The 7-stage roadmap for human-AI collaboration
Brian's most comprehensive framework for understanding how AI enters knowledge work. Each stage builds on the previous.
Published: June 24, 2025 — Original post
The stages
Stage 1: Simple tasks, "prompt and paste" (2023) Workers use AI for one-and-done tasks: summarizing, drafting emails, translating. They type a prompt, get output, copy it into their real workflow. Low-effort, low-stakes experimentation.
Stage 2: Deeper thinking, true collaboration (2024–2025) AI becomes a strategic thinking partner. Workers upload document stacks, engage in multi-turn discussions, develop strategy documents, analyze data. This is real work, not throwaway tasks. Workers bifurcate into "AI superhumans" and "workers who might get left behind."
Stage 3: The AI watches your screen (2H 2025) AI gains ambient awareness of what's on a worker's screen via tools like Microsoft Copilot Vision or Google Gemini in Chrome. The shift from reactive to ambient. Serious privacy and trust concerns emerge.
Stage 4: The AI uses your computer for you (2026) Computer-using agents (CUAs) operate the mouse and keyboard, navigating GUIs like a human would. Humans still provide scaffolding; AI fills in the gaps.
Stage 5: AI uses your computer without you watching (2026+) Trust improves, workers let agents run independently. Agents get their own login identities. The mental shift: assign tasks to a workspace, not a person.
Stage 6: Multi-agent AI communication (2027?+) Multiple AI agents coordinate, with one agent kicking off another, handling subtasks, returning results. Workers shift from doers to orchestrators.
Stage 7: AI-orchestrated work (2028?+) AI orchestrates large-scale workflows across people, agents, and systems. Humans provide strategy, governance, and goal-setting. Any UIs needed by humans are generated by AI in whatever form makes sense.
How to use this framework
The specific timing will vary, but the trajectory is clear. Most workers are in Stage 2 and experimenting with Stage 3. The top 0.1% are already pushing into Stages 4-5.
The priority is structural readiness—designing for disposability, since tools/agents/orchestrations implemented today will be obsolete in 18 months. Don't build for Stage 7. Build infrastructure flexible enough to handle whatever comes next.
The self-correction
Brian publicly corrected this framework in February 2026. The roadmap describes evolution within the existing model of work. But second brains and personal AI knowledge systems suggest the model itself is dissolving. The top practitioners aren't progressing through stages—they're working in a fundamentally different way. The 7-stage roadmap isn't wrong. It's incomplete. It describes the visible 20%. The frontier work is in the invisible 80%.