Ten weeks. One operator.

Topics

• AI vs. ML vs. LLMs vs. Generative AI: what the words actually mean
• How transformers and tokens work — conceptually, no math required
• Why models hallucinate and what to do about it
• The current model landscape: Claude, GPT, Gemini, Llama, open source
• The Outpost AI thesis: why we are AI-native and what that means for clients

Resources

• But what is a GPT?
  3Blue1Brown
• Intro to Large Language Models
  Andrej Karpathy
• The model landscape, today
  Outpost Internal
• Elements of AI
  University of Helsinki
• Anthropic Use Case Gallery
  Anthropic

Output

One-page client-ready explainer memo. Prompt: "If a client asked me what AI is and why it matters for their business, what would I say?"

Topics

• Anatomy of a strong prompt: role, context, task, constraints, format
• Few-shot prompting with examples
• Chain of thought and reasoning prompts
• Structured outputs: JSON, tables, schemas
• Common failure modes and how to debug them
• When to iterate vs. start over with a fresh prompt
• Best-practices checklist for shipping prompts

Resources

• Anthropic prompt engineering guide
  Anthropic
• ChatGPT Prompt Engineering for Developers
  DeepLearning.AI
• OpenAI prompting cookbook
  OpenAI
• The Prompt Report (taxonomy of techniques)
  arXiv

Output

A 10-prompt personal library for the hire's role, submitted to the Outpost shared library.

Topics

• Claude: web, desktop, Projects, Artifacts, Skills, MCP connectors
• ChatGPT: Projects, Custom GPTs, memory features
• Cursor: AI-native editing for code and markdown — relevant for non-engineers too
• Choosing the right tool: speed of iteration vs. depth of reasoning vs. tool access
• Voice mode, mobile, and other interfaces that change how you work

Resources

• Anthropic Claude product documentation
  Anthropic
• Claude Code documentation
  Anthropic
• OpenAI Custom GPT and Projects docs
  OpenAI
• Cursor onboarding guide
  Cursor
• Outpost tool matrix
  Outpost Internal

Output

Personal AI Stack memo — one page covering which tools the hire uses, for what, and why.

Topics

• Context windows: what they are, why size matters, where they break
• The difference between context, memory, and retrieval
• File-based context: feeding AI exactly what it needs, no more
• Project knowledge and persistent context across Claude, ChatGPT, Cursor
• Skill files: writing instructions that turn a general model into a domain specialist
• Memory management for long-running tasks and how to prevent context drift
• The Outpost Skills library: how we use Skills internally and contribution standards

Resources

• Extend Claude with skills
  Anthropic
• Agent Skills open standard
  agentskills.io
• ChatGPT Custom Instructions & memory
  OpenAI
• Cursor Rules files
  Cursor
• Outpost Skill file examples
  Outpost Internal

Output

Build one Skill file or Custom GPT that captures a piece of the hire's role expertise.

Topics

• What RAG is and what problem it actually solves
• Vector embeddings explained without math
• Vector databases: Pinecone, Chroma, Supabase Vector — what to pick when
• When to RAG vs. when to just paste context
• Chunking, embedding, retrieval, re-ranking — the four-stage pipeline
• RAG failure modes: bad chunks, irrelevant retrieval, stale data
• Building and evaluating a real pipeline end-to-end

Resources

• Contextual Retrieval
  Anthropic
• What is Retrieval-Augmented Generation?
  NVIDIA
• What is Retrieval-Augmented Generation?
  IBM
• What is Retrieval-Augmented Generation?
  AWS
• LangChain RAG tutorial
  LangChain
• Supabase Vector quickstart
  Supabase
• White Label Storage case study
  Outpost Internal

Output

Working RAG bot over a real Outpost knowledge base — must answer 5 evaluation queries correctly with cited sources.

Topics

• What MCP (Model Context Protocol) is and why it changes the game
• Function calling: the underlying mechanic every agent uses
• Connecting Claude / Cursor to Gmail, Drive, Calendar, HubSpot, Slack, Ramp, Todoist
• Building a connected workspace — the Outpost stack walkthrough
• Authorization, scopes, and what to never connect
• MCP vs. custom integration vs. Zapier vs. Skills — the decision tree

Resources

• MCP documentation
  Anthropic
• Function calling guide
  OpenAI
• Public MCP server directory
  Community
• Claude connector directory
  Anthropic
• Anthropic Help Center: connectors
  Anthropic
• Outpost connected stack walkthrough
  Outpost Internal

Output

A working workflow that uses at least 3 connected tools end-to-end. Documented, demo-able, and at least one teammate could re-run it.

Topics

• What is an agent, what is not, and why the distinction matters
• The agent loop: plan, act, observe, repeat
• Single-step vs. multi-step agents
• Agent vs. workflow vs. assistant — the choice that determines cost and reliability
• Frameworks: Claude Agent SDK, OpenAI Agents, Cursor agents, n8n
• Cost, latency, and reliability — the three things agents always cost more on
• Failure modes: infinite loops, hallucinated tool calls, scope creep

Resources

• Claude Agent SDK
  Anthropic
• Building effective agents
  Anthropic
• n8n quickstart
  n8n
• Claude Research — Anthropic Help Center
  Anthropic
• Outpost agent case studies
  Outpost Internal

Output

A working agent prototype that completes a real multi-step task in the hire's domain. Must include a 'stop condition' and at least one human-in-the-loop checkpoint.

Topics

• Why evals matter — especially for client-facing AI work
• Types of evals: deterministic, LLM-as-judge, human review
• Writing test cases and building a golden dataset
• Iterating prompts and systems using eval feedback (the inner loop)
• Eval tooling: Braintrust, LangSmith, or a simple Python harness
• Wiring evals into client deliverables and acceptance criteria
• Eval anti-patterns: cherry-picking, judge-model bias, vanity metrics

Resources

• Your AI Product Needs Evals
  Hamel Husain
• Evals guide
  Anthropic
• Braintrust intro
  Braintrust
• Outpost eval examples
  Outpost Internal

Output

Eval suite (5–10 cases) for the Week 7 agent. Iterate the agent based on results. Document the deltas.

Topics

• Track A — Data & AI: production RAG, fine-tuning vs. prompting, dbt + Databricks AI patterns, White Label Storage case study
• Track B — Finance & Ops: financial modeling with AI, categorization & reconciliation automation, Ramp audit case studies
• Track C — Marketing & Brand: content engine workflows, SEO automation, personalization at scale, Outpost outbound case study
• Track D — Search Ops & BD: ICP qualification, industry research automation, outreach personalization, vertical research patterns
• Cross-track: deciding when to specialise vs. when to stay general
• Cross-track: how track tools become firm-wide assets

Resources

• Track A mentor: Sameer / Maaz (Data)
  Outpost Internal
• Track B mentor: Taimoor / Fazeela (Finance)
  Outpost Internal
• Track C mentor: Ilma (Marketing)
  Outpost Internal
• Track D mentor: Zahra (Search Ops)
  Outpost Internal
• Track-specific case study packs
  Outpost Internal

Output

A working tool or workflow that solves a real problem in the hire's track. This is the seed of the Week 10 capstone — over-invest here.

Topics

• Days 1–3: polish the capstone — real workflow, real evals, real docs
• Day 4: Demo Day — 10-minute presentation + Q&A to managers and the wider team
• Day 5: write your Personal AI Operating Manual; submit to the Outpost playbook
• Capstone bar: 2+ of (prompts, Skills, RAG, MCP, agents), an eval suite, and documentation
• Demo Day craft: tell the story in 4 acts — problem, approach, demo, what's next
• The Operating Manual: distil 10 weeks of habits into a 2-page contribution

Resources

• Capstone rubric
  Outpost Internal
• Demo Day examples (past cohorts)
  Outpost Internal
• Operating Manual template
  Outpost Internal
• Outpost playbook
  Outpost Internal

Output

(1) Capstone shipped + documented. (2) Demo Day talk delivered. (3) Personal AI Operating Manual contributed to the Outpost playbook. Outpost AI Operator status confirmed.