A small bug fix can take eight hours of engineering work yet reach the customer ten days later. The gap may be filled with queues, handoffs, context switching, and rework. Each step can look responsible on its own while the customer continues to experience the bug. Waste becomes visible when you examine the whole path from request to working software, not just the time spent coding. section 2.1 section 1.2.2
Waste is unnecessary non-value-adding work: effort that creates no customer value, does not meaningfully reduce risk, and could be removed without harming delivery or its outcomes. Do not equate it with all work that customers cannot see. Automated tests, security reviews, and compliance checks may be necessary to deliver safely. Whether an activity is waste depends on its purpose, the system, and the risk of removing it—not on how busy or sophisticated it looks. section 2.3 section 2.3.3 section 2.3.5
Look for recurring friction in everyday engineering work:
- Waiting for builds, reviews, approvals, or environments; repeatedly handing work between people or tools. section 2.3.2
- Reworking code after requirements or defects surface late, or debugging behavior made unclear by poor observability. section 2.3.2 section 1.2.2
- Building more than the request needs: unused logic, speculative abstractions, or AI-generated rules that others must review, test, and maintain. section 3.3 section 3.3.2
A pull-request audit
Choose a recent PR that took longer than expected. List the work it required, then mark each item as value-adding, necessary non-value-adding, or waste. For each suspected waste, ask why it existed, what system constraint produced it, and whether it was avoidable. Finish by asking which small system change would remove the largest source of friction. The aim is to see a pattern, not blame the developer or automatically remove a safeguard. section 2.6 section 2.3.3
Table 2.1 Comparison table of common activities and their classification
|
Activity
|
Classification
|
Why
|
|
Implementing a user-visible feature
|
value-adding
|
Directly changes the behavior the customer cares about
|
|
Fixing a production defect
|
value-adding
|
Restores the expected customer experience
|
|
Performance optimization on a critical path
|
value-adding
|
Improves the latency or throughput that users notice
|
|
Writing automated tests to protect core behavior
|
Necessary Non-Value-Adding
|
Does not change behavior, but reduces risk
|
|
Security scanning and dependency checks
|
Necessary Non-Value-Adding
|
Enables safe delivery and compliance
|
|
CI/CD pipeline execution
|
Necessary Non-Value-Adding
|
Required to deliver changes reliably
|
|
Waiting for builds, reviews, or approvals
|
Non-value-adding
|
Pure delay with no value created
|
|
Manual environment setup
|
Non-value-adding
|
Exists due to system limitations
|
|
Re-work caused by late feedback
|
Non-value-adding
|
Corrects preventable errors
|
|
Maintaining duplicate documentation
|
Unnecessary Non-Value-Adding
|
The source of truth is unclear
|
|
Manually triaging unreliable tests instead of fixing them
|
Unnecessary Non-Value-Adding
|
Compensates rather than improves
|
|
Adding defensive checks for conditions that the system already guarantees
|
Unnecessary Non-Value-Adding
|
Repeats validation
|
The activity comparison gives concrete cases for testing the boundary between customer-visible value, necessary protection, and delay. Treat its classifications as prompts for judgment, not permanent labels. section 2.3.6
AI assistance can multiply waste when a vague request becomes a large, plausible diff. Extra behavior creates decisions that reviewers inherit and maintainers may have to reconstruct later. Make the intended behavior and constraints explicit; compare generated code with the specification, and question additions that serve no current need. The useful question is not how much code was produced, but whether the change reduces downstream work while preserving the outcome. section 3.1.3 section 3.6 section 4.5.3

