~/series/
Multi-Agent Systems in 30 Days
29 lessons published · new lesson added daily.
1
Day 1
How to Delegate to Subagents More Effectively
2
Day 2
Comparing a 4-Agent Swarm to One Agent With a Scratchpad File
3
Day 3
Why two correct agents can still give a wrong answer
4
Day 4
How a small planner prompt change broke downstream agents
5
Day 5
How an agent crash caused three duplicate refunds
6
Day 6
Why a PASS/FAIL verifier agent weakens code review
7
Day 7
Why agent traces show more calls than the plan you approved
8
Day 8
Why agent memory needs to forget overturned decisions
9
Day 9
The blackboard concurrency bug that passes code review
10
Day 10
How an Undocumented API Field Rename Broke Our Agents
11
Day 11
How Removing handoff() Reopened a Fixed Support Ticket
12
Day 12
How two green PRs create a race git conflict detection can't see
13
Day 13
Why majority vote across agents fails when their errors aren't independent
14
Day 14
Why multi-agent voting fails when agents share a model
15
Day 15
Why feeding a reviewer agent the author's rationale hides bugs
16
Day 16
How a vague tool description made an agent pass the wrong ID
17
Day 17
Function calling as grammar constraint
18
Day 18
Designing the Observe Step
19
Day 19
Reasoning vs Routing in ReAct Agents
20
Day 20
Reflection steps in agent pipelines
21
Day 21
Guardrails on agent handoffs
22
Day 22
Prompt injection between agents
23
Day 23
Fan-out and fan-in
24
Day 24
Pipeline vs swarm
25
Day 25
Replay in multi-agent systems
26
Day 26
Human-in-the-loop for multi-agent systems
27
Day 27
Agents gaming each other
28
Day 28
Token cost in multi-agent fleets
29
Day 29
Evaluating agent handoffs