The Task Intelligence Maturity Model: Five levels from ad-hoc to autonomous.
Use these five proposed levels to discuss ownership, task classification and operating evidence. They are a diagnostic framework, not a measured progression followed by every company or a certification of agent performance.
Why a maturity model, why now
Usage measures and operating evidence answer different questions.
Most AI progress in enterprises gets tracked the wrong way. Seat utilization. Query volume. Number of Copilot licenses deployed. Those are inputs. They say nothing about whether the work has moved. A company can be at 100% adoption of every AI tool on the market and still sit at Level 1 of operational change.
What the board actually wants to know is a different question. Did the work change. Is a task that used to take a person now done by an agent. Is the agentic ratio moving in any function. Did the P&L reshape, or did the company just add a software line item.
The proposed levels organize those questions around ownership, task mapping, testing and operating evidence. They help a team describe its gaps. They do not measure the current maturity of the industry or establish that most enterprises occupy a particular level.
Ad-hoc. Measured. Classified. Orchestrated. Autonomous.
The company has bought seats. Copilot, Claude, Gemini, maybe one enterprise contract. Adoption is a personal-productivity story, not an operational one. Every team reports different metrics, mostly anecdotal. There is no classification of the work and no owner for the outcome. The CEO cannot answer a board question about what changed in the P&L. The CHRO cannot answer a question about which roles will move. Progress comes from individuals, not systems.
- AI success stories are shared in all-hands; none of them show up in the operating plan.
- Nobody can name the three workflows where AI moved a metric.
- The CFO treats AI spend as IT cost, not line-of-business investment.
- HR and IT are both claiming ownership of AI enablement, and nothing is being enabled.
Name a single owner of task intelligence. A VP-level executive with budget and a direct line to the COO or CTO. That one appointment unlocks every subsequent level.
An AI program office exists. Dashboards track seat utilization, query volume, model spend. This is progress over Level 1 because now there is a number. But the number is an input, not an outcome. Nobody can tell you, task by task, what the AI is actually doing. The work is a black box that has 'AI adoption.' Rank-and-file employees are using AI, but the operating model is unchanged. Same roles, same workflows, same accountability.
- Weekly AI adoption reports show usage curves going up and to the right.
- No dashboard shows which specific tasks moved from human to agent.
- Procurement has a view on AI spend; operations does not.
- AI training is running, but nobody can tell you what the training is for.
Stop measuring usage. Start classifying tasks. Pick one workflow, decompose it into its actual tasks, and label each one automate, augment, or human.
The target workflow has a reviewed task map with proposed automation, augmentation and human responsibilities. HR, operations and IT can examine the same assumptions, permissions and evidence gaps. The classification informs capability planning and trial design; it does not demonstrate that an agent can perform the work.
- The COO can explain the proposed task mix and the evidence still needed for a priority workflow.
- Job descriptions for new hires reference task-level classifications, not just skills.
- Capability planning uses the classification as the input, not a hiring spreadsheet.
- When a workflow changes, the classification is the first artifact updated.
Choose a proposed change, agree its checks and run representative trials. Review permissions, exceptions and handoffs before assigning operating responsibility.
Classification is live in the operating model. Agents run a meaningful share of the automate-bucket tasks. Humans have new tools for the augment bucket. The human-only bucket is smaller and sharper, with the people in it having more agent-generated context than they used to. The org chart is starting to look different. Fewer process-executing middle layers. More orchestrators, more specialists at the top, more agent operations at the bottom. The CFO has an agentic-ratio metric, agents per human per function. The P&L shape is shifting.
- A function has moved to a materially different agentic ratio, measured and reported.
- The org has created new roles that did not exist three years ago, such as agent operators or AI workflow owners.
- The middle of the pyramid has compressed without layoffs, through redeployment.
- Monthly business reviews include agent metrics next to human metrics.
Move from single-function orchestration to enterprise-wide operation. Close the feedback loops. Measure agent quality, not just adoption. Build the capability to redesign any workflow in weeks, not quarters.
The company is operating in the new shape. A diamond, not a pyramid. Agents run a majority of the automate layer without incident. Human work is concentrated on judgment, customer intimacy, and the orchestration of complex agent systems. The CEO runs board conversations on agentic ratio by function, on P&L reshape, and on capital reallocation, not on tool adoption. New workflows are born agentic. Task classification is continuous, not a one-time exercise. The operating model is self-updating.
- New workflows are designed agentic-first. No legacy 'before AI' version exists.
- Capital allocation at the board level explicitly tracks agent investment.
- The CFO reports agentic ratio as a standard line in quarterly reviews.
- Onboarding for new employees assumes they will work alongside agents from day one.
You are the reference case other companies are studying. Maintain the classification discipline. Keep measuring. The ratio will keep moving as frontier models advance.
A ten-minute self-diagnostic
Five questions. Answer honestly. Your lowest yes is your level.
Can your COO name, in one sentence, which three workflows AI moved a metric on last quarter?
Is there a single VP-level owner of task classification in your company?
Can you show, for any priority workflow, a task-by-task classification of automate, augment, and human?
Have agents been deployed against a specific automate-bucket task in production, with a measured before-and-after?
Does your CFO report agentic ratio (agents per human per function) alongside revenue per employee?
Your level is the lowest question at which you answered yes. Most enterprises answer yes to Q1 or Q2 and no to Q3. That is Level 2. The honest read is usually one level lower than the leadership narrative suggests.
What to do after you locate yourself
The next move depends on where you are, not where you want to be.
Use the levels to identify the next evidence gap in a selected workflow. Ownership, measurement, design and testing may develop together. Review what is missing before expanding deployment rather than treating the model as a compulsory sequence.
At every level, the single most leveraged action is the next move listed in its card above. Appoint the owner at Level 1. Replace the usage dashboard with a task classification at Level 2. Move from classification to orchestration at Level 3. The moves are specific. That is on purpose.
A practical starting point is one workflow, an accountable owner and a reviewed task map. Record the assumptions and tests needed before moving from a classification to an operating decision.
Common questions about the maturity model
Straight answers from field conversations.