Steer Complex AI Reasoners Mid-Response to Save Iteration Time
Don't wait—steer the 'Thinking' models while they process your request.
Use the steering features in GPT-5.4/5.5 Thinking models to provide feedback on the model's 'upfront plan' before the final response is generated.
The Scenario
You are asking the model to design a multi-step project plan or a complex software architecture. You notice in the initial 'thinking' phase that it is heading down the wrong technical path.
Before & after
Previously, you had to wait for the entire response to finish before prompting for changes, often cycling through 3-4 iterations over 30 minutes.
By providing mid-flight instructions, you can refine the output as it generates, reducing a complex technical drafting process to 5-10 minutes.
The Prompt
[INPUT_YOUR_COMPLEX_REQUEST_HERE] (As the model provides its plan, use the 'adjust course' feature to say: 'Actually, focus more on the [SPECIFIC_COMPONENT] and skip the [UNNECESSARY_DETAIL] section.')
The GPT-5.4 Thinking model introduced an upfront plan and the ability to steer the model while it is still 'thinking'. This prevents the common frustration of waiting for a long, incorrect output to finish before being able to correct it.
Source
Model Release Notes | OpenAI Help Center"you can adjust course mid-response while it’s working, and arrive at a final output that’s more closely aligned with what you need."