AI workflows
Your token count isn’t your AI bill.
Before comparing AI usage, check what the report counts, where the day begins, and which records are missing.
Read the noteAI, software & building
Small experiments. Clearer decisions.
Featured · AI workflows
Your token count isn’t your AI bill.
Before comparing AI usage, check what the report counts, where the day begins, and which records are missing.
Read the noteFrom the notebook
10 notes
Software engineering
Before debugging the value, check the context.
Why the same JavaScript characters can produce different results—and what to preserve in a useful reproduction.
Read the note
Building products
A rollback button is not a recovery plan.
Application code, database state, and uploaded files can travel back to different points in time. Make the boundary explicit.
Read the note
Software engineering
Twenty characters. According to which counter?
A name field needs a definition of character before it needs a validation library. Compare code units, code points, visible clusters and storage bytes.
Read the note
Software engineering
The test was green before the fix, too.
A unique database row can hide repeated attempts and repeated side effects. Choose the observation that matches the bug.
Read the note
Software engineering
No value is not one instruction.
An update request can mean keep, clear or reject. Preserve that distinction through form state, JSON serialization and server validation.
Read the note
AI workflows
After Cancel, the old answer can still arrive.
Transport cancellation and permission to update the screen are separate. Use request identity to stop a late answer from replacing newer input.
Read the note
Software engineering
Does a still 3D scene need another frame?
A static product viewer and an animation have different inputs. Name the changes that require a redraw before changing the render loop.
Read the note
Building products
The container finished. Where did the file go?
A successful exit does not establish that a report survives outside its container. Trace the output path before cleaning up the run.
Read the note
Software engineering
Average first, convert later? The order changes the answer.
The average of converted measurements can differ from the conversion of their average. A two-number example makes the choice visible.
Read the noteNo notes match that search. Try another word or choose a different topic.
A small experimentSame characters.
Same characters.
Different context.
Explore a recorded JavaScript result. Change the context before blaming the value.
Read the explanation{} + []Resultnumber
Recorded in Node.js v26.0.0 using vm.runInNewContext. Each input was evaluated as a separate Script.
Useful ideas leave room
for the conditions.
A clear example, a source you can inspect, and an honest boundary around what was tested. That’s the shape of a MatthewLab note.
Meet the notebook