Tasks. Evidence.
Human judgment.
We show how the work changes, where AI contributes, and what a person still needs to decide. No job-replacement scores.
Why we look at tasks
A role combines many activities, contexts, and decisions. Looking at an individual task makes it possible to show where AI enters a workflow and what a professional still needs to do.
Version 1 deliberately covers only Software Engineering, Data Analyst, Project Management, and Product Management. They provide contrasting forms of technical, analytical, delivery, and customer-focused work. This is a bounded educational scope, not a ranking of occupations.
Ten recognizable tasks per role were selected from the supplied editorial scope and checked against related occupational descriptions and professional guidance. The selection is representative, not exhaustive. Product Management has no one-to-one occupational mapping here; the market-research reference supports only that part of the role.
How to read the two workflows
A representative sequence before widespread generative AI assistance. It may include IDEs, search, spreadsheets, automation scripts, statistical software, and collaboration. It does not describe a world without technology.
An illustrative sequence using documented classes of AI assistance such as drafting, synthesis, and generated code. Availability is not the same as adoption. Tools, teams, access controls, and practices vary, and these sequences are not measured averages.
What the evidence labels mean
Established Practice means a routine assisted action is directly documented in mature professional guidance. Growing Practice means sources support the capability but how it is integrated varies. Emerging Practice means an exploratory or especially context-dependent extension with limited direct workflow evidence. Labels are editorial judgments, not scores or adoption measurements. Current task examples use Growing and Emerging labels; a filter may have no matching records.
How we identify human decisions and skills
At each generated output or action, ask what can go wrong, who has the relevant context, what independent evidence is needed, and who has authority to accept the result. Mark the points where someone must verify, interpret, decide, prioritize, approve, communicate, or escalate.
Compare the work needed to produce an initial output with the work needed to specify, evaluate, and integrate an AI proposal. The shift adds verification and collaboration demands; it never makes foundational skills obsolete.
Sources and review process
Prefer occupational references, peer-reviewed and university research, professional organizations, official documentation, and credible industry research. Source notes distinguish task baselines, available capabilities, educational frameworks, and wider research. We do not infer universal practice from a vendor demonstration.
Each page displays the month its source alignment and learning example were reviewed. This initial edition was prepared in September 2026. Future revisions should check changed tool capabilities, broken sources, and substantive corrections, then update the review date and update log. No automatic research monitoring is running.
What this guide can — and can’t — tell you
This resource is an editorial educational synthesis, not a systematic review or a longitudinal workplace study. Examples are fictional and simplified. It does not represent every sector, geography, organization, or tool configuration. Source references do not imply endorsement, affiliation, or a university partnership.
The website does NOT estimate job replacement, automation probability, or AI exposure. It has no risk scores or replacement predictions.