If you were at the Asana London Work Innovation Summit, or if you looked at the different posts on social media, you saw that AI Teammates was the big announcement recently. We started to play with it and there is definitely potential there!
As soon as the announcement was made and we saw a couple of screenshots, I had lots of questions. The product team was willing to answer those questions, and I’ll share my initial reactions below as well.
Just a little disclaimer, AI Teammates is very young, it’s still in the early days of the feature. So whatever you read below might not be applicable in the future if you come back to that post later.
What is powering AI Teammate? AI Studio credits?
Asana: In our beta period, we’re providing access to AI Teammates for AI Studio Pro customers on a request basis. AI Teammates usage is currently tracked by credits but these credits will be a separate pool that does not affect existing AI Studio credits.
Bastien: Credits being a finite resource, it will be important to have a clean management and people in charge of watching over those credits.
Asana: We’re very open to learning how AI Teammates are used with our beta customers to land on a pricing model that works best for AI Teammates. During beta, only users on the Pro builder list will have the ability to create AI Teammates
Can you pick a profile picture for your teammate?
Asana: Yes you can customize the name and profile appearance of your AI Teammate though we will always have a unique design treatment in the interface to differentiate AI Teammates from human users.
Bastien: surprisingly, if you look at all the demos and the screenshots, Asana never uses someone’s face or name for their teammate. I’ve had a discussion with them about this, and their initial feeling was that it would be harder for the team to remember your teammates. Their initial thinking was that the teammate name should be based on the role it has. On the other hand, my initial thinking was that I wanted to give my teammates a face and a name, but I definitely understand how that could be creepy or hard to remember.
Asana: We’ve seen different users with different preferences and within our own Asana instance there are AI Teammates that are anchored more on a clear, descriptive name while others have opted more for a memorable persona.
Can you have them think « longer » if needed?
Asana: The amount of thinking will be correlated with the complexity of task you assign to the AI Teammate
Bastien: I guess we can trust Asana and the teammate to think for how long is necessary. However, when using chatGPT, sometimes you get a better answer by asking it to think longer. So I’m wondering if this is something we’ll need within Asana as well
Asana: Our goal is for AI Teammates to be able to provide quality answers while not wasting excessive thinking costs.
How long do they take to answer?
Asana: It varies. AI Teammates will give an instant acknowledgement that they’ve received your task by liking the task. They then craft an initial plan, and will write a comment outlining the plan. - this usually takes a minute.. They then go and execute on the subtasks. - This might take 5-15 minutes depending on complexity. AI Teammates are designed for deeper collaboration on work as opposed to a chat tool like ChatGPT for answering quick questions.
Bastien: with tools like ChatGPT now taking longer to think, maybe everybody’s used to wait. So just be careful, do not expect teammates to have a quick chat with you and help you, for example, during a meeting. Because if it takes a minute or several minutes to answer, it’s not going to be a real-time collaboration. So it’s going to get some use to it.
Asana: Definitely, just like how Asana and Slack are used differently, AI Teammates provide a unique way for work to be assigned to AI to execute on.
Can you pick the model it uses?
Asana: Not right now, we automatically optimize for quality based on complexity of task. We may decide to allow this later on if we feel like multiple models are able to do the job. We’ve found only a few of the latest frontier models to be capable of performing the reasoning we’d like AI Teammates to be capable of.
Bastien: One of the key elements when using AI Studio will be to pick the right model in order to have the best output but not use too many credits. We’ll see how important it’s going to be with teammates. Not being able to choose the model is nice because it makes things easier, but at the same time, it turns the teammate into kind of a black box where you can’t fine-tune to consume less credits.
Do you have to assign or mention them to get an answer or do they understand you are talking to them if they are a collaborator?
Asana: Both can work but assign/mention is more predictable. The latter only works if you have made it explicit. Basically, it will only respond to you if you are an approved user or making it obvious you are talking to it (e.g. you are the only one in the task with it, or have been very explicit).
Bastien: What this made me think instantly is whether or not it’s going to consume credit to understand you are talking to it. But I guess that could be done in a very cheap way. Maybe one of the best practices we’ll see will be to remove the teammate from a task once you got the help you needed.
Do they read their documents and resources every single time you need them?
Asana: No, the teammate will only load documents they have access to into their context if there is relevant data within those documents, which will be informed via a combination of contextual search at runtime and memory. Optimizing how AI Teammates consume context and memory is an area of ongoing development.
Bastien: This is going to be key because unlike AI Studio, where you were able to use AI for a very specific role requiring specific documents, when it comes to teammates, you want the teammate to know more and more over time and have a memory that grows. But you don’t want the cost of a question to skyrocket because the teammate became “too smart”.
Can you create a loop where two agents talk to each other and consume all credits?
Asana: right now agents won’t respond to other agents so no.
Bastien: I guess this is not surprising, but at the same time, some people might expect a teammate to collaborate and come up with an answer together. So just keep in mind that it’s not the case (yet).
And again, a little disclaimer coming directly from Asana: “Keep in mind everything is still being evolved so these answers may change!“.
I really want to take a moment to thank the product team for taking the time to put those answers together and allow me to share those early questions and answers with you.
Bastien, Asana Expert
i.DO (Asana Partner: Services & Licenses)
