The biggest roadblock to scaling AI is not the technology. It is us.
I have been asked repeatedly, since starting to publish on AI-native org design, for real cases and concrete learnings — what actually works, in what order. One of the most useful frames I have come across is the “Steps of AI Adoption” described by Boris Cherny, the creator of Claude Code at Anthropic. It maps how organisations genuinely scale from zero agents to a thousand and more, and it resonates closely with what I am writing in The Living Organization.
What makes the ladder valuable is not the taxonomy. It is that at every step, your role changes — and so does the bottleneck. The constraint migrates: from your attention, to your review capacity, to trust itself. Most organisations get stuck because they try to climb a step while still managing the previous step’s bottleneck.
Step 0 — Gated
Blocked by red tape. The tools are not approved, the data cannot be touched, the policy has not been written, and the security review is queued behind eleven other things. Nothing is happening, and the reason has nothing to do with capability.
This step is more common in large enterprises than anyone admits publicly, and it is frequently mislabelled as caution. It is worth distinguishing between the two: real caution produces a decision with conditions attached. Gating produces no decision at all, indefinitely, while the organisation waits for a level of certainty that will never arrive. The bottleneck here is purely organisational — and it is the cheapest one to fix, because it costs a decision rather than a capability.
Step 1 — Assisted
One human, one AI. The familiar pattern: you write the prompt, you get the draft, you edit it. The AI is a better tool in an unchanged job.
Real gains appear here, and this is where the majority of organisations currently sit — often mistaking it for the destination. The bottleneck is your attention. You can only work on one thing at a time, so the AI can only ever make that one thing faster. This is why licence-based ROI calculations at this step consistently disappoint: you have bought a faster tool for a serial process.
Step 2 — Parallel
You become an orchestrator. Several agents work simultaneously on different pieces, and your job shifts from doing to directing — setting up work, checking in, integrating results.
This is the first genuine change in the nature of the job, and it is the step most people find psychologically difficult. The satisfaction of craft — making the thing yourself — goes away, and is replaced by something that feels, at first, like doing less. The bottleneck moves to your review capacity: you can now generate far more work than you can meaningfully check.
Step 3 — Supervised Autonomy
You are managing managers. Agents run longer chains of work with less intervention, and you supervise outcomes rather than steps.
Every experienced manager already knows the skill this requires, because it is the same one: the shift from checking the work to checking the result, and the discipline to intervene on pattern rather than instance. The bottleneck here is trust — and trust is not a feeling to be talked into. It is built the way it is built with people: through calibration, visible track record, clear escalation paths, and the knowledge that failures will surface rather than hide.
This is precisely where a governance instrument earns its keep. Deciding in advance what an agent may do autonomously, what needs a human in the loop, and what must never leave human hands is what makes trust a design decision rather than a nerve.
Step 4 — AI-Native
Steering by intent. You describe what should be true, and the system organises itself to get there. Structure forms around the work rather than the work being pushed through a fixed structure.
Very few organisations are here, and I am wary of anyone who claims to be. What is worth noticing is that this step is not a technology milestone at all. Nothing about it requires a model that does not already exist. It requires an organisation in which authority is genuinely distributed, purpose is legible enough that intent can be stated, and accountability survives without step-by-step supervision. Those are org design problems, every one of them.
Why the ladder is really about us
Look at the sequence of bottlenecks again: red tape, attention, review capacity, trust. Not one of them is a model capability. Each is a property of how the organisation is built and how the people in it are asked to work.
Which is the whole point. You cannot buy your way up this ladder. Every step is a change in the human role, and organisations that skip the human change and install the technology anyway land back at Step 1 with a larger licence bill.
Where does your organisation actually sit on this ladder — and which bottleneck are you still managing?
Whether you lead thousands or a team of one, the ladder is the same. So is the invitation: to build organisations, and a working life, that flourish.
Frame credit: Boris Cherny, creator of Claude Code at Anthropic, “Steps of AI Adoption”.
📚 Start here: AI-Native Organization Design — the full research hub, with every article and video in one place.
More in this series
- 67% of AI Adoption Blockers Are Organizational, Not Individual
- $2M Revenue Per Employee: The Case for the Living Organization
- The Living Organization: Five Pillars of an AI-Native Operating Model
- Why Hierarchy Existed — and Why AI Just Removed the Reason
- Is HR Ready to Design Human and AI Agent Productivity Together?
- How Much Does a CEO Really Need to Understand About AI?
- Your AI Isn’t Failing — Your Specification Is
- Stop Measuring Tomorrow’s Talent With Yesterday’s Rulers
Watch the full series
This article accompanies a video from AI-Native Operating Model and Org Design — a series on how organisations actually absorb AI, and where they break.
- ▶️ Watch the full playlist — AI-Native Operating Model and Org Design
- 📺 Subscribe on YouTube — @BreezeDONG
Chunfeng “Breeze” Dong is an executive coach (ICF PCC, CPCC) and founder of Springbreeze Ventures, with twenty years in organisational development inside Fortune 100 companies — Roland Berger, Siemens, ABB and Roche. She writes on AI-native organisation design, human–agent governance and change leadership.
📘 The Living Organization · 📘 A Soulful Transition · 🔗 LinkedIn
