Illustrative example
-
01
Start with an open question
Use Spike when you need to see or test something before deciding. How dense should this dashboard be? Does this animation help? Would a different approach produce better results?
> /spike Can we make this dashboard more compact without making urgent issues harder to spot? Show me a few approaches. -
02
Compare the evidence
Claude builds the experiment separately from your app and gathers the outputs in one comparison. Depending on the question, that could be rendered alternatives, measured results, or both.
Review the options, ask for another pass, and compare the tradeoffs before changing the product.
Deploy finishedBackup completePayment API errors3 new signupsDisk 94% on db-2Weekly report readyCache clearedInvoice sentCertificate renewedACompact everything
More information fits on screen, but urgent issues compete with routine updates.
Payment API errors312 failed since 2:14 PMDisk 94% on db-2Rising 2% an hourDeploy finishedBackup complete3 new signupsWeekly report readyBCompact the routine items
Everyday information takes less space, while urgent issues stay prominent.
2urgent7updates12resolvedUrgent (2)Updates (7)Resolved (12)CShow summaries first
The dashboard is easier to scan, but individual details take another interaction to reach.
-
03
Make the call
Choose a direction, reject the approach, or leave it for later. Claude waits for your decision before bringing anything into the app.
If you move forward, the chosen approach is rebuilt to meet your project’s standards.
> Go with B. Keep the urgent items expanded and make the routine updates more compact.⏺ Built B into the dashboard. Routine updates are compact, and urgent issues stay expanded. M src/dashboard/Dashboard.tsx M src/dashboard/dashboard.css A src/dashboard/CompactRow.tsx M studies/dashboard-density/README.md Outcome: SHIPPED
-
04
Keep what you learned
Each experiment leaves a record of the question, method, results, and decision. Its outcome appears at the top, so a future session can see immediately whether it shipped, was disproved, or was parked.
A rejected idea still leaves something useful: evidence of what didn’t work and why.
Shipped: Compact routine updates; keep urgent issues expanded.
We chose B to reduce dashboard density while keeping urgent issues prominent. A gave urgent issues too little emphasis. C put individual details behind another interaction.