Five stages of AI. One agent edge.

Most agents are still using stage one of AI to advise clients on real money. Stage two, giving the tool an actual job against your real documents, is where the edge lives right now, and the gap will widen fast over the next few years.

The map of where we are.

Artificial intelligence is moving through five recognisable stages. For an agent, the map is practical, not philosophical. Each stage is the shape of what you can hand to a machine. The agent who knows which stage they're in is the one who stops asking AI for opinions and starts giving it work.

The agent who gives AI a job, instead of asking it questions, will out-research and out-execute the rest of the market for years.

Stage by stage.

01

Stage one: the chat box

Ask a question, get an answer drawn from public training data. Useful for generic questions. No access to your client files, your deal, or your CRM.

02

Stage two: agentic AI

You stop asking questions and start handing over jobs against your actual documents. This is the stage most of this library is built for.

03

Stage three: multi-agent teams

A team of specialised agents delegates to each other overnight: sourcing, research, valuation, negotiation, reporting, without you in the loop on every step.

04

Stages four and five

AGI and superintelligence are stages the frontier labs argue about. Worth knowing they exist. Not worth waiting for. The agents who matter when they arrive are the ones fluent in stage two today.

What it still gets wrong.

i.

The line between stages is debated

Researchers disagree on where stage three ends and stage four begins. Plan around what's usable today.

ii.

Most agents never leave stage one

Treating a chat box like a smarter search engine is the single biggest reason agents dismiss AI as unhelpful for real client work.

iii.

Stage two takes setup

Projects, connectors, and grounded documents are what separate a stage-two workflow from a stage-one question.

iv.

Stage three is arriving, not arrived

Orchestrated multi-agent teams are being built by the major labs now. Fluency in stage two is what lets you adopt stage three fastest.

Next Playbook