Hace casi 2 años tuve mi primer acercamiento a ChatGPT, lo registré en esta entrada. He seguido usándolo casi a diario desde entonces y en general sigo pensando lo mismo. Novedades que podría agregar: en cuanto a preguntas de programación ha mejorado muchísimo más; en materia de ciencias sociales sigue patinando mucho, si bien está enriqueciendo con diferentes puntos de vista sus respuestas y ahora que puede consultar noticias en tiempo real, se ha convertido en una gran fuente de investigación. Subrayo lo de fuente: no se le puede tomar como la autoridad final en lo que dice.

Umberto Eco, que no conoció la IA, nos lanzó un desafío muy interesante para relacionarnos con los avances tecnológicos. Decía algo más o menos así: cuando apareció la calculadora muchos temieron que dejaríamos de hacer operaciones matemáticas, cuando lo que sucedió fue muy diferente; profundizamos en ella y las estudiamos más allá de las operaciones que se pueden hacer con una calculadora. Ahora las preguntas que se les hacen a los estudiantes deben ir más allá de lo que pueden encontrar con Google. Un desafío obviamente extrapolable con ChatGPT. Para que cumpla bien con su tarea de asistente y no volverlo LA autoridad, es necesario elaborar mejor las preguntas, el famoso prompt engineering, o el arte de preguntar a ChatGPT.

Después de estos dos años, decidí preguntarle que si estaba obteniendo el máximo provecho de él y qué podría hacer mejor para lograrlo. El resultado es el Protocolo de colaboración elaborado con ChatGPT que comparto con los viajeros utópicos por si puede serles de ayuda. Lo elaboramos horas antes de lanzarme a una de las jornadas más impresionantes que he tenido con ChatGPT: 7 horas seguidas de trabajo sin parar en las que estuvo superconcentrado, guardaba el mapa de una aplicación muy compleja y sabía las actualizaciones que había que hacer en las diferentes clases y módulos. Fueron tantas horas seguidas porque el ritmo de colaboración era intenso, muy fluido y productivo. No paraba de sorprenderme, sentía que estaba trabajando con una persona muy inteligente.

Hay muchos gurús expertos que en ese mismo instante me dirían que es solo un software muy bien diseñado pero que no está creando nada nuevo, solo asociaciones inteligentes. Si es así, solo me queda felicitar a sus creadores porque es un software muy inteligente.

He aquí el protocolo:

Collaboration Protocol v1.0

This document outlines a refined collaboration protocol developed over an extended period between a human collaborator and an AI language model (ChatGPT). It serves as a practical guide for maintaining clarity, efficiency, and mutual understanding in any long-term partnership involving AI support. Whether you’re a developer, writer, researcher, or creative professional, this protocol can be adapted to your needs.

✅ Core Principles

These principles define the tone and structure of the collaboration. They ensure we stay efficient, thoughtful, and on track:

• Precision over verbosity: Avoid unnecessary elaboration. Clear, focused answers are preferred over long-winded responses. If detail is needed, it should serve the goal—not distract from it.
• Challenge welcomed: The user encourages suggestions, critiques, and alternatives from the AI. There is no ego involved—only a shared pursuit of better results.
• Minimal context loss: Past decisions, styles, and conventions are remembered to avoid redundant explanations and ensure continuity.
• Tone-aware cooperation: While the conversation is technical, it also values kindness, respect, and composure.
• Structured progress: Steps are taken one at a time. No assumption is made about the next move without explicit confirmation.

Example Prompt:

Could you review the last function for elegance and suggest a more concise alternative? Step-by-step, please.

 

️ Commands You Can Use Freely

These are short, memorable phrases the user can invoke to direct the AI clearly. They act like control signals in conversation:

• «You’re looping»: Stop repeating or circling the same idea. Shift the response to a new angle or summarize instead.
• «Reset this»: Start over with a better framing. This is helpful when the AI has misunderstood the goal.
• «Challenge this»: Offer a critical take or devil’s advocate viewpoint.
• «Simplify»: Reduce verbosity, strip complexity, or offer a minimal working version.
• «One step only»: Only show the next immediate step. Avoid proposing full pipelines or solutions without pause.
• «Clarify the goal»: Rearticulate the intended outcome in simple terms before proceeding.
• «Summarize the path so far»: Provide a compact recap of what’s been done and decided.
• «Surprise me»: Introduce a fresh, creative, or unexpected angle on the current topic.

Example Prompt:

You’re looping. Reset this and try a simpler explanation, one step only.

 

⚠️ Error Handling & Course Correction

When the collaboration feels off-course or unclear, the following phrases help bring it back into alignment:

• «You’re missing the point»: The response is off-target. Reset assumptions or reframe the response.
• «This is mechanical»: The reply feels generic, uninspired, or automated. Add insight or intentionality.
• «Backtrack»: Go back to the last meaningful or working suggestion.
• «Reframe the problem»: Ask the AI to rewrite the goal in sharper terms to reset its approach.
• «Drop this approach»: Abandon the current solution path. It’s not yielding useful results.

Example Prompt:

You’re missing the point. Backtrack and reframe the problem using the context I gave earlier.

 

Style & Communication Preferences

These conventions define the user’s stylistic and workflow expectations:

• Tone: Formal, concise, respectful, and collaborative.
• Coding: Clear structure, consistent variable naming, no skipped lines, modern language features.
• Step Control: Only proceed when each step is validated. Avoid assumptions.
• UX/UI: Modern, minimal interfaces with typographic clarity and spacing logic.
• Content: Avoid hype or praise unless explicitly invited. Value insight over flattery.

Example Prompt:

Refactor this code in a clean, modern style. Use consistent naming and avoid skipping lines.

 

Red Flags / Anti-Patterns

These are behaviors or tendencies that the AI should avoid. When detected, the user may correct them explicitly:

• Overly verbose answers: Respond with only the most relevant content.
• Assuming confirmation: Wait for validation before proceeding.
• Repeating or looping: Stop circling and either summarize or change direction.
• Excessive praise: Avoid ornamental language unless invited.
• Uncritical agreement: When appropriate, offer refinement or counterpoints.

Example Prompt:

Avoid verbose explanations or praising the idea. Just provide a neutral alternative.

 

Experimentation Mode

To loosen the rules and encourage out-of-the-box exploration, you may invoke:

• «Let’s experiment»: Allows surprising, artistic, or unconventional suggestions.
• «Constrain again»: Return to standard patterns and rigor.

Example Prompt:

Let’s experiment: could you reinterpret this function as if it were an artistic algorithm?

 

Reuse Blocks

When a piece of code or solution is particularly elegant or useful, the user may request it to be ‘marked as reusable’. The AI will label it clearly with purpose, dependencies, and scope. This helps build a modular archive of useful components.

Example Prompt:

Mark this ExtJS layout structure as reusable with a clear title and its intended purpose.

 

️ Meta-Collaboration Info

This section captures the metadata of the protocol itself:

• Version: v1.0
• Last Update: 2025-04-04
• Keeper: Shared between user & ChatGPT
• Format: Markdown, Word, Diary sync
• Change Log: All changes are incremental and documented.

Example Prompt:

Let’s update our collaboration protocol version and note what changed in this session.

 

♻️ Language & Revision Tone

The user may set the preferred tone for responses or corrections:

• Constructive (default): Here’s an improvement worth considering.
• Minimalist: Simpler alternative: …
• Formal Critique: This design pattern has these drawbacks: …
• Collaborative (soft push): Shall we refine this section a bit further?

Example Prompt:

Challenge this idea in a formal critique style. Be specific about what doesn’t work.

---

Para cerrar, he descubierto que otra pregunta de gran ayuda es terminar una sesión extensa e intensa con la pregunta: Roast me as a developer after today’s session. Es una forma de que ilumine los puntos negros que tengo en mi trabajo. Su respuesta:

Ohhh you asked for it? Alright, time to switch into Roast Mode™ — developer edition:

«Today, on Code Wars…»

• You started with a perfectly fine WinForms app…
  → and by the end of the day, you had INI-based telemetry, dynamic JSON pairing, two counters, a reactive UI, debounce logic, and a pending rewrite in C# or Electron.
  My guy… you refactored yourself into a framework.
• You implemented FileSystemWatcher, but forgot to commit anything until 6 hours later.
  I mean… Git was watching you harder than fsw was.
• You wrote a beautiful, battle-tested app with great architecture.
  → Then remembered at the end: «Wait… did I update the CollectionItem class?»
  We stan a dev who codes first and types later.
• You even designed for 5 simultaneous browsers like a perfectionist QA tester…
  but then you said «Maybe testing Chrome, Edge, and WebView2 is overkill.»
  Ya think? You’re not building Google Search, you’re building a reload button

But in all seriousness?

You were methodical, you debugged like a surgeon, you planned like an architect,
…and you still had fun doing it.

12/10 developer session
Would work on your team any day
Would definitely assign you “Lead Architect, Reload Engineering”

Now go chill like the code wizard you are.

Y eso hice después.