Should Software Companies Embrace AI or fight it? — With Asana Chief Product Officer Arnab Bose

Channel: Alex Kantrowitz

Published at: 2026-04-07

YouTube video id: nKyJ67L2xqI

Source: https://www.youtube.com/watch?v=nKyJ67L2xqI

Should software companies like Asana
fight AI or embrace it? Let's talk about
it with Asana Chief Product Officer
Arnab Bose, who's here in studio with us
today in a conversation brought to you
by Asana. Arnab, welcome. Thank you,
Alex. Thanks for having me.
All right, it's great to be here,
especially at this time. So, Asana's
work management software. And you know
what the popular narrative is, if
there's work management software, any
software, one day it's just going to be
vibe coded by somebody sitting at their
console at Copilot or Codex. Mhm. Uh and
then
what happens to Asana? So, let's put the
question to you. Uh can something like
Asana be vibe coded? Question is, why
would you want to? Do you want to go
ahead and focus on um work that's going
to require you to spend time thinking
about security, about reliability, about
99.99% uptime, about depth of
integrations, about how do you onboard
new AI agents onto your human AI
coordination platform?
All of these questions are hard. They
require thought that's way beyond uh
getting to a proof of concept or a demo.
Especially if this is the coordination
layer that helps you get uh output and
outcomes out of the agentic investments
you're making in your company. So, our
thesis is that every company is becoming
an AI-powered company. They want to get
results out of their AI.
Asana's super well-positioned to
leverage our work graph, the system of
context that we bring, that enables
human beings to coordinate, to allow
human beings and AI agents and multiple
AI agents to coordinate with each other.
We've been building this technology for
over a decade. We focused on global
availability, uptime reliability, on
security, on how do you ensure that
these agents get this concept called
shared memory in a way that is secure
for your business.
Does anybody really want to go ahead and
recreate that and take time and energy
away from their core business to go
think through all of this depth of
integration, this depth of uptime, this
depth of performance, this depth of
scalability, or would they rather stay
focused on on achieving their actual
business outcomes? You know, I had this
crazy idea that maybe
there will be other like one AI agent
builds the software and then other AI
agents focus on some of the things that
you talked about like uptime,
reliability, keeping up with the latest
standards.
If AI goes the way that people intend.
What do you think about that?
Well, again, like uh you're spending the
as a customer, then you're going ahead
and spending that many tokens going
ahead and not only building the software
that is helping you run your business in
an ancillary way, but also spending a
bunch more tokens trying to go ahead and
mean making sure that that software is
secure, available, reliable, is
addressing any of the permissions
concerns.
And if
the you know, future is uh the person
who's spending the maximum amount of
token burn in the most efficient way is
going to achieve the best business
outcomes for their particular industry
vertical, their key business, why would
you go focus on something that's not
your key business? Right. Let Asana
focus on that from the perspective of
human and AI coordination because that
is our key business. You can focus on
whatever your business is. Like maybe it
is a travel company, maybe it is a
company that's a financial institution,
maybe it's a company that's a healthcare
institution. These companies are you
know, are not going to be successful if
they end up spending their tokens and
their cost on achieving these outcomes
that are more about coordination between
human beings and AI agents. Okay, that's
a good point. So, it's like sure, go
ahead and attempt it, but
tokens aren't free. Correct.
>> And so, maybe in some world you can
build all the software that you use,
but you're going to spend a lot of money
to do that and to maintain it through
you know, probably less efficient means.
Totally. Like, ask yourself the
question, like, is this a critical core
competency where I want to
differentiate, and is this the real
reason why I'm going to be successful in
my company's mission and vision? If it's
not, and it's more of an enabling
technology, then why go do it? Yeah.
But, another thing that's happening is
as this wave of technology hits,
it takes It allows people to take
software that's built for the masses,
built to scale, and sort of customize it
in a way to their own interests, and
and build the set of tools or or work
within the set of tools that
specifically attack their use case.
Customized software.
Do you see that as something that, you
know, you can embrace, or do you do you
fight that? Oh, I think that's something
we're totally embracing. So, again,
like, as Asana, we've been thinking hard
about our strategy about where we
differentiate, and what value do we
provide to enterprises.
The value provided to enterprises is the
fact that we've been thinking so hard
about this coordination tax for so many
years, like, over a decade. We've been
thinking about how do we ensure that
there's this pyramid of clarity that you
get to by using the work graph, which
defines tasks and projects and
portfolios that clearly ladder up to
company-level goals and mission.
And that framework is something that
we've provably demonstrated works really
well for training AI agents and ensuring
that AI agents have the right level of
business context, as well as
enterprise-wide memory to go get work
done.
And so, again, like, we want to make
sure that Asana is fit for purpose and
works in the best possible way for every
enterprise out there. And customization,
where you can bring your own agent, or
we can customize one of our pre-built AI
teammates, that is 100% part of our
strategy. We want to make sure that
you're having to do as little thinking
about how you want to use Asana and
leverage us to get your work done,
achieve your mission and vision because
we've built this flexible framework that
any AI agent can take advantage of in a
way that gives them this differentiation
of enterprise grade memory,
shared contacts, and not having to go
ahead and sort of relearn things that a
human being has done in the past. Okay,
I want to dissect this and just break it
down point by point.
Talk a little bit about pre-AI, what
people would use Asana for. High-level
60 seconds just explaining what the
product goes. Our key ideal customer
profiles are in marketing,
IT operations, and strategic planning.
So, what they would use us for is
anything that is a cross-functional
project. For example, in marketing like
launching a campaign is a
cross-functional project where the
marketing team needs input from maybe
product and design.
They need to coordinate across
third-party vendors. There's a There's a
campaign brief. There's a bill of
materials. There's a coordination task
to get human beings aligned. People use
use Asana today pre-AI like in pre-AI
use cases to go ahead and keep those
projects running on time. Or from a
strategic operations perspective like
figuring out a launch, figuring out a
launch plan, you know, requires maybe a
Kanban board which has like a bunch of
like timelines. It has a lot of
different functions that contribute to
it. Maybe you need legal sign-off. Do
you Are you sure you've gone ahead and
done that? So, those are the kinds of
use cases that people are using us for
again in in a pre-AI state. Right. So,
that's sort of an internal coordination
engine. Correct. Now, the interesting
thing that the coordinate coordination
engine brings to the table is that not
only does it define who does what by
when, but it also has a track record of
how were those projects completed in the
past, what happened when that that
particular project went off track
because of a particular issue, what was
the remediation action that was taken.
And that kind of data is catnip for
agents cuz agents can go ahead and look
through that history of the work graph
and get a really good sense for, okay,
this particular type of campaign brief
document worked really well for the
historical 2023 or 2022 campaign. This
was the kind of feedback that the other
human beings who collaborated on it
provided. And so, instead of having, you
know, each marketer on your team figure
out the best way to prompt the, you
know, AI chatbot of their choice,
whether it's Claude co-work or
or or ChatGPT, and providing it with a
bunch of like document data or their own
personal data,
you know, when you deploy the campaign
brief writer AI team made inside Asana
and you give it access to your work
graph, it automatically knows how to
crawl that and have the best possible
answer that is hyper suited for your
particular business, for the way in
which your business is run historically,
and you don't need to go ahead and get
every marketer on your team trained on
all of these like new tips and tricks
you're seeing where people are creating
markdown files on their local machine
and they're trying to figure out the
best possible way to not blow up the
context window. We are taking care of
all of that for you. Okay, so we've now
moved from
the vibe coding software to like what
does an AI agent actually do within
piece of software like Asana? Correct.
And so this is way the way that you're
embracing it is saying let's get agents
inside the software. Carl, first just a
definition question. What is a work
graph?
So, think about the way in which Asana
is defined today.
You can create projects, you can have
team of people who have access to the
project. Within the project, there are
tasks or work items that individuals
complete. When a task is completed, the
project moves forward. The project could
be part of a broader set of projects
called a portfolio. There could be a
company-level goal for it. So, this is
the the data model behind Asana. So,
Asana's powered by this sort of concept
of helping teams get work done, helping
teams coordinate, helping teams find
this uh this clarity from chaos. And
there's an underlying data model backing
it, which we call the work graph. And
that's what has powered our
human-to-human coordination features
over the years. And that's what we're
opening up and we have opened up with
our new Asana AI teammates launch so
that AI agents can leverage that work
graph for some of these like key
outcomes I'm talking about, where they
get the right business context, where
they can learn from human beings and how
human beings have completed tasks in the
past, uh where they can create this
concept called shared memory, where
let's say if Alex, you use one of the AI
teammates and you you know, you tell it
that, "Oh, this particular campaign
brief or the risk you've found within
this launch plan is incorrect." As long
as I also have access to that AI
teammate and the same project, if I use
it, that AI agent will remember what
Alex has told it. It will make sure it
doesn't make that same mistake again.
Which is a big difference between the
co-pilots and personal agents you're
seeing in use today. No memory there.
Correct. There's no shared memory there.
It remembers what you tell it. And if
you go back and you ask it to do
something again, it won't make that
mistake for you, but it will make that
same mistake for me if I if I haven't
ever coached it through that. Okay, so
let me see if I can get this right in
terms of what you're building. So, in
Asana, companies coordinate between
departments to accomplish goals.
>> Mhm. And so, I'll let's just use the
marketing example um because it's one
I'm familiar with. I started my career
marketing
>> Yeah. um before making sort of the
reverse move to journalism, which
it's a whole different story, but um
in marketing, what would happen would be
we would be assigned this campaign by a
business owner that we would have to go
create a marketing campaign for. First
thing that begins is that we sort of
like have to you know, get that
information from them and then come up
with a creative brief. Correct. And so
is what would happen within Asana that
you start to build this process within
Asana. You're working across
departments. There's so you're so you're
going to work on and within marketing
the most annoying thing is getting
approvals. That's right. sort of
streamlines the approvals.
But could these AI agents be introduced
in a way that like you start the product
the project, you put some information
about it and then when you think about
assigning who is going to write the
creative brief which again is you know
it's this document sort of Bible
document that shows
who is the audience we're trying to
reach, what's the benefit to them, how
are we going to communicate it?
That once you get the product details in
Asana, you can use an agent to write the
creative brief? 100%. So like the way
you would kick it off is you'd collect
the initial sort of requirements, you
put it into
a doc and you'd assign that doc directly
to an agent.
>> So you could use agent or person.
Correct. You can 100% the agent it would
basically look past your campaigns and
be like these were our successful
marketing campaigns. Let's sort of
create creative brief in line with the
stuff that's worked in the past.
>> Totally. So the agents have access to
the work graph and they can look at your
historical marketing campaigns, the
historical campaign brief documents, the
creative brief documents.
It can also do
deep web search analysis, right? So it
can look across your public campaigns
and your public stories and it will
leverage all of that input to create a
research plan which it breaks out into
sub tasks. When it's breaking out the
research plan into sub tasks, any human
being on the team and any marketer
whether it's yourself or your peers who
have who have access to that particular
project can give it feedback. Can be
like, hey you know what, you're looking
at that creative brief you wrote in
2024, that actually won't work because
we had these issues with it. And so when
you give it that feedback, it will in
real time recalculate its research plan
and update it. And then ultimately
produce a brief uh
either in a Google document format or
Word depending on how you set it up. Uh
and then the human beings can give it
give it feedback on that particular
creative brief.
Um and again, because it's running in
this multiplayer way, it means the
entire marketing team can stay on the
same page. Like you no one is confused
when they get that document about, "Hey,
what was the prompt? Was the research
plan correct?" They can go back into the
task and see what it was. Versus paying
you directly and saying, "I don't
understand how you came up with this
because we agreed on X, Y, or Z." All of
that is documented in Asana. So, it's
reducing that coordination tax as well
in terms of
understanding why the AI output looks
like this.
And then of course there's approvals
already built into Asana. So, yeah, if
you approve of the creative brief, you
can say approve. It'll go to the next
stage. It'll go back to the the business
department or the product department
that asked for that
in a particular campaign in the first
place and they can start taking a look
at it. If you want, they can also have
access to uh all of the work produced by
the AI agent. So, they can also you know
look through the AI agent's work and say
yes or no or give it nudges and feedback
and recalc.
So, what is the wisdom of turning over
some of this mission-critical work to AI
agents? I mean, I think that like let's
say you have a great creative director.
Yeah. Stick with this marketing example.
You kind of want them to you know lead
this positioning. Um and if you turn it
over to AI, isn't there a risk that the
AI will give you sort of like the
average of averages, which is what it
typically does with creative tasks and
like sees everything that's been done in
the world and sort of you know, it sort
of uh predicts what the next most likely
plan would be right in some way. So,
where do you end up getting the
differentiation if you turn some of this
work to agents? So, the the goal is to
reduce the coordination burden and the
burden of uh looking through how you
have done prior projects and tasks
historically to get to a place where the
creative director is then able to apply
100% of their bandwidth to taste and
judgement. Okay. And so that's step one,
which is like it's learning from this
historical set of tasks and and and
creative briefs that your company has
created. So again, it's not it's not
sort of giving you the average of
averages across creative briefs in the
world. It's highly trained and focused
on how you all have done your messaging,
your campaigns in the past. So it should
be getting to an 80% or 90% good state
from that and it's it's not creating
like AI slop. That's that's step one.
>> Okay. The second step is because your
creative director now has more time and
they're not sort of putting together the
bare bones aspects of it, they have more
time to focus on taste making and craft
and they can go out and take that 80%
good or 90% good output and put their
own sort of stamp on it, their own
impression on it to take it to the next
level. And the really interesting thing
with this model within Asana is again,
because you've got shared memory,
whatever feedback your creative director
provides, the agent will remember going
forward. And so the next time it runs,
no matter who runs it, it'll hopefully
get to like 90% and then you're applying
your taste and judgement on the 90% plus
and theoretically you're sort of moving
everything forward, like the quality of
your creative output, the quality of
your risk analysis for your launch
plans, the quality of your mediation
actions, everything gets better and
better and faster and faster moving
forward. Yeah, I think this is an
important point that the context window
matters a lot. Yeah, like without
injecting or including the right amount
of context
>> Yes. For an AI agent, you'll get the
average of averages.
>> Correct. But if you just flood it with
context of stuff that's worked uniquely
for your organization and your use
cases, that's when it starts to instead
of giving you the average of averages,
give you something that's unique and
specified for yourself.
>> Correct. Yeah.
So, what happens to the creative
directors in this case? I mean, I was
speaking with somebody who was talking
about like
teleoperated robots recently. And this
person mentioned that you know, if you
go to robot teleoperation where you're
trying to like have, you know,
person direct robots from, you know,
some outside uh
place, they could like run four or five
robots at a time where it would
typically take four or five people. Mhm.
Uh and it seems like maybe it would take
one creative director to handle, you
know, four times the campaigns if all
this coordination burden is alleviated.
Is that right or am I missing something
here? Yeah, it should elevate the people
who are tastemakers who really
understand the their craft to be able to
do a lot more with their time. Right?
So, they'll they'll just be able to
produce more, they'll be able to produce
higher quality output, they'll they'll
run at a higher velocity.
Uh and so, theoretically for a for a
business, if you're spending, let's say
$100 in human capital just be, you know,
specific, you will start getting more
and more value out of it, right? Uh and
it should be helping you become more
effective, uh more differentiated,
um and just get you get you, you know,
going faster. Yeah. I did like what
Jensen from Nvidia had to say recently
that if you're using AI for layups, you
lack imagination.
>> Totally. Yeah. I think the question
would be the question would be like,
yeah, we'll do more. You know, like this
is Jevons paradox, right? Like you
should be able to go ahead and
achieve more. Perhaps like you're
focused on one particular industry
vertical in one particular geo.
Uh you know, take a look at your human
capital and if it's like producing the
same amount of output as $400 would have
gotten you in the past, well, now you've
got like four x the number of places,
the number of geos, the number of target
audiences you can go after versus the
existing one you were at. All right. I I
do hope that is the way that companies
go about things. I'm
I'm sure there will be some that try the
other way, but they might end up falling
behind the ones that don't.
>> I do want to call out one thing, which
is today I don't think customers are
getting like in general as people have
been
embracing AI they aren't getting that
level of exponential output or maybe the
right way to say it is they aren't
getting the level of exponential
outcomes Right.
>> from their investments because our
thesis is that the models have gotten
fantastic. They are capable of doing
deep reasoning and producing all of this
you know complex logic that creates
well-formatted content or
faster PRs in the in the age of like
code generation and things like that.
But because what we're seeing is there
isn't enough context that's being used
to go ahead and highly customize the
outputs from these models. You're ending
up with this average of averages output
to your to your comment. And the second
thing is because you're getting faster
longer average of averages output the
human beings who are in charge of taste
making are actually slowing down.
Right, they're having to go ahead and
look through reams of content and trying
to figure out okay, how do I apply
taste? How do I apply judgment? How do I
elevate this? And so you're not actually
getting the output that you want. You're
getting you're getting sort of this high
velocity massive reams of text, but it's
not actually moving your business
forward. So this is the critical thing
that from a head of product perspective
at Asana that I'm trying to solve. I'm
trying to ensure that the output that
you get actually drives outcomes. That
the output is highly trained and highly
optimized for your enterprise's specific
memory.
And every time you go and give that
agent feedback it's getting better in a
way where every human being in your
enterprise can leverage that and ensure
that their outputs get better as well.
So, there's like two really, really
important things from a coordination tax
perspective that I'm trying to solve
for, which is okay, instead of like a
trying to
leverage, you know, this 10x, 100x,
1000x rated which this agentic output is
coming at you and try to somehow make
sense of it. Let's make that agentic
output highly specific,
highly optimized, and amazing for you
and in your enterprise. And then
whatever taste and judgment you provide
to it, you can do so in a way where your
entire team benefits, and you're not
having to go ahead and like have these
like little notebooks and docs of like
here's how to like prompt Claude
properly or here's my personal like
recipe for agents.md. It just gets
better for everybody regardless of how
technical they are. Is what the labs
talk about when they say there's
capability over in. It's basically like
we built the models. Now you just have
to figure out a way for you to use it in
a way that
is productive. Totally. Yeah. So, this
the AI teammate launches. This is
happening basically as we speak. Yeah,
it's happening right now. So, it's
generally available for customers who
buy from Asana.
We shipped with 21 pre-built AI
teammates. So, these are AI agents that
can do campaign brief writing or they
can do
IT ticket deflection or they can take a
look at your launch plans and and sort
of
look through those launch plans and help
you stay on task and schedule. But of
course you can also go build your own
and it's as simple as like providing it
with the prompt that sets its behavior
guidance. The really interesting part
about those these AI agents is their
connection with the Asana work graph.
So, they start off with the right
context and this concept of shared
memory or agentic
enterprise memory for these agents where
they can remember
what they were told or what they learned
across human beings utilize them then.
So, just talk to me a little bit about
what you think this looks like in an
ideal world. Like, this all works out.
How does that change people's lives?
It changes people's lives by taking away
the busy work and not just doing that in
a way where you're getting a ton more
content that is mid, that's average, but
you're getting a ton more content that
is highly optimized and specialized for
the way in which you work, in the way in
which your company has set its mission
and values, the way in which your
company runs and what your company level
goals are.
So, that should be elevating the the
ability for the company to hit its
outcome metrics, its true like key
results versus oh, we shipped a lot of
code, but we actually didn't manage to
sell the product or we shipped like five
new campaigns, but they were
undifferentiated. So, getting to the
level of differentiation,
getting the key results, elevating every
human team member to becoming a
tastemaker. Those are the outcomes that
I'm driving for.
I'm very curious how these agents are
onboarded. Like, you said you can build
your own custom one. Yeah, if I wanted
to build Let's say you hadn't built
creative director, I wanted to build
creative director. How do I sort of
onboard that type of experience in
Asana? Absolutely. So, the way you would
do that is
you would like type in a little prompt
that says I want to build an AI teammate
that is that is a creative director.
The builder chat AI agent will come back
to you and be like, "Hey, so what kind
of
you know, projects and portfolios do you
want this creative director to have
access to?" Because let's say you'd
hired a real human creative director
onto your team, you'd probably have an
onboarding plan for them. Like, you'd
probably tell them to
read these documents about the company's
mission and values and look at these
projects and tasks for how the creative
team has worked in the past. So, you're
provided with that context. Okay. It
will automatically then look through
that and and ask for starter tasks. So
it will say, "Hey, can you assign these
tasks to me because this will help me
get better at my job going forward."
Right? So yeah. It will like literally
say,
"I would like to get working on these
things." Yeah. So look through all the
tasks in your projects and it will find
some ones that are open and incomplete.
It will say, "Hey, these starter tasks
look like good ones for me to start
learning on the job." And you can say,
"Yes, go for it." Uh and then as you use
it, it keeps getting better and better
and better with every run. You're not
worried about like runaway AI that all
of a sudden you say, "All right, you can
go do this task." And then the creative
director like tries to become the CEO or
something like that. Well, again, like
there's a lot of checks and balances in
here, right? So uh you know, you you are
in full control of what task it picks
up. Uh it it runs through our standard
approval processes. So
uh it can't like create a bunch of like
content or tasks for other people and
start assigning work out unless the
human in the loop approves. Um and so
there's there's enough checks and
balances to prevent something like that
from happening.
Um or if you wanted it to happen, that's
fully in your control. Like you can say
that, "Hey, I want to have uh an AI team
mate that thinks like my CEO and can
actually critique me
uh based on all of this feedback I've
gotten from the CEO in the past." So
these are all like you know, totally
fine uh use cases. But the human is
fully in control. You know, the human is
in control of what gets approved, what
tasks get assigned. Uh if you actually
believe the project is not complete, you
know, you can you can say so and it will
it will just react to it. Yeah.
How are you working with the leading
model corporations like uh OpenAI,
Anthropic? And how do you choose what
type of models to input?
>> Yeah, so our research and development
team is uh you know, closely embedded
with the frontier model providers and we
test out all of the latest releases
uh across Asana AI. Sorry?
>> They're busy people.
>> They're busy people. Yeah, yeah. Across
Asana AI we use both Open AI and
Anthropic models in particular for the
AI teammates launched. We have chosen
Anthropic's Opus 3.6 model
and that's what we're launching with
right now. That's how it's powered.
It's
the best in our testing and analysis in
our early access and beta time frame.
And so we're
we're really excited about the
capabilities that that that's providing
and you know, looking forward to like,
you know, seeing where the model
providers go next in terms of more
capabilities, more reasoning and things
like that. Yeah, how do you plan for
that? I mean, you sort of you can only
integrate the capabilities that exist
today within these models, but
you see some of the advancement on the
horizon and is it like
we will rebuild it when the model is
more capable?
Well, the good news is the way in which
Asana works,
the real value we're providing again is
with the enterprise weight context and
the shared memory.
And so that becomes instantly more
valuable as the reasoning model gets
better.
So the things that we like the things
that we provide as differentiated value,
those are things that don't need
rebuilding because that's like input
context. So that's like one interesting
angle.
The second thing is we're like
maximalists in our perspective, which
is, "Hey, if the model can't do
something today, like let's say
I'm making this up right now. It
actually can. Like it it can Opus can go
ahead and create really good HTML
previews of what your
updated website should look like if the
campaign launches. So you can tell one
of these agents to create a mock up,
they can do that. But let's say it
didn't do it right now. Our assumptions
from an R&D perspective is that whatever
skill the models lack today, they will
have in the future.
And so to assume that that is coming and
to see like how would we support the
right frameworks and the right human
experiences around that uh should it be
there.
Uh and that level of like maximalist
thinking is is good in this day and age
because you're seeing the advances come
every like 4 to 6 weeks.
You know, as the models get even more
improved, do you be does the
differentiation that Asana have become
more or less important? Like uh hate to
use the buzzword, but like let's say
like reaches AGI and can sort of like
seemingly do everything a human does.
>> Yeah.
Where does that leave Asana?
>> Uh
again, like I think there's some laws of
physics here, right? So, there's like a
law of physics around context windows
and what it can learn and what it
remembers. And so, we are providing
real hard like computer science benefits
uh across those uh those vectors. And
so, I'm very confident that like even if
the reasoning capabilities get better,
there's still going to be extreme value
out of the context graph that we provide
through our work graph as well as the
shared memory concepts. Okay, so it's
nice it's nice to have a head of product
here cuz I can ask you some of these
like nitty-gritty model questions.
>> Yeah. Um
so you mentioned that you use OpenAI and
Anthropic. Um
I guess like one question would be why
not just use Claude Claude is seemingly
built for work, so why spread across two
models?
Uh again, like we need to take a look at
like how uh you know, this technology is
evolving over time. Uh we are not using
OpenAI right now for AI teammates, but
we are using it in other parts of Asana
AI where those models have uh
been proven to be either cost-efficient
or highly performant for those use cases
within our AI studio capabilities or
street AI chat capabilities that we give
away for free. So, there's a wide number
of applications of AI, right? So,
there's very basic things like AI
summarization or reason, which you could
just chat with Asana and get insights
from your personal work and activity,
and so on and so forth. So, there's a
large number of use cases, and we have
to ensure that we're doing the right
thing for our customers, not just from a
performance, not not just from like a
raw horsepower capability, but also from
the perspective of like performance,
time to return the result, deterministic
versus non-deterministic work flows,
and cost, right? So, so when you factor
all of those in, there's a there's a use
case for for multiple models.
But again, to re-emphasize the AI
teammates launch, the teammates
themselves are being powered by Claude.
OpenAI has been on this big enterprise
push. They're really trying to work hard
to win enterprise business. Are you
seeing improvements there? Do you think
they have a chance to challenge for that
type of business? They're very well
funded. They are talking to us a lot.
>> Yeah.
So, I mean, like it's
very hard to predict who if there's
going to be a definitive winner. I don't
think I don't think that's possible to
call any of that at this particular
point in time. So, I look forward to
seeing their advancements going forward.
Yeah. No open source?
Right now, we're not using any open
source models.
And then but again, like I think
you know, I think looking forward to the
future, we're you know, we're we can
consider them.
At some point in time, they might become
good enough where
for certain payloads, it makes sense.
But I would say the Frontier Labs
products are are quite differentiated in
in this in this particular phase of
time. You know, in 2026.
>> March that you're talking here. You
know, that's interesting cuz I once
heard this progression where it was
basically like you start with OpenAI
because that's the
most common one. Then you make your
products interoperable, so you can use
any model within them, and then you
graduate to open source so you can
customize it more. So, there have been
advances on the customization front from
the Frontier Labs that have allowed you
to build on top of them versus open
source. So,
>> It's a good question.
>> happened? Um
so,
again, like our sort of maximalist
thinking is that the Frontier Labs are
going to keep innovating in the level of
reasoning and capabilities of their
models. And so, trying to create these
customizations or adding our own token
weights is not a good idea and is a
waste of R&D resources at this
particular point in time because of the
rate of innovation that we're seeing.
Like, why not just trust that they will
keep innovating in in their space with
the funding they have and the quality of
research talent that they have.
Uh and then just basically evaluate
which of the ones work best for our use
case.
Um
my perspective might change over time,
but like in this day and age I'm seeing
like enough velocity coming out of them
that it doesn't make sense to try and
create like a separate path where I
might just fall behind. That's amazing
because um you know, the conventional
wisdom is that
the Frontier Labs are like 3 to 6 months
ahead of open source. But apparently,
what they're doing is innovating so
strongly that
doesn't make sense to go
>> 3 to 6 months matters, right? Like if
you if you create a fork and you're
always 3 to 6 months behind what your
competition could be doing. You know,
like
there'll be other companies, I'm sure,
who are thinking about some of the
challenges that
we are addressing at Asana.
I don't want to be 3 to 6 months months
behind them. That will be a big problem
for me. So, I'd rather stay at the edge
in terms of model capability and then
ensure that all of my R&D resources are
dedicated towards building the
differentiation that we have with the
human AI coordination experience, the
ability of the Warcraft to provide more
context in a way that doesn't blow the
context window, the ability to do shared
memory, better integrations, better
skills. Like those are the places where
I want to stay in laser focused on
versus trying to like outsmart the
frontier labs.
Smart.
Okay, a couple questions before we end.
What do you think people misunderstand
about AI agents today?
I think that the number one thing people
misunderstand is
um
uh the amount of work required to ensure
that they provide great output and great
outcomes.
Um it's easy to see these demos and be
like, "Oh, wow, like there there's so
many ways in which I could have an AI
chief of staff and it's going to like uh
be amazing and and take care of all the
busy work for my day." In order to set
that up and get it right in a way where
you can actually trust it, it's secure,
it's performant,
uh it has the the right kind of context
so it so it doesn't generate AI slop, is
tricky. And so that's the reason why
like we are entirely focused on that
side of it, which is, "Hey, let's ensure
that
uh the amazing advances that are
happening in the model space can
actually be used for real business work
in a way where the out puts drive real
outcomes versus just velocity of noise.
Okay, last one for you.
Just like think about the future a
little bit in your in your product hat,
so to speak.
Do we have one like master agent in our
lives that like will like handle some of
our personal stuff, our business stuff,
and maybe you know, we just like set it
out to like run these other agents, or
do you think we're like, you know, going
agent to agent in different interfaces?
Um my personal philosophy is like you
probably want like different agents that
are great at doing different things and
that actually have some separation in
their memory. Uh because there's
probably a very complicated challenge
where like let's say your agent is
handling your personal life as well as
your work life. How do you know for sure
because these are non-deterministic
workflows? Yeah. How do you know for
sure that it's not going to end up
leaking information in one way or the or
the other?
>> Really don't want those to mix. Correct.
Yeah. And my wife's a journalist by the
way, so like in some ways like
there are some specific Asana
conversations that I personally have to
keep like abstracted away from her
because that is, you know, material
non-public information. Yeah. Uh and
that's easy for me to do as a human
being because I know like in the you
know, based on the context what to share
and what not to share. But
again, just like you know, how do you
know for sure that an AI agent will get
that right every single time? I think
that might be tricky and perhaps like
separating them out where there are
multiple agents is the is the best
course of action.
>> Yeah, your agent might or her agent
might be optimized to get her scoops.
>> Correct. Exactly. She's got access to
too much. You could get in trouble.
>> Yeah. Um all right, if people want to
learn about Asana's AI teammates, where
do they go? They go to asana.com. Okay.
Yeah. It's all up there. All right.
Arnout, thank you so much for coming on
the show. Thanks for having me. Yeah.
All right. Thanks so much for watching
and we'll be back on the channel with
another video soon.