What to actually ask an AI assistant about your workload
By The Serena Team
Most first attempts at using an AI assistant for work planning go the same way. You ask what you should focus on today, get back a confident and reasonable-sounding ranking, notice that it does not account for the thing you promised a colleague on Tuesday, and conclude the tool is not useful.
The diagnosis is half right. That question is a poor fit, because prioritization depends on things the assistant cannot see: political weight, how much credibility you have with a particular stakeholder, whether a deadline is real or performative. But there is a category of question these tools answer genuinely well, and it is the category that requires reading across more material than you can hold in your head at once.
The split that determines whether a question works
An AI assistant with access to your tasks, notes, and calendar has one strong advantage over you: it can read all of it at once, without fatigue and without the recency bias that makes last Thursday feel more important than last month. It has one structural disadvantage: it does not know what matters, because most of what determines importance was never written down.
So the questions that work are retrieval and synthesis questions, where breadth of reading is the hard part. The questions that fail are judgment questions, where the hard part is weighing things that only exist in your head. A useful heuristic: if a diligent new assistant could answer it correctly after a week of reading everything you have written, an AI can probably answer it. If they would need to have sat in the meetings, it cannot.
Pattern one: find the inconsistencies
This is the highest-value pattern and the most underused. You are asking the assistant to cross-reference things that live in different places, which is exactly the work that reliably slips.
- Which commitments in my meeting notes from the last month never became tasks?
- Which of my tasks reference a deadline that has already passed?
- Where do my notes mention waiting on someone without a corresponding follow-up?
- Which projects have had no activity in three weeks but are still marked active?
- Are there tasks that appear to duplicate each other in different words?
These have verifiable answers. You can check each result in a few seconds, which means a wrong answer costs you almost nothing and a right one saves a genuine dropped ball. That asymmetry is what makes the pattern worth leaning on.
Pattern two: reconstruct a history you have lost
Coming back to a project after three weeks away is expensive. So is being asked for a status update on something you have not touched since April. Both are reading problems.
- Summarize everything recorded about this project, in order, and tell me what the last decision was.
- What changed on this project between the start of June and now?
- What was the reasoning behind the approach we chose here?
- Which open questions from earlier in this project were never resolved?
The last one is quietly the most valuable. Unresolved questions do not announce themselves; they sit in old notes while the project moves past them, and they tend to resurface at the worst possible moment.
Pattern three: break down work you are avoiding
Tasks stall for two reasons: they are genuinely blocked, or the first step is unclear. Decomposition addresses the second, and an assistant is a reasonable tool for it because a mediocre breakdown is still easier to edit than a blank page.
- Break this into steps where each one could be finished in under thirty minutes.
- What is the smallest first step that would make progress here visible?
- What would need to be true before this task could be started?
- This has been on my list for two months. What might be making it hard to start?
Pattern four: check your plan against arithmetic
Assistants are unsentimental about capacity in a way that people planning their own week are not. This makes them useful for a specific, narrow check.
- Given the meetings on my calendar this week, roughly how much unscheduled time do I have?
- I have committed to these five things this week. Which is most likely to slip, and why?
- Which of these tasks depend on someone else responding first?
- What did I plan last week that did not get done?
Note the framing on the second one. Asking which item will slip invites analysis of dependencies and size. Asking which item is most important invites a confident guess about your priorities, which is the failure mode from the opening paragraph.
Failure modes worth knowing about
These tools fail in characteristic ways rather than randomly, and knowing the patterns makes the output much easier to use.
Confident answers built on partial context
If the assistant can only see some of your work, it will answer as though it can see all of it, without flagging the gap. Ask what it looked at when an answer surprises you.
Plausible invention
Details can be fabricated in a form that reads exactly like the real ones. This is why verifiable questions are safer than judgment questions: you can check a date or a task reference immediately.
Agreeableness
Ask whether your plan is reasonable and you will usually be told it is. Ask what is most likely to go wrong with it and you get something more useful. Frame questions so agreement is not the easy answer.
Recency weighting
Recent items tend to dominate summaries regardless of importance. When something older matters, name the time range explicitly.
None of these make the tool unusable. They do mean the sensible division of labor is stable: let the assistant read broadly and surface what it finds, and keep the deciding for yourself. That split also happens to be the one where the output is easiest to verify, which is not a coincidence. Serena AI is built around it: read across tasks, notes, projects, and the calendar to surface what is there, and leave the deciding to the person.
Key takeaways
- Retrieval and synthesis questions work; judgment questions do not, because importance is mostly unwritten.
- The highest-value pattern is finding inconsistencies across places, since every answer is cheap to verify.
- Use it to reconstruct project history, especially open questions that were never resolved.
- Treat generated task breakdowns as drafts to edit, not plans to follow.
- Frame questions so agreement is not the easy answer: ask what will slip, not what is important.
Frequently asked questions
Should I let an AI assistant reorganize my task list automatically?
Not without review. Reorganization involves exactly the judgment calls these tools are weakest at, and the cost of a wrong bulk change is high because it is tedious to unpick. Ask for proposed changes and apply the ones you agree with.
Why does it recommend things I have already decided against?
Because the decision and its reasoning probably were not written down anywhere it can read. If a rejected approach keeps resurfacing, recording the reason once in a project note usually resolves it, and is worth doing for human colleagues regardless.
Is it worth the setup time if I only have a small task list?
The value scales with how scattered your work is, not how much of it there is. If everything already fits on one page you read daily, an assistant has little to find. If commitments live across notes, chat, and a calendar, the cross-referencing pattern pays for itself quickly.
What should I never delegate to it?
Anything where being confidently wrong is expensive and hard to check: final prioritization calls, communications you have not read, and irreversible bulk edits. The useful boundary is verifiability rather than importance.