Running a social media account may take a year. Handling an Apple developer account can involve repeated follow-up between support replies, account reviews, and payment. Improving a website's performance may mean working on the same set of problems for three months. These tasks have a clear direction, but their progress keeps changing. Someone needs to remember where the work stands and keep arranging the next step.
We are building Sparse Agent to take on this kind of long-term responsibility. The user defines the desired result with a Goal, explains how progress should be checked through a KPI, and establishes the scope of independent work through Constraints. The Agent plans around the current situation, reports regularly, and asks proactively when it needs the user's participation.
The following three scenarios help explain what work is a good fit.
A year of social media operations
Social media operations involve many rounds of work: understanding the product, choosing topics, publishing content, observing the response, and adjusting the next round. Building a useful body of content and a steady readership requires someone to connect those rounds over time.
You can give an operations Agent a one-year responsibility: keep running the product's accounts on Weibo, Zhihu, or X so that people interested in the field understand the product better. Provide product materials, existing content, and working preferences saved in Memory, then agree on the KPI and publishing scope. It now has work it can keep advancing.
The KPI might be monthly verifiable public reach or another account metric relevant to the current goal. Agree on the target, reporting period, and calculation method first. Recording how much content was published helps track the workload; observing reach and reader responses helps assess whether the topics and presentation are producing the intended effect.
Within that scope, the Agent can choose topics, try different approaches, and adjust the posting schedule. When the product gains a capability, it can turn the new materials into future content. If a type of topic receives a weak response, it can keep examining the audience, presentation, and publication time. The user does not need to assign the day's posts again every morning. The Agent plans the next round using existing materials and the latest results.
Our earlier social media operations case used monthly verifiable reach as its KPI. Daily reports tracked cumulative progress, the remaining gap, and upcoming work, showing the user how that responsibility was advancing. The user could also point out content that needed adjustment, and the Agent applied those requirements to later work.
A year is a period of responsibility you can set. Reaching one month's KPI does not end the ongoing operations; the results from each stage can inform adjustments in the next.
An Apple developer account task spanning two weeks
Another kind of work has a clear finish line but involves many waits and exchanges with other people. Apple developer account work is one example we have actually carried out.
On July 16, 2026, I assigned the Agent a task: handle the Apple developer account and complete developer signing for the PythonMiao desktop application within a month. We also agreed on progress reports covering the current status, the next day's plan, and the estimated time remaining.
As the work unfolded, it involved registration and renewal, account activation, support tickets, communication with Apple Support, payment confirmation, and then certificates, signing, and notarization. If a step failed, it had to keep investigating. If Support had not replied, it had to preserve the progress and arrange follow-up. When it needed the user to log in, confirm their identity, or authorize a payment, it had to contact them proactively and explain what was missing.
After the user completed the necessary action, the Agent continued with the following steps. It kept track of where the task was blocked, what had already been tried, and whose reply it was waiting for. It also reminded the user to complete actions that affected progress. The user could delegate routine coordination and participate when a specific question or request arrived.
According to the project's case record, the task was completed on July 29, about 13 days after the assignment. The final checks covered a usable account and membership, along with successful Developer ID signing, Apple notarization, and distribution verification for the software. Much of the elapsed time was spent waiting, but the task still needed follow-up until those results were confirmed.
For this kind of work, the KPI can take the form of a set of completion criteria. The number of support tickets handled records activity. Whether the account works and the final delivery passes its checks indicates how far the goal has progressed. Once the agreed acceptance criteria are met, the specific task can be closed.
Three months of website performance improvement
A development team can also delegate a sustained improvement responsibility to an Agent. For example, it can focus on improving a website's performance over the next three months.
Before starting, identify the website, key pages, or APIs to improve, then agree on the metrics to observe. Page loading time, response time for key APIs, and error rates can all provide directions for assessment. Define the current baseline, desired improvements, and measurement method together, so that later changes can be evaluated.
During that period, the Agent can inspect the running system, use data to find opportunities for improvement, reproduce problems, prepare code changes, and compare the results before and after optimization. After a release, it continues observing actual behavior, checking whether the problem is resolved and whether other functionality is affected, then planning the next round.
An explicit constraint can apply here: submit changes for human review and release them after approval. The Agent handles monitoring, analysis, preparing reviewable changes, and tracking results; people retain the review and release decisions. When it needs additional information, access, or judgment, it raises the issue proactively.
For those three months, it keeps its attention on the same performance goal. After one improvement is complete, it can examine other problems affecting the results. When new data changes an earlier judgment, it can adjust its priorities. Progress reports show the user the current metrics, completed changes, and what it plans to address next.
This is a scenario a team can define: delegating a three-month performance responsibility to an Agent.
All three require continuing responsibility
These types of work differ in duration and how they end. Operations are an ongoing role assessed in stages. An Apple account task can end when its acceptance criteria are met. Performance improvement repeatedly addresses a set of metrics within an agreed period.
They all require follow-up. After a round of work, progress needs to be saved. When circumstances change, the situation needs to be reassessed. When waiting is necessary, the next check needs to be scheduled. When something is missing, the person who can help needs to be contacted proactively.
The Goal keeps these actions focused on the same result. The KPI helps assess current progress, and Constraints establish where the Agent may proceed independently. Schedules and events can provide conditions for starting work, while the next actions also depend on the goal, existing materials, and the latest results.
The user can add information and correct the direction during the collaboration. The Agent's responsibility is to incorporate those changes into subsequent work and keep it moving. Reports, feedback, and necessary participation keep the user informed and involved.
Write a clear assignment before delegating
A workable assignment should make clear what the task needs to achieve, how it will be checked, and where assistance will be needed.
The Goal states the responsibility. Explain whom the Agent is working for, the desired result, and the duration. An ongoing role needs a period of responsibility; a specific task needs acceptance criteria for completion.
The KPI explains how progress is assessed. Operations need a metric definition and target. Performance work needs a baseline and improvement metrics. An account task can list the conditions that must be met. A clear KPI lets the Agent adjust subsequent actions based on results and lets the user check progress through reports.
Constraints establish the agreed working scope. Specify which actions may be completed independently, which content requires review, and which matters need the user to confirm personally. When a login, identity confirmation, payment authorization, or code release requires the user's participation, the Agent should make an explicit request and explain how it will continue afterward.
The Sparse Agent assignment form. The operations goal shown is an example; no Agent has been created.
The reporting schedule can also be agreed at the start. The user needs to see what has been completed, how far the work is from the goal, where it is blocked, and what is planned next. Feedback and new materials feed into later work. Requirements and lessons worth keeping can also be recorded in Memory for relevant work to reuse.
Start with a clearly defined responsibility
A question, a text revision, or a file that can be checked immediately can first be handled in a single conversation. Tasks requiring coordination across days, projects advancing week by week, and roles managed month by month are more likely to benefit from a long-term Agent.
If the goal and assessment method are still unclear, discuss them with the Agent first and organize the task before delegating ongoing work. A goal lasting months or a year also requires an available model, cloud environment, and account authorization, so that later work has what it needs to continue.
Sparse Agent is suited to responsibilities with a clear direction, progress that can be checked, and a continuing need for follow-up. The user provides the goal, KPI, and working scope. The Agent takes responsibility for carrying the work through successive rounds and contacting the user when participation is needed.
Further reading: Real Digital Workers Have Arrived.