1. What has been completed when a weekly report is finished?
“Turn these materials into a weekly project report.”
AI reads group chats, meeting notes, and several spreadsheets, then produces a clearly structured, well-formatted document. What previously took two hours now takes ten minutes. Judged against that instruction, the task is complete and efficiency has improved.
But why is the report needed?
A manager may need to identify projects likely to be delayed so resources can be coordinated. A customer may need an update on delivery progress. Or the report may simply follow an established reporting routine. The same document format can serve different purposes.
If the manager really needs timely detection of delay risks, compiling materials every Friday may already be too late. If key facts were never recorded, even excellent summarization cannot reconstruct reliable project status from the materials.
“Producing a weekly report” and “enabling the manager to address project risks in time” are completion criteria for different scopes of work.
Before discussing whether AI can help people complete tasks, we need to establish where we draw the task boundary.
2. The tasks users describe have their own origins
Tasks often appear as concrete actions: writing documents, filling in forms, sending messages, or approving requests. These are easy to assign and check for completion, but usually sit within longer work processes.
A project manager writes weekly reports because they have responsibility for project coordination. Procurement staff follow up with a supplier because a batch's arrival date affects production. Customer service staff organize complaints so problems can be addressed and areas for service improvement identified.
Tracing upstream leads to why the organization needs this function, what commitments it has made to whom, and what remains missing between the current situation and the requirements. Following that chain down to specific activities brings us to the actions a particular person performs today with particular tools.
Existing conditions also shape those actions. Information scattered across systems requires copying and aggregation. States that do not update promptly require repeated inquiries. Responsibilities that are not explicit require searching for the right people.
A task list therefore contains both work required by the business itself and work added to compensate for current organizational arrangements and system capabilities.
Existing tasks are important clues for understanding work, but both their existence and their form need to be explained in the environment that produces them.
3. “Better” depends on the goal the task serves
Understanding a task's origins shows that doing it “better” involves more than speed.
For project risk management, faster report writing reduces compilation costs; earlier risk detection may buy time to adjust plans. These improvements occur at different points and need separate evaluation.
Local efficiency gains can also increase burdens elsewhere. If AI generates a long project report every day, the writer saves time while readers spend more effort filtering it. More output does not necessarily improve coordination.
Evaluation therefore needs to consider the whole work process: whether information is trustworthy, whether recipients can act on it, how much additional checking and correction cost it creates, and whether the intended outcomes improve.
“Helping people” also requires identifying whom we are helping. Applicants want fewer fields to fill in, approvers need sufficient material for judgment, and those carrying out the work need complete information. A single design must consider these connected needs.
Personal tasks are similar, though their goals may be more open-ended. Someone asking AI to help write an article may want to publish quickly or may want to clarify their thinking through writing. Generating a finished draft directly helps the first goal but may bypass the most valuable part of the second.
Completion criteria depend on the goal a task serves; the process itself may be part of that goal.
4. Understanding a task provides a basis for redesigning it
Return to the weekly project report.
If project status is continuously updated, relevant dependencies connected, and managers can directly see risks and arrange responses, the tasks of asking everyone for progress, copying it into a spreadsheet, and compiling a report may change.
Some steps can disappear, others can be combined, and some remain worth keeping. A system may continuously aggregate progress, while the manager still needs to explain important changes, resolve conflicts, or confirm revised delivery arrangements with customers. External communication may still require a reviewed progress report.
These changes should follow from what each activity contributes and what remains necessary under the new conditions.
There are two different scopes of design: making a step easier within an existing way of working, and reorganizing work so the goal is reached by a more appropriate path. The first is usually easier to begin with a single instruction. The second requires understanding upstream and downstream work, as well as the corresponding decision authority and collaboration.
If AI identifies potentially unnecessary work, it can help present supporting reasons and alternatives. A change to organizational processes still needs confirmation from someone authorized to decide.
We therefore cannot predict which tasks will disappear simply by looking at how much time people spend creating presentations, working in Excel, or entering information into systems. Presentations and Excel spreadsheets are formats in which work is performed, and the same format can serve different tasks. A presentation can support a sales proposal, training, or a business decision; a spreadsheet can support scheduling, account reconciliation, or procurement calculations. Their information needs, judgments, and completion criteria differ.
AI can speed up producing these artifacts, and new system capabilities may replace some of them. But judging how work will change requires tracing the specific tasks they support. After tool operations are reduced, the underlying analysis, coordination, and decisions may remain, or may be redistributed between people and systems.
Understanding why a task exists provides a basis for deciding whether to accelerate, redesign, or eliminate it.
5. Agent design requires an explicit scope of work
“Build a project management agent” does not, by itself, define the scope of work.
The agent might only compile weekly reports, collect and verify project status, or continuously identify risks and recommend responses. These scopes require different information, permissions, and collaboration mechanisms, with different success criteria.
If it only generates reports, we should check fidelity to the source materials. If it identifies risks, it needs project commitments, current states, and key dependencies. If it can also move responses forward, we must define which actions it can execute and which situations require a decision from whom.
Clearer task boundaries make it easier to decide what tools to give an agent, when its work is complete, and where people need to participate. Otherwise, a system may keep acting toward an ambiguous goal while users cannot judge how much responsibility it has actually taken on.
These questions also concern model training and evaluation. Execution following a given instruction is one category of evaluation. Understanding task context, detecting missing conditions, and recognizing when clarification is needed are other capabilities worth evaluating.
An enterprise digital twin provides a concrete setting for further analysis. Once organizational functions, goals, responsibilities, business objects, and current states are explicitly represented, we have an opportunity to reason about why tasks arise, how far they need to be carried through, and who can undertake them. The next article will follow this line to discuss how tasks arise from the business.
To help people complete tasks better with AI, we need to understand both how the work is done and why it needs to be done.