Triangulation is expensive in attention, and attention is the scarce resource in this method. Running three platforms on a question that needed one is not rigor — it is ceremony, and it burns the focus you'll want when a question genuinely warrants the session.
Before you open anything, ask one question: what does this task most require — breadth, depth, or grounding?
Most real tasks need more than one. But they usually have a dominant need, and naming it decides both who you call and whether you need everyone.
When the question is ambiguous or poorly defined and needs to be made tractable before anything else can happen. When you need a structured document that will serve as the foundation for the rest of the work. When the stakes involve ethical or long-term dimensions. When you have raw material — notes, transcripts, a pile of sources — that needs organizing into an argument.
When you need a first draft on the page. When you want variations on a message or an approach. When you're in early-stage exploration and the goal is to widen the field rather than narrow it. When you need the same concept explained at three different levels of complexity.
When you suspect you're stuck in a frame and can't see its edges. When you have a tentative plan and want it stress-tested before you commit. When the other platforms have been agreeable and you don't trust that agreement. When you want the unconventional angle that a more cautious system won't volunteer.
When the answer has to be connected to sources. When facts need checking before anything is built on them. When the question touches a domain where training-data limits could produce plausible but outdated information — pricing, regulations, current status, anything that moves.
Brief the full ensemble when at least one of these is true:
You will often start with one platform and discover mid-session that you needed three. Escalate when a single response surfaces complexity you didn't anticipate, when it contradicts an assumption you'd been treating as settled, or when the stakes changed while you were working.
Escalating is not an admission that you routed badly. It is the routing decision working — you learned something that changed the shape of the question.
CONTEXT: [Brief description of the plan, decision, or recommendation to be tested.] TASK: I have a tentative direction on the above. Before I commit, I want you to stress-test it. Please: · Identify the weakest assumptions in my reasoning · Describe what could go wrong, and how I would know early · Surface any angles I have not considered · Tell me what you would need to see to change your assessment Do not soften this. I need the friction, not reassurance. If the plan is sound, say so — but only after you have genuinely tried to break it.
A weak answer does not always mean you asked badly. Before you rewrite the prompt for the fourth time, run the diagnosis: is the problem the prompt, the platform, the task, or the routing choice?
Sometimes the task is simply pressing against the platform's native architecture. Prompting can improve your access to a system's strengths; it cannot manufacture strengths the system doesn't have. Architecture sets the reachable range; prompting determines how much of that range you can actually use. If the limitation is architectural, stop polishing and change the instrument.
And be careful with reputation. A platform's standing in your memory reflects the version you used months ago, filtered through whatever you were doing at the time. Capability moves faster than reputation does, in both directions. Audition the platform in the task, not in your memory of its brand.
Good conductors know when not to conduct. Once the question is framed well, a single platform can carry a low-stakes task without full ensemble review. Stepping back is not carelessness — it is part of the discipline.