The State of Task Intelligence 2026
This Q2 2026 report introduces role and process views of work. The web edition distinguishes dated corpus coverage, the book’s classification framework and illustrative scenarios. Use them to frame questions for your enterprise project, then test the actual work.
Source coverage is not a single study: the published posting corpus and workflow-reference corpus use different records and units. The worked scores below are illustrative, not measured enterprise readiness.
The PDF is the original Q2 2026 edition and has not been revised. It retains older counts and scenario and vendor statements that differ from this page. Use this web edition for the corrected scope; the PDF is not a current readiness benchmark.
5M
Posting-task records
published source coverage
8
Example task mixes
not a sample of deployed enterprises
7,113
Workflow references
separate process corpus
3.1
Mean example score
not an observed enterprise baseline
Task Intelligence: Two Paths
Review the people, agents and connected workflow against their relevant criteria.
People Path
Map a selected team’s responsibilities into tasks, then examine where people, tools and agents contribute. Use this to define changed responsibilities, building work and preparation needs.
Process Path
Follow a selected case through its records, systems, decisions and handoffs. Define proposed changes, permitted actions, exception handling and checks before building.
Use role and process evidence together. A task map can inform a proposed design, but operating impact depends on the implemented workflow and its review burden, errors and recovery. Assess those against an agreed enterprise baseline.
Five planning points
Separate published coverage, proposed designs and evidence from actual operating use.
The book’s task split is a framework, not a universal automation rate
The Agentic Enterprise (2026) uses a 30/40/30 framing for automation candidates, human and AI collaboration, and human-led work, alongside its study of 2,143,500 tasks across six industries, 60 roles and 11 companies. Use the framing as a planning reference, not a measured deployment rate or a target for your workflow.
Connect role responsibilities with the whole workflow
A role view shows who contributes which tasks. A process view shows how records and decisions pass between people and systems. For invoice generation, examine both the billing analyst’s work and the approvals linking orders, delivery evidence and invoice issue.
Enterprise context changes design assumptions
The 2.2 and 3.7 figures are illustrative scenario values for different capability assumptions. They are not observations of deployed enterprises or a measured increase. Data access, rules, integration and evaluation must be established for the actual workflow.
Review the capabilities of the tools you already use
A workflow platform may already provide automation, agents, controls or process analysis. Check its current capabilities and configuration against the work you need to change. This report does not establish that classification is absent from every vendor’s products.
Measure the complete workflow, including review and exceptions
Using AI in one step does not establish better operating results. Check the complete case: evidence preparation, proposed action, review, exception handling and recovery. Compare results with an agreed baseline rather than converting task labels into financial impact.
The book’s 30/40/30 framework
From the enterprise-AI discussion in The Agentic Enterprise (2026), not a universal split measured across all work.
30%
Automate
Tasks considered candidates for software execution under agreed rules and controls. Owners retain operating accountability.
40%
Augment
Tasks where a person uses AI for work such as analysis or drafting and checks the result against the available evidence.
30%
Human-Only
Tasks assigned to people in the proposed design because of judgment, physical requirements or responsibility.
The book’s framing is a planning reference. For a separate capability comparison, review the assumptions in the capability-tier comparison in Explore before applying it to your workflow.
Illustrative NTI task mixes
Worked examples on the 1–5 weighting scale, not measured industry readiness or adoption.
Technology
The example includes code generation and testing. Requirements, defect checks and release approval still need independent review.
Financial Services
The example includes structured financial work with review and regulatory decisions retained by authorized owners.
Retail & E-Commerce
The example combines digital operations with stock, store and customer-facing responsibilities.
Consulting
The example assigns much work to AI support; consultants still need to check evidence and recommendations.
Insurance
The example includes document preparation; coverage and underwriting decisions still need authorized review.
Healthcare
The example distinguishes administrative support from clinical decisions and physical care.
Energy & Utilities
The example includes field and safety constraints that need explicit authority and checks.
Manufacturing
The example includes plant and hands-on work. Its score does not measure business importance or readiness.
A modelled enterprise-context comparison
Illustrative scenario values, not observed deployment results or a forecast.
Tier 1: Today's AI
2.2
Illustrative score under a scenario without enterprise-specific data and system access. This is not a measured baseline for most organizations.
Tier 3: Enterprise AI
3.7
Illustrative score under assumptions about enterprise data, policies and connected systems. Those assumptions need to be implemented and tested; they do not establish capability by themselves.
A scenario difference is a question to test
The numerical difference describes this modelled comparison. It does not show that an enterprise has deployed agents or achieved a corresponding gain. Test whether current records, tool permissions and review rules allow the proposed tasks, including failures and exceptions.
Assess the tools in your workflow
Check capabilities and controls in your actual configuration.
The published workflow corpus contains 7,113 heterogeneous references. A record can describe a vendor capability or process example; it is not a product audit. Evaluate the specific platform version, configured permissions, actions and controls relevant to your project.
ERP / CRM / HCM
SAP, Workday, Salesforce, Oracle
Check what records and actions are available, and who can approve or restrict them.
Process Mining
Celonis, UiPath, ServiceNow
Check event coverage, process variants and whether observations reflect the workflow you plan to change.
Vertical SaaS
Epic, Guidewire, Toast, Veeva
Check the exact agent scope, source evidence, authority boundaries and review requirements.
Market Context
What the broader research says about the state of enterprise AI.
Read each report for its population, observation period and method. Anthropic’s internal-use observations, a proposed workflow design and a controlled operating comparison answer different questions. None establishes a result for your enterprise without relevant evidence.
Methodology
How Nuvepro classifies tasks and calculates NTI scores.
Data sources
- U.S. Department of Labor O*NET occupational data across 894 occupations
- Real job postings from 2,400+ companies
- Industry-standard workflow databases
- Structured task references with their source provenance retained
- AI-generated decomposition, market research, web search, audit history
Posting, occupational and workflow references have separate provenance. Do not add their counts into one claimed reviewed classification population.
NTI formula
NTI = (Auto% x 5.0 + Augment% x 3.0 + Human% x 1.0) / 100
The weights summarize category assumptions. They are not evidence of deployment, business impact or readiness. The illustrated Tier 3 comparison is modelled rather than observed.
Classification criteria
Automate signals
Repetitive, rule-based, data-heavy, low ambiguity, structured input/output, no novel situations, measurable success criteria.
Augment signals
Requires context, some judgment, stakeholder interaction, variable input quality, benefits from AI drafting with human review.
Human-only signals
High ambiguity, interpersonal judgment, novel situations, physical presence, accountability and liability, ethical reasoning.