Classify the task before touching the artifact. A generative task may be created freely; a transformative task preserves the existing artifact and modifies only the authorized surface. The classification decides the method, and it is made before any work starts.
By Pawan Nayar · from the LLOS AI Learning record
Classify the task before touching the artifact.
A generative task may be created freely; a transformative task preserves the existing artifact and modifies only the authorized surface. The classification decides the method, and it is made before any work starts.
Classify the task before touching anything, because the classification decides the method. If the task is **generative**, the model may create freely. If the task is **transformative**, the model preserves the existing artifact and modifies only the requested surface. This is decided at step zero. The common failure is not a misunderstood design detail discovered late — it is the wrong transformation strategy chosen at the start, which then produces a dozen small drifts nobody asked for. **The instruction pattern for a transformation task** State plainly: this is a controlled transformation task, not a reconstruction task. Then: 1. Start from the supplied source artifact itself. 2. Duplicate it unchanged. 3. Identify only the regions authorized for modification. 4. Replace content inside those existing regions. 5. Delete only regions explicitly marked for removal. 6. Do not recreate existing structure. 7. Do not introduce replacement wrappers, layout systems, CSS, components or interactions unless requested. 8. Anything not explicitly named for modification is immutable. 9. Preserve the existing artifact even when another implementation appears cleaner or easier. 10. Verify structural preservation, not merely visual similarity. **Required method: COPY → REPLACE → DELETE → PRESERVE.** Point 9 is the one that gets broken most often, and it is the one that costs the most. "Cleaner" is the model's judgment substituted for the user's approved work.
A user provides a carefully refined reference page: "Follow this exactly. Copy the page, replace section content one section at a time, and remove sections that are not needed." The model instead reads the reference, extracts its CSS and visual patterns, creates a new body or component tree, inserts the new content, recreates decorative elements, adjusts spacing, changes proportions — and describes the result as a clone. The output looks broadly similar. But cards become narrower, text measures change, wrappers appear, spacing shifts, sections change order, decorative elements behave differently, responsive behaviour changes, controls move, and inherited details disappear. The model did not fail because it misunderstood every design detail. It failed earlier, at the point where it chose to recreate rather than duplicate.
Resolved from this lesson's own record — real definitions, not invented for this page.
More from the same record, pulled by shared topic — not written new for this page.
Because AI models are strongly optimized for generation. Given a reference artifact and asked to make another version, the model behaves like an author: infer the pattern, reconstruct the structure, generate a fresh version, fill in the new content, smooth inconsistencies, and make local decisions along the way.
An AI model must show results before making changes. The user decides if something is wrong.
Classify the task before touching the artifact. A generative task may be created freely; a transformative task preserves the existing artifact and modifies only the authorized surface. The classification decides the method, and it is made before any work starts.