2/15
The 7 Levels
of AI Partnership
L1 Ask & answer · L2 Draft & edit · L3 In-tool copilot
L4 Agent — plans, executes, checks
L5 Team of specialists — memory, roles
L6 Autonomous ops — humans at judgment points
L7 Dark Factory — fully autonomous software
15 30 60 🎯 FRAMEWORK · 4 min · Step through levels
There are seven levels of working with AI. Let me walk you through them.
These first three are familiar. You ask AI questions. It drafts for you. It works inside your tools. Most of us live here.
Level 4 is where it changes. You describe a task — AI plans, executes, and checks its own work. This is not autocomplete. This is an operator.
Level 5 — you're directing a team of specialist agents. Each with its own expertise, memory, and role. Frameworks like Microsoft Agent Framework, CrewAI, and Squad make this possible today.
Level 6 — agents run autonomously. They discover, triage, execute. Humans step in at the judgment points. Think autonomous DevOps, AI-powered content pipelines, self-healing systems.
Level 7 — the Dark Factory. Fully autonomous software that builds and operates itself. We're not there yet. But look how close we got.
Now — before I show you something, I want to know where you are.
🔴 SQUAD DEMO — Full Script
What the audience will see
You talk to an AI team. It thinks about what you need, proposes specialists, and when you approve — starts building the project. Files appear in the folders in real-time. Then you jump ahead 6 weeks to show the finished result. The contrast between the half-built scaffold and the polished production site is the wow moment.
Before the talk (pre-staging)
✓
Open terminal (PowerShell 7) — large font (20pt+), dark theme, clean prompt
✓
Run demosquad projectname — creates folder, git init, squad init, launches Agency + Copilot + Squad
✓
Once loaded, press Ctrl+L to clear init clutter (keeps context)
✓
Open file explorer next to terminal — showing projects\demo\projectname (snap left/right)
✓
Type squad as first message — triggers welcome greeting. Leave it waiting for input.
✓
Browser tab pre-loaded with aka.ms/agenticlibrary (hidden behind terminal)
✓
Enable voice-to-text (Windows dictation: Win+H). Test it once before the talk.
✓
Do a full test run tonight. Time it. Know where your pivot point is.
Phase 1 — The Prompt (~60 sec)
1
Narrate to audience: "I have an empty project. I'm not going to write code. I'm going to describe what I want to build — and let a team of AI agents figure out how."
2
Activate voice-to-text (Win+H) and speak this prompt:
🎤 The prompt (speak naturally, one phrase at a time):
"I want to build a curated content library for AI resources.
It should be a website where a community can browse and contribute links.
I want it automated.
New content should be discovered from feeds, triaged by topic, quality-checked, and published — with a human approving at the right moment.
Help me think this through.
What team do I need, what should the architecture look like, and what do we build first?"
Phase 2 — The Team Proposal (~60 sec)
3
Wait. Squad thinks and proposes team members with names and roles (10-20 sec). Stay quiet — let the audience read. The silence is the drama.
4
React naturally: "Look at this. It didn't just list roles — it designed a team. Each specialist has a name, a charter, and will remember what it learns across sessions. It's a thinking partner — helping me design the architecture before a single line of code exists."
5
Say or type: "This looks great. Hire them and start building the foundation."
Phase 3 — Watch It Build (~60-90 sec)
6
Squad creates team files — charters, memory files appear in .squad/agents/. Point to file explorer: "Watch the folders."
7
Squad starts building — project files appear: folder structure, configs, workflow definitions, agent prompts. Don't narrate every file — let the visual speak. Occasionally point: "There's the workflow." "There's the agent definition."
8
After 30-60 seconds of files streaming, narrate: "This is still going. It's designing the workflows, writing the prompts, setting up the automation. I could sit here and watch it finish — or I could let it run and review later. That's how I actually work."
Phase 4 — The Pivot
9
THE PIVOT LINE (rehearse this!): "I did exactly this. Six weeks ago. Same empty folder. Same conversation. I let it build, I reviewed, I iterated — and I kept going. Let me show you where this ends up."
10
Switch to browser. Browse the live site — categories, search, a link entry. "This is in production. Real users. Real agents running overnight. Built from a conversation like the one you just watched."
Fallback plan
A
If Squad is slow proposing the team (>30 sec): "It's thinking — designing the right team for this problem. While it works, let me show you where this conversation leads—" → pivot to site early. Come back to terminal later if it finishes.
B
If Squad proposes team but stalls on building: Pivot after the team proposal. "These specialists are now ready to build. I'll let them work — but let me skip ahead to show you the end result." The team proposal alone is a strong demo.
C
If voice-to-text fails: Type the prompt. You've rehearsed it. Keep talking while typing: "I'm describing what I want — a content library, automated discovery, quality gates..."
Rehearsal notes
!
Do a full test run tonight. Time each phase. Know how long Squad takes to propose, approve, and start building.
!
Practice the voice prompt 3×. Windows dictation needs clear speech, slight pauses between phrases.
!
The silence while Squad thinks IS the drama. Don't fill it. Let the audience wonder.
!
The pivot line is the most important sentence. Rehearse it cold: "I did exactly this. Six weeks ago. Let me show you where this ends up."
!
File explorer is your visual proof. The audience sees an empty folder become a project. Even if they can't read the file names, the tree growing is the signal.
7/15
AI Was a Great Advisor.
Now It Is Also a Great Operator.
- AI can now reason across complex tasks
- AI can now use tools — MCP, custom skills, extensible toolboxes
- AI can now self-correct — check, iterate, and improve in loops
🧠 Understand → 📋 Plan → 🔧 Act → ✅ Check → 🔄 Iterate
Wrapped in: memory · guardrails · human review
15 30 60 ⚡ THE SPINE · 3 min · This is the WHY
So why is this happening now? One shift. AI crossed a threshold recently. Coding agents became good enough to ship real work. The ripple effects are everywhere.
Three capabilities changed: reasoning across complex tasks, tool use through MCP and custom skills, and self-correction — agents that check, iterate, and improve in loops.
The agent loop: understand, plan, act, check, iterate. Wrapped in memory, guardrails, and human review. That's why this is a real system, not a party trick.