~/series/
One Concept a Day — The AI-Era Engineer's Glossary
30 lessons published · new lesson added daily.
1
Day 1
Why a Healthy Load Balancer Can Still Have High p99 Latency
2
Day 2
Why Agents Break 'Never' Rules in System Prompts
3
Day 3
Why Agent Mistakes Often Come From Stale Context, Not Hallucination
4
Day 4
How a 95% Cache Hit Ratio Can Still Overload Your Origin Server
5
Day 5
How Two Agents Editing the Same Service Conflict at the API Gateway
6
Day 6
Why rate limiting an AI agent can cause a request swarm
7
Day 7
Why a healthy queue can still lose a message
8
Day 8
Why retrying a tool call can double-charge with no bug in the code
9
Day 9
Why circuit breakers miss agent drift
10
Day 10
Why an empty dead letter queue can hide data loss
11
Day 11
Why adding a database index didn't speed up reads
12
Day 12
Multi-agent bottlenecks explained as a hot shard problem
13
Day 13
Why Passing Evals Once Doesn't Mean an Agent Is Reliable
14
Day 14
Why Rebalancing Agent Sessions Raises Your Token Bill
15
Day 15
How a bloom filter skips vector-DB lookups for duplicate chunks
16
Day 16
Why a webhook is a public endpoint anyone can POST to
17
Day 17
Why feature flags roll back faster than git revert
18
Day 18
Why blue-green deployment can't roll back a database migration
19
Day 19
Why OTel traces can't show which agent changed shared state, and when
20
Day 20
How missing backpressure breaks both a sprint board and an agent fleet
21
Day 21
The Saga Pattern
22
Day 22
Event Sourcing
23
Day 23
CQRS and Projection Lag
24
Day 24
Service mesh for AI agents
25
Day 25
Vector databases
26
Day 26
Embeddings
27
Day 27
RAG (Retrieval-Augmented Generation)
28
Day 28
The orchestrator agent
29
Day 29
Verifier agents
30
Day 30
Agent guardrails