Repository navigation
Cost Log
Real measured costs, contributed by people who ran the material. Not estimates.
Why this is a wiki page.
docs/setup/cost-controls.mdis the guidance — where money goes, and the teardown checklist. It deliberately quotes no prices, because stale numbers are worse than none. This page is the observed data, with dates attached so you can judge how stale it is.
Before any table: cost here is not mostly inference.
- Inference is cheap. Cents per exercise, a few dollars per module.
- Idle infrastructure is not. OpenSearch Serverless collections and AgentCore runtimes bill for existing, not for use. A collection created on Tuesday and forgotten is still charging on Friday.
Almost every "this cost far more than expected" report traces to #2, not #1. Teardown checklist.
| Module | Reported | Torn down same day? | Region | Date | By |
|---|---|---|---|---|---|
| 02 Bedrock Essentials | — | — | — | — | — |
| 03 Bedrock Agents | — | — | — | — | — |
| 10 RAG / OpenSearch | — | — | — | — | — |
| 11 AgentCore | — | — | — | — | — |
| 14 End-to-end | — | — | — | — | — |
Add a row after you finish a module. The "torn down same day" column is the most informative one — the same module costs wildly different amounts depending on it.
| Approach | Total | Period | Notes | By |
|---|---|---|---|---|
| — | — | — | — | — |
Useful to state: whether you tore down as you went, which modules you skipped, and whether anything was left running by accident (that one is the most useful of all).
Genuinely free, no account needed: Modules 00, 01 and 15, every L.A.B. lab, rag_by_hand.py, quality_gate.py, and the whole field guide. Roughly 20 hours of real work.
- Budget alarm at 50/80/100%. It does not stop spending — it tells you
- A sandbox account, not production
- The teardown checklist somewhere you will actually see it
- Know that Modules 10, 11 and 14 create infrastructure that bills for existing
| Driver | Controlled by | Depth |
|---|---|---|
| Tokens per turn | Prompt and schema discipline | Token Tax Ledger |
| Turns per task | Topology — every handoff re-sends context | Handoff Multiplier |
| Non-linear jumps | Eight named cliffs | Cost Cliff Map |
| Idle infrastructure | The teardown checklist | Cost controls |
Turns per task is almost always the highest-sensitivity term. It multiplies four of the six token taxes at once, which is why cost surprises are usually topology surprises.
If you got a surprise bill, please add it. A row saying "left an OpenSearch collection up for three weeks, £—" is more useful to the next person than every carefully-torn-down row on this page. No judgement — it is the most common mistake here and the whole reason the checklist exists.
This wiki is the connective layer. The repository is canonical and versioned; these pages are the cross-surface maps and the notes that change too often to live in a reviewed file. Anyone can edit — see Editing this wiki.
▶ Start here · 55 lessons, on the site ↗
- Getting started · 2
- Agentic PDLC fundamentals · 9
- Methods decoded · 7
- Running delivery · 13
- By role · 9
- Teams and organisation · 4
- Practice · 4
- Interviews and careers · 7
The method
The Agentic PDLC
The Eight Loops
Gates & Governance
The Evidence Pack
The journey · end to end
Product manager
Solution architect
Engineering lead
QA lead
DevOps & platform
The method · by role
Product manager
Solution architect
Engineering lead
QA lead
DevOps and platform
Sponsor
How to
Run an NFR workshop
Design an agent on paper
Build, buy or borrow
Cut sprints into bolts
Review by risk band
Prove the bar
Control the token bill
Hold the security boundary
Run a missing-control postmortem
Reference
Mental Models
Glossary
Formulas & Calculators
Decision Trees
Scenario Library
Exercises & Answers
Anti-Patterns
Sources & Confidence
Where do I find…?
The 16 modules
- Agentic Foundations
- LLM Intuition and the AWS Bridge
- Amazon Bedrock Essentials
- Amazon Bedrock Agents
- Agent Builder, Knowledge Bases and Guardrails
- The Agent Loop: No Framework to Strands
- Strands Foundations: Tools, Memory and MCP
- Multi-Agent Patterns with Strands
- LangChain and LangGraph
- LLM Memory Mechanics
- RAG, OpenSearch and LiteLLM
- Amazon Bedrock AgentCore
- A2A and A2UI: Agent Interoperability
- Agentic QA and Evaluation
- End-to-End Production Pipeline
- Agentic Product Lifecycle
Learning paths · Field guide
Study Plans
Eight sessions
Living notes
Error Index
Community Answers
Field Notes
Model & Region Notes
Cost Log
Scoreboard 🧪
The project
Roadmap
Editing this wiki
Maintainer Runbook
📘 The manual · live
🛫 Ninety Days, the simulator
🧰 The workbench
👔 For leadership
🧠 Mental models
Templates · Prompts
Frameworks & acronyms
The repository · Setup
Pinned discussions
Discussion map — where to post what
Exercise & lab index
Simulator Arena — get a lab graded
Pitch an idea
All discussions →