~/series/
Software Engineering in the AI Era
29 lessons published · new lesson added daily.
1
Day 1
Why Faster Coding With Agents Leaves PRs Waiting Longer in Review
2
Day 2
Why the Same Prompt Fails When a Different Person Runs It
3
Day 3
Why a second AI reviewer on the same model doesn't help
4
Day 4
How an agent can fix the test instead of the bug
5
Day 5
Why free context doesn't stop an agent from using a deleted file
6
Day 6
How an agent reasoned past a 'never force-push' rule
7
Day 7
How an ambiguous spec led an agent to a GDPR violation
8
Day 8
Why AI-written PRs get less review scrutiny than human ones
9
Day 9
Catching a coding agent's plan before it goes wrong
10
Day 10
What makes a code review actually catch agent bugs
11
Day 11
How Agent-Written Tests Missed a Stacked-Discount Bug
12
Day 12
How a Correct PR Still Took Down Three Services
13
Day 13
Why splitting agent tasks by role breaks shared context at merge
14
Day 14
Why approval fatigue makes you stop reading agent permission prompts
15
Day 15
Why git revert can't undo an agent's API calls, emails, and charges
16
Day 16
Why trace-and-span dashboards can't catch agent task drift
17
Day 17
How to calculate the real cost per task of an AI agent
18
Day 18
Debugging agent loops
19
Day 19
Same-pass verification bias
20
Day 20
Code Knowledge Base Staleness
21
Day 21
Exemplar audits
22
Day 22
Cold-start agent smoke test
23
Day 23
Measuring flow
24
Day 24
Reviewing agent output
25
Day 25
Interrupt literacy
26
Day 26
Engineering taste
27
Day 27
The many hands problem
28
Day 28
Prompt injection in legacy repos
29
Day 29
Handoff radius