AI IN MANAGEMENT CONSULTING

Get oriented with the course structure, meet your instructor and fellow consultants, and share which AI tool you already use as you set expectations for the six weeks ahead.

• Course structure and expectations
• Instructor introduction
• Tool-usage baseline poll

Explore how large language models actually work under the hood, compare closed APIs against open-weight and local models, and learn to pick the right tool for each consulting task.

• The AI map
• Tokens, context windows, cut-offs and temperature
• Hallucinations and the consultant's golden rule
• Productivity tool landscape
• Closed APIs vs open-weight vs local models
• Demo: Live tokenizer walkthrough

Assignment #1: Model Comparison & Critical Evaluation 
Run three real consulting questions through at least two LLMs, then analyse each model's personality, hallucinations and best-fit use cases.

Master the CRAFT framework for building effective prompts, navigate common prompting pitfalls using the Goldilocks zone, and unpack Claude's Constitution to see how model values shape your system prompt.

• CRAFT framework: Context, Role, Action, Format, Tone
• Zero-shot, one-shot, few-shot & Chain-of-Thought
• 5 common prompting challenges and fixes
• Anti-patterns and the Goldilocks zone
• Claude's Constitution and the operator layer
• Demo: Live CRAFT before/after prompt runs

Assignment #1B: CRAFT Optimisation
Rewrite your Assignment #1 prompts using the CRAFT framework and compare the before-and-after results in a short video or PDF.

Shift from prompt engineering to context engineering, learn to manage your attention budget, and combine instructions, knowledge, tools and memory into a context-rich specialist assistant that delivers reliably.

• Context rot and the attention budget
• Four dimensions: instructions, knowledge, tools, memory
• SKILLs, RAG, Agent RAG and OKF
• Four knowledge routes: PE, RAG, agentic-JIT, fine-tuning
• Scaling the context stack

Assignment #2A: Context-Rich Specialist Assistant
Write and iteratively refine a detailed prompt for a specialised daily task, then document the process you used to get consistent results.

Examine the anatomy of a reusable AI Skill, compare how Claude, GPTs and Gems handle them differently, and learn the supply-chain security lessons behind the OpenClaw poisoning incident.

• Anatomy of a Skill
• Progressive disclosure and composability
• Platform differences: Claude, GPTs, Gems
• ClawHub/OpenClaw poisoning incident
• Demo: Building a brand-guideline PowerPoint with SKILLs
• Workshop: Breakout brainstorming and live practice drafting an AI Skill

Assignment #2B: Skill Conversion & Repetitive Application 
Convert your Assignment #2A prompt into one or more standalone Skills, then submit proof of it working consistently across more than one task.

Learn to spot repetitive workflows worth automating using the SEE, SPLIT and SPOT framework, climb the outcome ladder from elimination to human intervention, and apply Bornet's L0–L5 progression scale.

• SEE: discovery and the 8-question battery
• SPLIT: atomic-chunk criteria
• SPOT: SPAR and the five filters
• Outcome ladder from eliminate to human intervention
• Bornet's Agentic AI Progression Framework (L0–L5)
• Case study: Mapping the email-triage process live

Assignment #3A: Process Decomposition
Apply the 3S framework to a real workflow you do repetitively, breaking it into a long-list of automation opportunities with at least four or five "agentifiable" steps.

Score and prioritise automation opportunities by impact, feasibility and effort, frame an honest ROI without falling into the x10 trap, and draft a one-page agent SPEC and Agent Canvas.

• Scoring by impact, feasibility, effort and time-to-value
• Honest ROI and avoiding the x10 trap
• The one-page agent SPEC: 7 key questions
• The 10-box Agent Canvas
• Guest Speaker: Partner & TMT Industry Leader, Deloitte Portugal
• Workshop: 80/20 Pareto prioritisation exercise

Assignment #3B: Process Prioritisation, ROI & SPEC
Score the opportunity from Assignment #3A, calculate an honest ROI, and draft a one-page SPEC for your chosen workflow.

Dissect a real marketing workflow of a research agent, writer and human gate, map the 4S method directly onto an agent canvas, and compare building it in Cowork versus n8n.

• Research agent, writer and human-gate workflow
• 4S method mapped onto the agent canvas
• Route A (Cowork/Projects) vs Route B (n8n)
• Guest Speaker: Partner & CEO, WikiBrains
• Case study: How the workflow broke at WikiBrains/Unitel Angola
• Demo: Building the same workflow twice, live

Build a functional CV-screening flow in n8n using synthetic data, apply an iterative development loop, and navigate the governance wall covering bias, GDPR, transparency and human sign-off before deployment.

• HR workflow and data rules with synthetic CVs
• Iterative development loop for agents
• Governance wall: bias, GDPR, transparency, sign-off
• Workshop: Team build of a CV-screening flow

Assignment #4: Workflow Automation
Build one of two or three instructor-provided n8n automation options and submit your finished workflow via the platform.

Break down an agent's reference architecture, from LLM and memory to planning and orchestration, explore six core agent patterns, and use sub-agents for context isolation in multi-agent systems.

• Reference architecture
• Six agent patterns
• Sub-agents and context isolation
• Security lessons from OpenClaw and Hermes
• Case study: A financial analyst as a 4-agent composition
• Demo: Multi-agent workflow build in Google ADK

Examine MCP as AI's USB-C-style connector, break down its servers, clients and resources under the latest specification, and connect a live MCP server using Claude Desktop and MCP Inspector.

• MCP v2.0.0: servers, clients, tools, resources
• Wider protocol stack: A2A, AG-UI, AP2
• Risk surfaces
• Demo: Connecting and inspecting an MCP server live

Assignment #5: Simple Agent with MCP Integration 
Configure a simple agent to connect with and retrieve structured information from an external MCP tool interface.

Contrast traditional SDLC with the Agent Development Life Cycle, explore evaluation frameworks like LangSmith and LangFuse, and learn to price and scope agent projects honestly for real clients.

• SDLC vs ADLC
• Prompt versioning, logging and token cost monitoring
• Evaluation frameworks
• AI code assistants for deployment pipelines
• Guest Speaker: Head of Data Analytics & Cloud Architect, NOS
• Case study: NOS's production story at enterprise scale

Assignment #6: The Final Capstone Project
Submit your complete capstone package: a working agent, a 4S Story, an honest ROI assessment, a governance note and a live or recorded pitch.

Explore production deployment scenarios from local servers to serverless, secure agent authentication, build a client governance framework covering accountability and audit trails, and chart your next career move.

• Deployment scenarios
• Securing agent authentication and authorisation
• From prototype to pilot: delegation with control
• Client governance
• The Applied AI Engineering career path
• Capstone Showcase: Volunteers pitch live to the class