Which Feature Fits
| Feature | Best for | Not for |
|---|---|---|
| Chat | One-off questions and quick tasks | Work that needs the same standing context every time |
| Project | Recurring work with shared instructions and knowledge | A single throwaway question |
| Research mode | Multi-source questions needing a synthesized, cited report | Quick lookups |
| Artifact | Standalone content you will reuse or revise | Short conversational answers |
Model Tiers
| Tier | Character | Typical fit |
|---|---|---|
| Haiku | Fastest and lightest | High-volume, simple, time-sensitive tasks |
| Sonnet | Balanced capability and speed | Most everyday business work |
| Opus | Most capable | Hard reasoning, complex analysis, high-stakes quality |
Decision Table
| If the question says... | The answer is likely... |
|---|---|
| Simple, repetitive, high volume | Smaller, faster tier |
| Everyday drafting and analysis | Balanced tier |
| Complex reasoning where quality matters most | Most capable tier |
| "Use the strongest model for everything" | Usually wrong. Match the tier to the task |
| Same instructions typed every week | Move them into a Project |
| Chat is long and early details are lost | Summarize and restart, or persist standing facts in a Project |
Traps
| Trap | Correct Answer |
|---|---|
| Largest model by default | Choose the smallest tier that passes your quality check |
| Keep going in an overlong chat | Summarize and restart; review the summary first |
| Research mode for a quick fact | Use chat for simple lookups |
| Memory replaces Project knowledge | Keep team-critical reference material in a Project |
If You Remember Nothing Else
- Match the model tier to the task; defaulting to the largest is often wrong.
- Projects hold what repeats; prompts hold what is one-off.
- Long chats degrade. Restart with a reviewed summary or persist standing information.
- Verify citations from research mode before relying on them.