Our relationship with technology is shifting. We’re moving from users of tools to collaborators with (and managers of) autonomous systems. Understanding this progression is crucial for how we structure work in the future.
I think about AI systems evolving along two fundamental axes: Capability and Agency.
Capability: what can the system do? How complex are its tasks? How sophisticated is its output?
Agency: How much autonomy does it have? Can it operate and make decisions independently, without step-by-step human direction?
Mapping AI's progression across these axes helps clarify its journey from basic tools to something potentially much more transformative. Here’s how I see the stages:
1. Technology
Early ML algorithms like clustering or anomaly detection have limited capability and no agency. They are useful for finding patterns or flagging outliers, but essentially they are sophisticated calculators requiring heavy human guidance.
2. Tools
Think of features like Magic Eraser or advanced photo editing filters. The capability sees a real jump – understanding image context to remove objects cleanly is complex. But agency remains low. You point, you decide, the tool executes a specific, user-initiated task. It’s a powerful lever for human intent, not an independent actor.
3. Appendage
This is where tools like NotebookLM or advanced research aids sit. Capability is high – summarizing documents, synthesizing information, drawing connections. Agency inches up slightly. They might surface unexpected insights or suggest relevant questions based on your inputs, acting like an extension of your own cognitive process. Still, they operate entirely within the bounds you set.
4. Assistant
This is where agentic AI concepts become relevant. You give it a goal ("Find the top 3 competitors in X market and summarize their strategies"), not just a command. The AI determines the steps, executes searches, maybe even interacts with other systems. Capability remains high, but agency takes a significant leap. It's moving from being told to do a thing to figuring out what needs to be done and then doing it. Gemini Deep Research is a good example, TypingMind with extensions is another example of an assistant.
5. Team Member
The next logical step, as I see it, is AI evolving into a contributing team member. Imagine an AI not just taking well-defined tasks, but identifying needs, proposing actions, collaborating with human counterparts, and taking initiative within a specific domain. Its capability is specialized; its agency allows semi-autonomous operation, requiring oversight more like managing a junior team member than using a tool.
6. Team
Tasked with complex, overarching goals, these AI teams could manage projects, allocate resources internally, and deliver outcomes, requiring only high-level strategic direction from humans. This represents peak capability and agency, fundamentally altering organizational structures.Why this matters: Democratizing Capability
The diagram above reveals an underlying trend: the democratization of capability. As AI systems move up these axes they make previously complex, expensive or expertise-gated abilities accessible to a much wider audience.
The playing field for creation, analysis, and execution gets leveled. Individuals and smaller teams can suddenly access capabilities previously reserved for large organizations or highly specialized professionals.