Self-study course · Syllabus v1 · October 2026

AI Transformation with Claude

A technical, hands-on path from Claude power user to independent AI transformation consultant for small and mid-size companies. Every module ends with something you build and can show to a client.

12 weeks8–10 h / weekPython-first12 portfolio deliverables1 capstone engagement

Who this is for and what you leave with

Written for one learner: a data scientist and Python developer who already runs an IT services company and wants to sell AI transformation work. The course skips ML fundamentals and spends the time on three things clients actually pay for: finding the right use cases, building reliable Claude-based systems quickly, and proving the value afterwards.

Technical outcomes

Design Claude solutions across apps, Claude Code, the API and the Agent SDK
Build and ship MCP servers that connect Claude to client systems
Write Skills and plugins that package a company's know-how
Run evals, cost routing and guardrails in production

Business outcomes

A repeatable AI readiness audit you can sell as a fixed-price entry offer
Proposal, SOW and ROI templates
A public portfolio of 12 working demos plus a case study
A positioning statement and outreach plan for Israeli SMBs

The Claude product map

Most SMB work is picking the lowest layer that solves the problem. Start at the top; go down only when you need more control.

Rule of thumb for proposals: a quick win lives in the top two layers; a durable system usually needs the bottom two.

Part 1 · Foundations

weeks 1–3

Week 1

The AI transformation landscape

What "AI transformation" means for a 10–200 person company, where the money is, and how Anthropic's products are positioned against the alternatives.

Model families (Opus, Sonnet, Haiku) and how to choose by task, latency and cost
Augmentation versus automation; reading the Anthropic Economic Index for task-level evidence
Data privacy, retention and plan types (Pro, Team, Enterprise, API) as they matter to a client

Claude appsEconomic Indexmodel selection

Deliverable: a one-page "Claude for SMBs" briefing you can send to prospects.

Week 2

Prompt and context engineering

The core craft. Structured prompts, examples, XML tags, system prompts, long context, and how Projects and memory change what you put in each message.

Prompt patterns: role, task, constraints, examples, output format
Context engineering: what goes in the Project, the Skill, the prompt, or a tool
Extended thinking and when it pays off

Projectssystem promptsprompt library

Lab: build a prompt library of 15 reusable business prompts (sales, finance, HR, ops) with test inputs and expected outputs.

Week 3

Claude apps for knowledge work

Most clients start here. Learn to deploy Claude to a team so it is used daily, not tried once.

Projects as team knowledge bases; artifacts, Docs, Slides, Sheets and dashboards
Connectors (Google Workspace, Microsoft 365, Atlassian, Slack) and research
Claude in Chrome and scheduled tasks for recurring work
Rollout playbook: champions, training, usage reviews

Team planconnectorsChromescheduled tasks

Lab: set up a demo "company workspace" for a fictional 30-person logistics firm with three Projects, two connectors and one weekly scheduled report.

Part 2 · Building with Claude

weeks 4–8

The reference architecture behind modules 4–8. Each module builds one box.

Week 4

Claude API in Python

The platform layer. Everything later is built on these calls.

Messages API, streaming, structured outputs, vision and PDF input
Tool use (function calling) and the manual tool loop
Prompt caching and the Batch API for cost control
Error handling, retries, rate limits, logging

anthropic SDKtool useprompt cachingbatches

Lab: an invoice and receipt extractor (PDF → validated JSON → CSV) that runs 500 documents through the Batch API and reports cost per document.

Week 5

Knowledge retrieval (RAG) for company documents

The most requested SMB use case: "ask our documents". Learn when long context is enough and when you need retrieval.

Chunking, embeddings, hybrid search, contextual retrieval
Citations and grounding to cut hallucinations
Hebrew and Russian documents: tokenisation and search quality

APIcitationsvector DB

Lab: a bilingual (Hebrew/English) policy assistant over 200 pages of HR and procedures, with cited answers.

Week 6

Model Context Protocol (MCP)

MCP is how you plug Claude into a client's systems once, and reuse that connection across apps, Claude Code and agents. This is a strong consulting differentiator.

MCP concepts: servers, tools, resources, prompts; local versus remote servers
Writing a Python MCP server (FastMCP), auth, permissions, read versus write tools
Publishing as a custom connector; security review checklist

MCPFastMCPcustom connectors

Lab: an MCP server for a public Israeli data source or a small CRM, used from the Claude app and from Claude Code.

Week 7

Claude Code as a delivery engine

Use Claude Code to build client solutions several times faster, and teach clients' developers to do the same.

CLAUDE.md, project memory, slash commands, hooks, subagents
Plan → implement → test loops; orchestrating cheaper models for sub-tasks
GitHub workflows, code review, CI integration

Claude CodesubagentshooksGitHub

Lab: a reusable project template (CLAUDE.md, skills, MCP config, tests) that you start every client engagement from.

Week 8

Agents, Skills and plugins

From assistants to systems that complete multi-step work on their own, packaged so the client can keep using them after you leave.

Claude Agent SDK and Managed Agents; when an agent is the wrong answer
Writing Skills that capture a company's procedures; bundling them as plugins
Human-in-the-loop approvals for anything that sends, pays or deletes

Agent SDKSkillspluginsapprovals

Lab: a lead-to-proposal agent: reads an inbound email, researches the company, drafts a proposal from a template, and waits for approval.

Part 3 · Running it in production

weeks 9–10

Week 9

Evaluation, cost and reliability

What separates a demo from a system a client will pay for every month.

Building eval sets from real client data; LLM-as-judge with care
Model routing (Haiku for triage, Sonnet for work, Opus for hard reasoning); caching; batching
Monitoring: logs, token dashboards, drift, alerting

evalsroutingobservability

Lab: add an eval suite and a cost dashboard to your module 4 or 5 project; cut cost by 40% without losing accuracy.

Week 10

Security, governance and compliance

Clients ask about this before they sign. Have clear answers.

Prompt injection, tool permissions, least privilege for MCP
Data handling, PII, retention; Israeli Privacy Protection Law and GDPR basics; EU AI Act overview
An AI acceptable-use policy and approval matrix for SMBs

securitypolicyrisk register

Deliverable: an AI usage policy template and a 25-point security checklist for Claude deployments.

Part 4 · The consulting practice

weeks 11–12

The maturity ladder you score clients against in the audit. Each step maps to one layer of the product map.

Week 11

Discovery, ROI and proposals

Turn technical skill into a sellable offer.

The AI readiness audit: interviews, process mapping, use-case scoring (value × feasibility × risk)
ROI model: hours saved, error reduction, revenue lift; baseline before you build
Packaging: audit → pilot → retainer; pricing fixed-scope versus time-based
Change management: training, champions, adoption metrics

auditROI modelSOW

Deliverable: audit questionnaire, use-case scoring sheet, ROI calculator, and a proposal template.

Week 12

Capstone engagement and positioning

Run a full engagement for a real or friendly company (a garage, a ceramics studio, a logistics firm), then publish it.

Audit → architecture → 2-week pilot build → handover → measured results
Write the case study; update website and LinkedIn; record a 5-minute demo
Outreach plan: 30 target companies, two content pieces a week, one workshop offer

capstonecase studygo-to-market

Deliverable: a published case study with before/after numbers and a live demo.

Weekly rhythm

Block	Time	What you do
Learn	2 h	Anthropic docs, Anthropic Academy courses, cookbook notebooks
Build	4–5 h	The module lab, in a dedicated repo, built with Claude Code
Package	1 h	README, 2-minute demo video, one-page summary for clients
Publish	1 h	One LinkedIn post or short article on what you built and learned

Assessment

Component	Weight	Pass criterion
Module labs (12)	40%	Runs from a clean clone; README explains the business problem
Eval and cost report	15%	Accuracy measured on 50+ cases; cost per task stated
Consulting toolkit	15%	Audit, ROI, policy and proposal templates ready to send
Capstone	30%	Real users, measured baseline and result, published case study

Toolkit and resources

Accounts

Claude Team or Max plan
Claude Console with API key and spend limit
Claude Code
GitHub

Python stack

anthropic, claude-agent-sdk
mcp / FastMCP
pydantic, FastAPI, pandas
a vector store (pgvector or Chroma)

Learning sources

docs.claude.com
Anthropic Academy
Anthropic cookbook (GitHub)
modelcontextprotocol.io
Anthropic Economic Index

Credentials to aim for

Anthropic Academy course completions
A public MCP server with real users
Two client case studies

Product names and features change quickly. Check each module's topics against docs.claude.com before you start it, and update this syllabus where something has moved.