Progress Bars for Open-Ended Work: How AI Can Estimate “How Far to Done”
Progress bars can orient open-ended knowledge work if AI represents uncertainty, milestones, and changing scope rather than inventing false precision about how close a task is to done.
A progress bar is easy when the denominator is real and difficult when the work can change
Downloading a file has a measurable denominator. Writing a strategy, debugging an unfamiliar system, or researching a market does not. Yet progress displays can still influence persistence. The endowed-progress literature shows that perceived advancement toward a goal can change completion behavior.[1] The design challenge is therefore not whether progress matters; it is whether the interface has earned the right to quantify it. An AI product should avoid presenting “72% done” when the remaining scope is unknown. It can instead represent completed milestones, confidence ranges, unresolved branches, and an estimate that is explicitly conditional on the current plan.
Precision should match knowledge
When the denominator is uncertain, a staged bar such as Discover → Decide → Build → Verify may be more honest than a percentage. Users need orientation, not decorative certainty.
People often accelerate as they perceive themselves getting closer to a reward
Kivetz, Urminsky, and Zheng documented goal-gradient effects in which effort increased as people approached a reward threshold, including in real customer behavior.[2] For open-ended work, an AI-generated progress display can use this principle by creating credible proximal milestones. The danger is manufacturing a finish line that moves every time the user reaches it. If “final review” suddenly expands into five new stages, the system destroys the motivational value of proximity. Progress estimates need scope-change events that explain why the denominator changed.
Moving the finish line should be an explicit event
When new work is discovered, show the cause: new requirement, failed test, added objective, or changed user request. Hidden denominator changes feel like manipulation even when technically justified.
Meaningful progress is psychologically useful even when it cannot be reduced to a percentage
Amabile and Kramer’s research on the progress principle emphasizes the motivational importance of making headway in meaningful work.[3] An AI progress system can therefore surface qualitative state changes: “research question narrowed,” “three risks retired,” “draft now covers all required sections,” or “dependency cleared.” These statements may be more motivating and informative than a numerical bar because they explain why the project is in a better state. The model’s role is to synthesize evidence, not to pretend all progress is linear.
Progress can be semantic, not merely numeric
A useful update tells the user what became known, completed, validated, or unblocked. It can coexist with a bar while preserving the richer meaning of the work.
Goals need enough specificity for a progress estimate to have a denominator
Goal-setting theory describes how specific goals focus attention and effort more effectively than vague “do your best” intentions in many settings.[4] A progress bar for “improve my career” is nonsense until the user and AI define a bounded objective such as “prepare and submit five targeted applications.” AI should therefore negotiate the unit of completion before estimating progress. The product can maintain a hierarchy in which broad ambitions contain measurable projects, which contain milestones and tasks.
Measure the project, not the aspiration
A broad life goal can remain intentionally open-ended. Progress bars belong on bounded subgoals where evidence of completion can be identified without pretending the larger aspiration has an endpoint.
Feedback should direct attention toward the task rather than toward anxiety about the bar
Kluger and DeNisi’s meta-analysis found that feedback interventions can help or hurt performance and proposed that effects depend in part on where attention is directed.[5] Constant percentage movement can become a meta-task: users watch the score instead of doing the work. A better AI system updates progress at meaningful transitions, explains what changed, and offers the next action. The display should reduce uncertainty about the task, not create a new obligation to make the indicator move.
AI can estimate progress from multiple evidence channels instead of one proxy
For complex work, a robust estimate may combine milestone completion, artifact quality checks, dependency status, unresolved decisions, user-defined acceptance criteria, and remaining risk. The weights should vary by project type. A research report may be mostly complete structurally but still fail because evidence is weak; a software feature may be coded but not tested. AI is useful precisely because it can integrate heterogeneous evidence, but the user should be able to inspect and correct the reasoning behind the estimate.
Uncertainty should be visible because discovery can increase the amount of known work
Open-ended projects often look less complete after a good discovery session because new risks and tasks become visible. That is not regression. A trustworthy interface can distinguish “work completed” from “scope understood.” One indicator may rise while another falls. For example, evidence coverage can improve even as the total number of required experiments increases. By showing uncertainty and scope separately, the product teaches users that learning can temporarily move the estimated finish line while still representing genuine progress.
The right progress bar is a navigation instrument, not a promise
The goal of a progress display is to answer three questions: where am I, what remains, and what should I do next? AI can make those answers far richer than a static percentage by generating milestone maps, explaining changes, and recalculating after new evidence. The design should prefer honest ranges and semantic milestones over false precision. When users can see both progress and uncertainty, the bar becomes a tool for orientation rather than a device for manufacturing confidence.
Works Cited
- 01
- 02
- 03
- 04Locke & Latham — Goal-Setting Theory pubmed.ncbi.nlm.nih.gov
- 05
CodeHistory is a living archive. Citations document the evidence used for this edition; later evidence may refine the account.