Differences in Execution Results Between Dash and Studio

Asana Community Ask the Community
,
1

When running with the same AI-Instructions, Dash and Studio produce different levels of results.

Execution Details: Check all tasks within a project, create a new task in another project, and write the task titles in a bulleted list in the new task’s description field.

As a result, Dash returns perfect execution results.

Studio has many omissions (even after changing the AI ​​model several times, there are omissions every time) (I tried changing the AI ​​model to Recommended, GPT-5, GPT-5.2, and GPT-5.4).

I consulted Dash about ways to improve it and changed the AI-Instructions, but it didn’t work well in Studio.

Additional information

Studio doesn’t always provide the answers I need, and it consumes credits.
Running Dash every single time is also a hassle.

I was able to achieve the same result using a script—which was created for me by Copilot (the free version). (Dash, for its part, stated that it cannot generate scripts.)

2

Hi,

What you’re seeing is expected for this type of job. Listing every task title is a deterministic task, where 100% completeness matters, and an LLM-based workflow is not the best tool for it. An AI Studio action reads project data within a limited context and number of steps, so on larger projects it can skip items, and changing the model won’t fully fix that.

Recommendations:

  1. Use a script action for deterministic work. You’ve already proven that a script gets the complete list. If your plan includes script actions in rules, run the script from a rule (for example on a schedule or when a task is added). It’s exact every time and doesn’t consume AI credits.

  2. Keep AI Studio for judgement-based steps such as summarising, categorising, drafting or deciding priority, where small variation is acceptable.

  3. If you want to stay in AI Studio, reduce the scope per run. Process one task at a time (for example ““when a task is added, append its title to the summary task””) instead of asking it to read the whole project at once.

  4. Add a self-check to the instructions, such as ““first count the tasks, then confirm the list has the same number of items””. It helps, but it won’t guarantee completeness the way a script does.

Hope this helps you choose the right tool for each step

3

Hi, @Preeti_Singh1 .

Thank you for the detailed advice.

After that, I consulted with Copilot and was able to script the task extraction process.

It’s something like this.

I have been running it continuously for a while now, and things are generally going well.