OpenAI found that removing repeated instructions from old prompts improved scores by 10-15% while cutting token usage by 66%. The complex rule lists built for older models now confuse systems like GPT-5.6, leading to worse and more expensive answers.
Instead of using vague adjectives like "high quality," give a model a concrete, checkable goal (e.g., "a stranger can't tell our render from the real photo"). Then, use a loop command to force the model to iterate and self-correct until it meets that high bar.
Intuit PM Christine Zhu argues the biggest productivity unlock comes from using models like Fable 5 for high-leverage "impact work," like product strategy, not just clearing a backlog of small tasks. Treat the AI as a sparring partner for your hardest problems.
Tariq from Cloud Code suggests the quality of work with models like Fable 5 is bottlenecked by the user's ability to clarify unknowns. Instead of perfect upfront planning, use the AI iteratively to surface your own blind spots, assumptions, and unstated requirements.
New models like GPT-5.6 Sol are more "tenacious" and will take more initiative. Users must set explicit boundaries (e.g., "don't send, just draft") to prevent them from taking unwanted actions or wasting resources on irrelevant tasks, which has real-world consequences.
Because new AI models thrive on context, rambling via voice dictation can be more effective than carefully crafted written prompts. This unstructured stream of consciousness provides richer background information that the AI can use to generate better results, making "ramblers" more effective prompters.