {
      "from": "result",
      "to": "model",
      "label": "next request"
     },
     {
      "from": "result",
      "to": "done"
     },
     {
      "from": "done",
      "to": "answer",
      "label": "yes"
     }
    ],
    "path": [
     "task",
     "model",
     "call",
     "harness",
     "result",
     "done",
     "answer"
    ]
   },
   "note": "The model never runs a tool: it writes a request; the program around it decides, runs it and sends the result back. If the answer to 'are we done?' is no, round again.",
   "source": "Lenses, roles and loops, Tools changed everything; Precisely"
  },
  {
   "kind": "spectrum",
   "setting": "none",
   "caption": "What the next request carries",
   "stops": [
    {
     "label": "Your question",
     "note": "is it raining in Lisbon?",
     "colour": "#e5e7eb"
    },
    {
     "label": "Model asks",
     "note": "call the weather service",
     "colour": "#fde68a"
    },
    {
     "label": "Harness",
     "note": "calls it: 'light rain'",
     "colour": "#fed7aa"
    },
    {
     "label": "All of it, resent",
     "note": "question + call + result",
     "colour": "#bfdbfe"
    },
    {
     "label": "The answer",
     "note": "take an umbrella",
     "colour": "#bbf7d0"
    }
   ],
   "cast": [
    {
     "who": "zara",
     "x": 0.5,
     "pose": "point",
     "mood": "determined"
    }
   ],
   "note": "Two requests for one answer; the second holds everything the first did, plus the tool's result. (A made-up example.)",
   "source": "Lenses, roles and loops, Tools changed everything"
  },
  {
   "kind": "spectrum",
   "setting": "none",
   "caption": "From training only, to tools",
   "stops": [
    {
     "label": "Nov 2022",
     "note": "ChatGPT, no tools",
     "colour": "#e5e7eb"
    },
    {
     "label": "Mar 2023",
     "note": "plugins",
     "colour": "#fde68a"
    },
    {
     "label": "Mid 2023",
     "note": "functions for developers",
     "colour": "#fed7aa"
    },
    {
     "label": "Nov 2024",
     "note": "Model Context Protocol",
     "colour": "#bfdbfe"
    },
    {
     "label": "2026",
     "note": "the loop, many tools, hours",
     "colour": "#bbf7d0"
    }
   ],
   "cast": [
    {
     "who": "guide",
     "x": 0.5,
     "pose": "stand",
     "mood": "happy"
    }
   ],
   "note": "Training is a compression of the world, frozen on the day it ended; tools let the model ask for what it does not hold.",
   "source": "Lenses, roles and loops, Tools changed everything"
  },
  {
   "kind": "scene",
   "setting": "night",
   "caption": "It remembers nothing",
   "cast": [
    {
     "who": "zara",
     "x": 0.24,
     "pose": "think",
     "mood": "surprised",
     "says": "So it doesn't remember what I said an hour ago?"
    },
    {
     "who": "guide",
     "x": 0.76,
     "pose": "stand",
     "mood": "happy",
     "flip": true,
     "says": "There is no memory."
    }
   ],
   "note": "Every request carries the whole conversation; that is why there is a context window, and why long chats get slower.",
   "source": "Lenses, roles and loops, It remembers nothing"
  },
  {
   "kind": "evidence",
   "setting": "none",
   "caption": "Long chats are compressed",
   "cast": [
    {
     "who": "agent",
     "x": 0.2,
     "pose": "think",
     "mood": "thinking"
    }
   ],
   "evidence": {
    "title": "What one summary keeps",
    "bars": [
     {
      "label": "the chat",
      "value": 100,
      "shown": "100%"
     },
     {
      "label": "its summary",
      "value": 1.7,
      "shown": "1.7%"
     }
    ]
   },
   "note": "One session, one month: summarised eighteen times, each summary keeping on average 1.7% of what it replaced (re-anchoring, newsroom). What the summary leaves out is not forgotten; it was never given back.",
   "source": "Lenses, roles and loops, Long chats are compressed"
  },
  {
   "kind": "evidence",
   "setting": "none",
   "caption": "The newsroom's answer",
   "cast": [
    {
     "who": "guide",
     "x": 0.2,
     "pose": "point",
     "mood": "happy"
    }
   ],
   "evidence": {
    "quote": "The rules are restored, not remembered.",
    "source": "Re-anchoring agent behaviour policies (newsroom)"
   },
   "note": "The rules live in a file printed back after every summary.",
   "source": "Lenses, roles and loops, Long chats are compressed"
  },
  {
   "kind": "graph",
   "caption": "Many focused threads, each with a name",
   "graph": {
    "nodes": [
     {
      "id": "dinis",
      "label": "a person",
      "x": 0.08,
      "y": 0.12,
      "kind": "person",
      "colour": "#ddd6fe"
     },
     {
      "id": "inbox",
      "label": "inbox.triage",
      "x": 0.1,
      "y": 0.9
     },
     {
      "id": "drafts",
      "label": "inbox.drafts",
      "x": 0.37,
      "y": 0.9
     },
     {
      "id": "editor",
      "label": "gen-z.editor",
      "x": 0.63,
      "y": 0.9
     },
     {
      "id": "journalist",
      "label": "gen-z.journalist",
      "x": 0.9,
      "y": 0.9
     },
     {
      "id": "files",
      "label": "the files",
      "x": 0.5,
      "y": 0.42,
      "colour": "#fde68a"
     }
    ],
    "edges": [
     {
      "from": "inbox",
      "to": "files",
      "label": "writes"
     },
     {
      "from": "drafts",
      "to": "files",
      "label": "reads"
     },
     {
      "from": "editor",
      "to": "files"
     },
     {
      "from": "journalist",
      "to": "files",
      "label": "briefs"
     },
     {
      "from": "files",
      "to": "dinis",
      "label": "drafts to send"
     }
    ]
   },
   "note": "One topic per session keeps what each model has to hold small; the sessions talk through files, never through memory. The thread names are illustrative.",
   "source": "Lenses, roles and loops, Many focused threads"
  },
  {
   "kind": "evidence",
   "setting": "none",
   "caption": "Where the memory is",
   "cast": [
    {
     "who": "tomas",
     "x": 0.2,
     "pose": "think",
     "mood": "thinking"
    }
   ],
   "evidence": {
    "quote": "Nothing is remembered by the model between sessions. Everything is remembered by the files.",
    "source": "Memory is not a spectator sport (newsroom)"
   },
   "note": "The principle the threads rely on.",
   "source": "Lenses, roles and loops, Long chats are compressed"
  },
  {
   "kind": "punchline",
   "setting": "none",
   "punchline": "Not magic: a lens, a role, a loop.",
   "cast": [
    {
     "who": "zara",
     "x": 0.12,
     "pose": "arms-up",
     "mood": "happy"
    },
    {
     "who": "priya",
     "x": 0.27,
     "pose": "stand",
     "mood": "happy"
    },
    {
     "who": "agent",
     "x": 0.42,
     "pose": "wave",
     "mood": "happy"
    }
   ],
   "note": "Understand it, and you are the one who sets the constraints.",
   "source": "Lenses, roles and loops, Why this matters"
  }
 ]
}