Use strong models for ambiguity
Planning, architecture, debugging unknowns, product decisions, risk review, and final quality checks.
Model choices
Use the strong model to decide. Use the faster model to execute.
Model choice is part of the workshop skill. Premium models are best when the problem is fuzzy. Faster or cheaper models are better once the task is specific.
The Rule
Match the model to the job.
Use faster models for bounded work
Formatting, filling templates, rewriting, small code edits, checklist execution, and repetitive fixes after the plan is clear.
Switch after the plan
Do not keep a premium model doing simple execution if the steps are already specific.
Workshop Picks
Which tool to use and why.
Claude / Claude Code
Best for long-context coding, repo understanding, careful writing, and guided builds. Use the best available model for planning, then switch to a faster execution model for narrow changes.
ChatGPT / Codex
Best for goal-based build loops, image generation, general reasoning, and working inside the Codex app. Use /plan for unclear work and /goal for one clear objective.
DeepSeek V3.2 / Speciale
Good for cost-effective reasoning and coding when available. Treat Speciale as a deep-reasoning option, not the default workshop tool: provider access and tool-calling support can vary. Use it for plans and hard debugging, then move bounded tasks to a faster option if your provider offers one. If API access is provided for your pod, treat it as workshop-only and never post the key in public chat or a repo.
Gemini / NotebookLM
Use Gemini for web-connected research and Google ecosystem work. Use NotebookLM to turn a repo or notes into study material, podcast-style summaries, or presentation prep.
Model Switch Checklist
Switch only when the work is bounded.
Stay on the strongest model
The problem is unclear, the file structure is unknown, the pod disagrees, or the change could break the demo.
Switch to a faster model
The plan is approved, the next step is specific, and the output can be checked quickly by a human or a test.
Switch back up
The model repeats mistakes, invents facts, loses context, or cannot explain why the work failed.
Plain Explanation
What an LLM is in workshop terms.
An LLM is a reasoning and drafting engine. It predicts useful next text from your instructions and context. It can help plan, write, code, debug, and explain. It is not automatically correct. Good workshop work means giving it enough context, asking it to plan before acting, checking the result, and keeping the task small enough to finish.
Do not give any model private tokens, passwords, customer data, or personal information during the workshop. Use fake examples unless a facilitator says the source is approved.