Nine questions, five minutes. You get your stage on the Agentic Curve and see which of the seven decisions before the first agent are still open for you. No email required.
How do you measure whether an AI initiative works?
04
Who is accountable when an agent acts wrongly?
05
How do you decide what an agent may do on its own?
06
What do you compare an agent’s performance with?
07
What happens when the model, the vendor or an interface changes?
08
Who holds the obligations under the AI Act, the CRA and the Machinery Regulation?
09
How do you choose the implementation partner?
9 question(s) left.
The model behind the check
The Agentic Curve
The Agentic Curve is a four-stage maturity model for agentic AI by Kai Hackbarth: AI-Aware, AI-Enabled, AI-Native and AI-First. It places how far AI agents already act on their own in a company’s processes, from the first conversation with no one owning a number to a business designed around agents. Each stage has its typical trap and one highest-leverage next move; the check places you on one.
Stage 1
AI-Aware
The conversation has started, but no one owns a number yet.
Trap: innovation theatre — demos, no owner, no metric.
Next move: name one business-owned problem with a number.
Stage 2
AI-Enabled
AI is paying off in a few workflows, but it only assists.
Trap: the copilot plateau — humans do every step, and the gains stay small.
Next move: from assistance to delegated action (one task, an approval gate).
Stage 3
AI-Native
Agents run real, multi-step processes end to end.
Trap: pilot purgatory — works in demo, fails on edge cases; trust never builds.
Next move: build the reliability layer (evals, guardrails, escalation).
Stage 4
AI-First
The business is being designed around agents.
Trap: a local maximum — optimising the old model instead of redesigning it.
Next move: redesign a core value-chain process with agents as the default.
Let’s have a conversation.
The introductory call is confidential, with no obligation.