~/series/
AI Fundamentals in 30 Days
29 lessons published · new lesson added daily.
1
Day 1
How to Estimate the Token Cost of an Agent Task
2
Day 2
Why an Agent Can Ignore a Clear Rule in a Long Spec
3
Day 3
What 'Attention Is All You Need' actually introduced
4
Day 4
Why quickly approving a clean agent PR is risky
5
Day 5
Why you can't reliably track changes to a prompt
6
Day 6
What a system prompt is and why every AI has one
7
Day 7
Why temperature=0 doesn't make an agent deterministic
8
Day 8
Why an LLM returns different JSON for the same input
9
Day 9
Why the same prompt costs different amounts across models
10
Day 10
How stale vector stores make RAG return outdated code
11
Day 11
Why vector search misses exact matches like SKUs
12
Day 12
How Vector Search Wired a Rate Limiter Into an Auth Route
13
Day 13
Why RAG Bots Give Wrong Answers from a Stale Index
14
Day 14
Why inference-time context beats fine-tuning when base models upgrade
15
Day 15
When to rewrite the prompt instead of fine-tuning a model
16
Day 16
Why to exhaust prompt engineering before fine-tuning a LoRA
17
Day 17
How process hallucination lets an agent claim tests ran when they didn't
18
Day 18
Tool Schema Versioning
19
Day 19
Vision encoders and token budgets
20
Day 20
Reasoning Tokens
21
Day 21
AI alignment
22
Day 22
RLHF and sycophancy
23
Day 23
Frequency bias in coding agents
24
Day 24
Prompt injection
25
Day 25
Context engineering
26
Day 26
Evals from real failures
27
Day 27
AI coding benchmarks
28
Day 28
Open vs closed models in agents
29
Day 29
AI agents and hand-backs