Compare Claude Opus 4.7 and ChatGPT 5.4 on design-heavy code

Set up two different kinds of challenge

Luke’s April video compares real-code design work and a flight simulator built from a one-shot request. The existing SaaS page provides constraints that a new app does not. Keep those tasks separate when deciding what the output says about each model’s usefulness for your own work.

1. Save the starting revision of the real project and give each tool the same redesign request. Identify the page’s audience, existing content and main action before inspecting the results.
2. For the simulator, keep the same initial request and record what it leaves unspecified. Judge the first attempt separately from any later clarification or repair.

Inspect the interface and the working behaviour

The timeline shows early progress, a simulator result at 7:18 and landing-page results at 9:40 and 10:56. Completion order is one observation; whether the app works and whether its design suits the task are different checks. The description does not supply a detailed scored winner for every requirement.

1. Open each generated result, inspect it at the same screen sizes and test the controls that support its main task. Record missing behaviour separately from visual preferences.
2. Compare the redesigned page with the original brief and content. Check hierarchy, readable text, navigation and calls to action rather than choosing solely from the first screen.

Use prompt quality as part of the diagnosis

Luke discusses poor prompts and autonomy in design-heavy workflows. A vague request reveals the model’s defaults, but a failure to infer an unstated preference is different from ignoring a clear requirement. That distinction makes the next revision more useful than a generic request to improve everything.

1. List the decisions you explicitly requested and those the model supplied. Choose one observable mismatch and give the same clarification to both tools if continuing the comparison.
2. Recheck the revised result against the task and preserve the original output. Keep timing, working functionality, design judgement and revision effort visible as separate observations.

The takeaway

Choose from the task you actually need to complete. A simulator challenge and a SaaS redesign reveal different strengths, and prompt changes belong in the comparison record.