The 30/40/30 Pattern: Why every industry converges on the same three buckets.
Across 2.1 million tasks, 894 occupations, and 2,400+ companies, the work inside every industry splits into the same three buckets. Automate. Augment. Human. The sizes of those buckets move with the function. The three-bucket structure does not.
From The Agentic Enterprise (2026), co-authored by Giridhar LV, Kashi KS, and Rajan. Available on Amazon Kindle.
The pattern, stated plainly
Three buckets. Meaningful weight in each. The ratio shifts by function, not the structure.
When you classify every task in a real role against the question "what should AI do with this, specifically," the answers do not distribute randomly. They cluster. Three clusters, in every industry we have measured.
Some tasks can be done entirely by AI today. Others are done faster and better by a person and an agent together. A smaller set has to stay fully human, because accountability, judgment, or trust cannot be delegated. Call them automate, augment, human.
30% automate. 40% augment. 30% human. The number is a mnemonic. The three-bucket structure is the finding.
Tasks in knowledge-work roles lean more automate-heavy. Tasks in frontline and operational roles lean more augment-heavy. Human-only tasks concentrate where accountability, patient contact, or unstructured judgment dominate the work. Every role has all three. No role we have ever classified has only two.
Anthropic found two buckets. Enterprises need three.
The Economic Index measured conversations. We classify tasks.
In November 2025 Anthropic published the Economic Index, an analysis of approximately 4 million Claude conversations. The headline finding: 52% of the work being done in Claude was augmentation, 45% was automation. That is a two-bucket model. There is no human-only category because by definition everything inside the sample was being done with an AI in the loop.
The two-bucket view is correct for its scope. It tells you, among work that is already being done with Claude, how it splits. It does not tell you about work that is not in Claude at all. For enterprise planning, that missing category is the important one. It is the oversight layer. The accountability layer. The judgment calls the board wants a human signing off on.
The 30/40/30 pattern takes the Anthropic split and adds the third bucket. Not to disagree with it. To extend it into a frame a CFO can staff against.
Today's ratio and tomorrow's ratio
The gap between them is the next five years of operational work.
There are two relevant ratios for any organization. The first is what the work looks like today, given the AI that is already deployed in that role. The second is what it looks like if the enterprise adopts the current generation of agentic AI in full. The gap between those two is the transition.
What the work looks like with commodity AI already in place. Augment-heavy. A meaningful human-only residual where judgment or accountability dominates.
What the ratio becomes when the enterprise deploys current-generation agentic AI in full. Automate nearly doubles. Human-only compresses to almost nothing inside knowledge work.
Same structure, different weights
Seven industries, three buckets, ratios that move with the work.
Cut the same data by industry and the three-bucket structure holds, but the weights shift. Healthcare has the largest human-only layer because clinical judgment and patient contact cannot be automated. Financial services and technology lean augment-heavy because most of the work is already digital. Energy sits between the two.
Source: Nuvepro canonical tasks database. 4,807 roles pulled from 152,000+ real job descriptions across 620 companies. Each role classified against the six-dimension framework. Ratios are averages at the role level.
The function decides the bucket weights
Knowledge work compresses human-only to near zero. Frontline work keeps augment dominant.
The cleanest view is by SOC major group, the US Labor Department's occupation taxonomy. Cut the Tier 3 ceiling data this way and a clear story emerges. Computer work, business and financial work, and office and administrative support all converge on 60 to 70% automate. Healthcare, construction, and production stay augment-dominant into the mid-80s. Nothing in the modern economy is 100% automate or 100% human.
At the role level, the numbers get specific
Seven roles, seven ratios. This is what the Tier 3 ceiling looks like in practice.
Aggregates are useful for board conversations. Operational planning happens one role at a time. Pick a role, classify every task in it, and the three-bucket pattern resolves to a specific ratio with specific tasks behind it. A few examples from the Nuvepro platform:
Market Research Analysts sit at 85% automate because almost every task in the role is language-, data-, or pattern-driven, all of which current agents can do end to end. Financial Analysts sit at 45% because the judgment layer, including stakeholder conversations and scenario selection, still needs a human at the wheel.
We did not name the three-bucket idea first
What is new is the measurement layer underneath it.
Versions of "the 30% rule" have been written about for at least two years, in essays on Medium and in analyst notes pointing at the share of tasks AI can take over. The World Economic Forum has published estimates of automatable task share by industry. Anthropic's Economic Index extended the work to observed Claude usage. None of those claims are new.
What Nuvepro contributes is not the naming. It is the data. 2.1 million tasks classified, each one with a rationale. 4,807 roles in the canonical database, pulled from real job descriptions rather than O*NET summaries. 894 O*NET occupations reclassified under a three-bucket framework. Every ratio on this page can be reproduced for any role, any industry, any workflow in under a minute.
The pattern has been described for a while. The measurement layer is what was missing.
Why the three-bucket structure holds
It is not about the capability of the model. It is about the nature of the work.
The temptation, when a new generation of AI lands, is to assume the ratio is about model capability. It is not. GPT-5 and Claude 4 can both clear every task that current frontier AI can clear. That does not move the Tier 3 ceiling very much. What moves it is not model quality. It is task type.
Some tasks are irreducibly human. A therapist holding space for a patient in crisis. A surgeon's hands. A leader telling a team that layoffs are coming. The work is not hard because the pattern is hard. It is hard because the interaction is the work. No agent removes that.
Some tasks are irreducibly hybrid. An analyst presenting scenarios to a CFO. A sales rep negotiating a contract with a long-standing customer. An engineer choosing between two architectural approaches. The AI can draft. The AI cannot close.
Some tasks are irreducibly automatable. Reconciling a balance sheet. Writing a first-draft contract. Updating a forecast model with last week's numbers. The judgment here is already delegated; the agent just removes the remaining typing.
Three structural categories. That is why the three-bucket pattern is structural, not accidental.
What a CXO actually does with this
Three steps. One workflow. 90 days.
Not one function, not one role. One workflow. Order-to-cash, clinical intake, contract review. Something end to end that the company runs every day.
One line per task. Label each one automate, augment, or human. Do not skip the last bucket. The human-only tasks are where accountability lives. If you get the classification wrong, you get the staffing plan wrong.
Automate tasks ship first. Augment tasks ship second, as the tooling matures and handoffs stabilize. Human tasks get fewer people and more support, including the agent-generated context they now have to validate. The roadmap writes itself once the classification is done.
Questions CXOs ask about the pattern
Straight answers. No hedging.