Ep #7: Going AI-first as a design org with Avi Ashkenazi, Senior Design Director at Deel

WEBVTT
NOTE
This file was generated by Descript
00:00:04.577 --> 00:00:06.277
Tom: Welcome back to
the Near Future Podcast.
00:00:06.637 --> 00:00:10.067
Today, we have our first
ever guest, Avi Ashkenazi.
00:00:10.627 --> 00:00:14.367
Avi is a long-time design leader and
builder who's worked at companies like
00:00:14.427 --> 00:00:19.277
Shopify and now Deel, where he focuses
on fintech, AI, design systems, and more.
00:00:19.937 --> 00:00:23.217
Johnny and I recently saw him speak
at Product Unleashed, an AI and design
00:00:23.217 --> 00:00:25.917
event in London, and spoke to him
afterwards about building software with
00:00:25.917 --> 00:00:28.857
AI to scale how he and his team operate.
00:00:29.287 --> 00:00:31.757
We're really excited for this
conversation and hope you enjoy it too.
00:00:33.578 --> 00:00:34.388
Avi Ashkenazi: my name is Avi.
00:00:34.428 --> 00:00:37.168
I'm a senior product
design director at Deal.
00:00:37.218 --> 00:00:42.998
I work across fintech, AI, design systems,
our mobile areas, and our homepage.
00:00:44.068 --> 00:00:50.798
Uh, I live in London, uh, I've been
doing this for maybe 16 years, 17 years,
00:00:50.808 --> 00:00:54.770
something like that I've been working
on a really interesting tool that is
00:00:54.770 --> 00:00:58.950
called Vidit, and does what it says.
00:00:59.280 --> 00:01:02.630
it allows you to vid it, videos.
00:01:03.120 --> 00:01:07.860
And the reason why I built it was
because, A, we pay at Deel quite a
00:01:07.860 --> 00:01:13.460
lot of money historically for Loom,
and B, because a lot of the designers,
00:01:13.460 --> 00:01:18.740
and there are 130 designers at Deel,
marketing people, wanted Screen Studio.
00:01:20.510 --> 00:01:23.440
Screen Studio costs some money,
but the company did not want
00:01:23.450 --> 00:01:25.180
to pay that amount of money.
00:01:25.620 --> 00:01:30.470
And I really like Screen Studio, but
I was like, "I can do something like
00:01:30.480 --> 00:01:33.440
it, something maybe better with all
the features that I always dreamt,"
00:01:33.440 --> 00:01:36.420
and I'm looking at the roadmap all the
time hoping that they would build the
00:01:36.420 --> 00:01:38.270
thing that I wish they would build.
00:01:38.950 --> 00:01:43.910
I bas- basically decided to embark
on this journey and build a backup,
00:01:44.180 --> 00:01:46.480
and that's essentially what I built.
00:03:49.447 --> 00:03:52.977
Jonny: So th- this is quite a complex
product that you've built, Abby.
00:03:52.977 --> 00:04:00.707
This isn't, a sort of a, quick tool to
nail an easing graph or, whatever it is.
00:04:00.707 --> 00:04:02.557
This is, this is-- It looks fairly…
00:04:03.267 --> 00:04:06.437
for those people listening, we
basically looked at something that
00:04:06.447 --> 00:04:08.307
looks very much like Screen Studio.
00:04:08.747 --> 00:04:12.067
it's like a dark,
beautiful dark UI with…
00:04:13.067 --> 00:04:14.747
it looks like Screen
Studio, but with more stuff.
00:04:14.747 --> 00:04:17.667
I'm a Screen Studio user, and I,
like I see features there that
00:04:17.777 --> 00:04:19.057
I don't have, and I want them.
00:04:20.087 --> 00:04:24.437
you built this for Deal directly?
00:04:24.687 --> 00:04:28.427
is this a Deal-only thing, or are
there other users outside Deal?
00:04:28.923 --> 00:04:30.733
Avi Ashkenazi: That's a really
good, interesting story.
00:04:30.743 --> 00:04:34.323
So one of the magics about
working in enterprise is that
00:04:34.333 --> 00:04:38.803
once you enter into enterprise,
you cannot exit the enterprise.
00:04:38.853 --> 00:04:44.453
Meaning because, this captive audience
and I built this tool, and that's the
00:04:44.483 --> 00:04:48.093
only tool that is available, I essentially
don't have competition, which ended up
00:04:48.093 --> 00:04:54.893
in me having six hundred daily users
this tool, recording, producing videos.
00:04:54.893 --> 00:04:59.803
I think so far about ten thousand
videos were already produced with this
00:04:59.803 --> 00:05:02.033
tool, and it's like my weekend project.
00:05:02.463 --> 00:05:09.723
Every, weekend, I go, and I sit there and
keep vibe coding my life away produce.
00:05:09.893 --> 00:05:13.943
for example, this weekend,
I added, touch gestures.
00:05:13.953 --> 00:05:19.313
So when when you do, pinch on an
iPhone or zoom in or today, when you
00:05:19.313 --> 00:05:22.063
were-- you plug in the phone to your
computer and you record it, you don't
00:05:22.073 --> 00:05:23.673
have those things, those visuals.
00:05:23.993 --> 00:05:28.713
And now in Viddit, you are able to
basically add them on screen, and you
00:05:28.713 --> 00:05:32.093
can-- people can see where you swipe
from where to you pinch, et cetera.
00:05:32.503 --> 00:05:36.993
Now, you mentioned this is only i-
like, in the internal environment.
00:05:37.003 --> 00:05:41.773
However, when I started testing it,
I basically utilized my own account
00:05:41.773 --> 00:05:44.743
because I wanted my friends to try it
out and give me feedback before I…
00:05:45.093 --> 00:05:48.353
it's a very high stake to
say to the company, "You guys
00:05:48.363 --> 00:05:49.943
all use this from now on."
00:05:50.517 --> 00:05:50.867
Jonny: Yeah
00:05:51.283 --> 00:05:54.363
Avi Ashkenazi: I already had, external
users, and over time, I convinced,
00:05:54.953 --> 00:06:01.123
Alex, the CEO, to basically do
what Lottie, done with, Airbnb.
00:06:02.073 --> 00:06:05.083
the, engineers who built Lottie were like,
"This could really change the internet.
00:06:05.093 --> 00:06:05.663
It helps us.
00:06:05.663 --> 00:06:07.733
It can help other people,
so we will open source it."
00:06:08.563 --> 00:06:13.843
In my case, we're not open sourcing it
because I don't know how I can control
00:06:13.843 --> 00:06:16.623
the fact that no one's gonna put a
shell on top of it and start selling
00:06:16.623 --> 00:06:18.783
it, and I do want to give it for free.
00:06:18.783 --> 00:06:20.383
I want to contribute it to the community.
00:06:20.383 --> 00:06:24.653
So I agreed with Alex that we're
gonna give it for free for anyone.
00:06:24.653 --> 00:06:25.823
So if you go to viddit.video,
00:06:25.823 --> 00:06:28.623
you will be able to download it.
00:06:29.483 --> 00:06:32.403
the website is still in the making,
but, there is a download link there.
00:06:32.413 --> 00:06:36.083
You can go and, download it
now, and anyone who wants to
00:06:36.083 --> 00:06:37.913
use it can use it for free.
00:06:38.941 --> 00:06:39.281
Jonny: Awesome.
00:06:39.421 --> 00:06:44.211
But I suppose having, the captive
audience, you get tons of feedback.
00:06:44.271 --> 00:06:44.891
You--
00:06:44.893 --> 00:06:44.943
Avi Ashkenazi: Thanks
00:06:45.561 --> 00:06:49.741
Jonny: but also they can literally
never leave, so you don't have to
00:06:49.751 --> 00:06:55.511
worry about one whole part of the
product stress, which is churn.
00:06:56.651 --> 00:06:57.351
in a sense,
00:06:57.935 --> 00:06:59.195
Avi Ashkenazi: you do have some…
00:06:59.405 --> 00:07:02.305
so for example, I'll give you
one of the features that I added
00:07:02.355 --> 00:07:04.565
was, it was called Instant Mode.
00:07:05.335 --> 00:07:09.085
happened was I got annoyed 'cause
I built this thing and spent a
00:07:09.085 --> 00:07:10.625
lot of my blood, sweat, and tears.
00:07:10.925 --> 00:07:14.445
I got annoyed that people keep recording
stuff on Slack, and then, someone provides
00:07:14.455 --> 00:07:18.675
feedback on a design, and are, like,
very obsessive about their feedback.
00:07:18.675 --> 00:07:23.825
So you have suddenly, videos of five
minutes ri- one after the other in Slack.
00:07:24.105 --> 00:07:26.295
And I'm like, "Why are
you not using FITED?"
00:07:26.615 --> 00:07:28.735
And they're like, "Look, friction.
00:07:28.815 --> 00:07:29.495
It is there.
00:07:29.525 --> 00:07:33.625
I press the button, and I don't need to
render or edit it or, tweak anything."
00:07:33.625 --> 00:07:36.775
So I made Instant Mode with
ba- which basically means it's
00:07:36.775 --> 00:07:37.865
recording everything live.
00:07:37.865 --> 00:07:39.055
You can preset the environment.
00:07:39.065 --> 00:07:40.815
You can say, "Hey, my
camera's gonna be a circle.
00:07:40.825 --> 00:07:41.945
It's gonna be in this corner.
00:07:42.865 --> 00:07:43.495
gonna be like this.
00:07:43.525 --> 00:07:44.825
The background's gonna be like that."
00:07:45.235 --> 00:07:47.535
And then when you finish recording,
you can just drop it into
00:07:47.535 --> 00:07:49.955
Slack when you record in Slack.
00:07:50.385 --> 00:07:53.855
And that brought me some other
ad-additional percentage of users
00:07:54.673 --> 00:07:55.053
Jonny: Okay
00:07:55.767 --> 00:07:56.317
Tom: Amazing.
00:07:57.257 --> 00:08:03.147
you mentioned about the sort of starting
point of this being 130 people across
00:08:03.147 --> 00:08:11.167
Deel who wanted to use this tool that you
decided to, build your own version of.
00:08:11.947 --> 00:08:16.187
curious how this has changed your ways of
working within the company or the team.
00:08:17.227 --> 00:08:21.967
hinted there at using Slack for design
reviews or maybe using this tool instead.
00:08:22.337 --> 00:08:27.057
How have you not only built this
tool, but how has this changed your
00:08:27.057 --> 00:08:30.657
ways of working as a team or, and
your role as a leader in that team?
00:08:32.877 --> 00:08:37.117
Avi Ashkenazi: I think first of all, it
allowed me to be way closer to designers.
00:08:37.147 --> 00:08:42.517
And from my perspective, I also-- One of
the good, amazing things about tools like
00:08:42.517 --> 00:08:48.587
this is that they enable you to share
stuff in a, such a high fidelity and such
00:08:48.587 --> 00:08:56.167
quality suddenly you have a marketing
video-like output at your fingertips.
00:08:57.137 --> 00:08:58.077
Maybe it's an overkill.
00:08:58.097 --> 00:08:59.277
For some people, it's an overkill.
00:08:59.277 --> 00:09:02.857
And I get, in the feedback channel,
many, questions quite often that
00:09:02.867 --> 00:09:04.667
is like, "Hey, can I do this?"
00:09:04.667 --> 00:09:06.437
And it's "Yeah, actually, you can do this.
00:09:06.447 --> 00:09:07.527
It already is built."
00:09:08.527 --> 00:09:12.187
and that, that tells me, the UX
is maybe discoverability is not
00:09:12.207 --> 00:09:13.477
good enough, et cetera, et cetera.
00:09:13.727 --> 00:09:19.107
But it also gave me and the team a
lot of confidence say, "Hey, we can
00:09:19.127 --> 00:09:21.617
go and build these amazing AI tools."
00:09:22.467 --> 00:09:26.777
I know you guys are doing this brilliant
workshop that, I also witnessed in the
00:09:26.777 --> 00:09:28.427
event that we were in two weeks ago.
00:09:30.787 --> 00:09:31.507
and people…
00:09:32.267 --> 00:09:35.527
It gave people much more
confidence to just go and try out.
00:09:35.547 --> 00:09:38.857
Because before people thought, "Oh, I can
just make a small Figma plugin," or, "I
00:09:38.857 --> 00:09:41.807
can summarize my review in terms of text."
00:09:41.837 --> 00:09:48.477
And that suddenly shows them, okay,
a person who is very, busy can go
00:09:48.477 --> 00:09:52.737
and, alongside do this thing, then
we can do it as well, and gave them
00:09:52.737 --> 00:09:54.057
a lot of confidence to build stuff.
00:09:55.199 --> 00:09:59.719
Jonny: Avi, it, might, be useful
context for people listening to
00:09:59.719 --> 00:10:02.659
describe what Deel is like to work at.
00:10:02.759 --> 00:10:07.489
And I think it's quite a unique
culture, in terms of the attitude
00:10:07.489 --> 00:10:12.149
towards build versus buy and AI
generally and things like that.
00:10:13.069 --> 00:10:16.499
and, there is one anecdote that I
know you've shared a couple of times,
00:10:16.809 --> 00:10:18.989
including on stage around Figma seats.
00:10:19.039 --> 00:10:22.669
So c-can you just give us a little bit of
a-- maybe the design lens, but also just
00:10:22.669 --> 00:10:24.329
generally how Deel works company-wide?
00:10:25.315 --> 00:10:25.545
Avi Ashkenazi: Yeah.
00:10:25.545 --> 00:10:30.675
Deel, trying to reduce as much as
possible politics or anything like that.
00:10:30.685 --> 00:10:34.955
Deel thrives on speed, which
basically means that it bu-- a lot
00:10:34.955 --> 00:10:36.615
of things bubble up from the bottom.
00:10:36.735 --> 00:10:40.065
Now, of course, this could be some
projects where, s-senior stakeholder
00:10:40.995 --> 00:10:47.095
will be involved, but gradually-- but
overall, would say that if you have an
00:10:47.095 --> 00:10:51.205
idea, just go and do it because there's
not enough people to do all the ideas and
00:10:51.205 --> 00:10:55.385
not enough people to do even the projects
that already are scoped, that anyone is
00:10:55.385 --> 00:11:00.405
welcome with any additional initiative
that they have to just go and contribute.
00:11:02.715 --> 00:11:08.795
I mentioned, the design team is about 130
people across the board, and we have this,
00:11:11.165 --> 00:11:16.015
AI process that enabled us in one click,
basically, to onboard people all the way
00:11:16.015 --> 00:11:22.795
from setting up their Amazon permissions
to, set up their repo database inside
00:11:23.385 --> 00:11:27.755
GitHub, and then install the cloud
skills that they need and connect it
00:11:27.755 --> 00:11:29.525
to the design system, essentially.
00:11:29.635 --> 00:11:33.405
So one-click button, you can basically
go and prototype anything that you want,
00:11:33.655 --> 00:11:37.215
Claude Design, Claude Code, whichever
tool that you feel like, VS Code,
00:11:37.325 --> 00:11:39.415
Cursor, whatever tool that you feel like.
00:11:39.645 --> 00:11:42.925
And we also have a hosting environment
internally called Nexus, which is
00:11:43.175 --> 00:11:47.045
like our own version of Vercel,
anyone can ship projects there
00:11:47.065 --> 00:11:48.985
and has many interesting features.
00:11:49.355 --> 00:11:54.565
That was built in two weeks by two
DevOps engineers, and now has been
00:11:54.565 --> 00:11:55.825
maintained in the last two months.
00:11:55.825 --> 00:11:58.615
It was pretty awful at the
beginning, and now it's fantastic.
00:11:59.643 --> 00:12:00.593
Jonny: Do they get some design feedback?
00:12:02.315 --> 00:12:04.375
Avi Ashkenazi: it got some design
feedback, but not enough, I would
00:12:04.375 --> 00:12:09.355
say because that's one of the
problems, like when people build AI,
00:12:09.355 --> 00:12:15.125
like sometimes it's there's like a,
small ego aspect there that you "But
00:12:15.125 --> 00:12:16.305
yeah, but I built all this thing.
00:12:16.305 --> 00:12:16.905
I don't need help.
00:12:16.905 --> 00:12:18.105
I don't want people to slow me down.
00:12:18.105 --> 00:12:21.725
I just wanna go and churn and do as
much as I can, as fast as I can."
00:12:22.495 --> 00:12:25.235
we got to a point where at the
beginning, people were freaking out,
00:12:25.255 --> 00:12:28.915
especially when, Claude released--
Anthropic released Claude Design.
00:12:30.505 --> 00:12:32.845
People were like, "I,
I-- We cannot keep up.
00:12:33.235 --> 00:12:34.515
The, pace is too fast.
00:12:34.535 --> 00:12:35.735
We don't know what to do."
00:12:35.855 --> 00:12:41.135
And so we organized twice a week
meetings with designers that are
00:12:41.415 --> 00:12:44.685
great at AI or that have achieved
something in AI that others haven't
00:12:45.365 --> 00:12:47.185
do this cross-pollination and teach.
00:12:47.955 --> 00:12:52.465
we got to a point where, at this point
of time, last, at the beginning of,
00:12:52.605 --> 00:12:56.945
July, I got a few messages from designers
like, "Hey, I lost my Figma license.
00:12:56.965 --> 00:12:57.685
Can you renew me?"
00:12:57.685 --> 00:13:00.275
And the reason why they lost
it is because they practically
00:13:00.545 --> 00:13:02.085
have not used it for 30 days.
00:13:02.115 --> 00:13:05.495
And we have an automation that if someone
doesn't use it for 30 days, we obviously
00:13:05.495 --> 00:13:09.965
don't wanna pay for it, so we clear the
seat from that person for someone else
00:13:09.965 --> 00:13:13.205
who might want to use or to basically
not pay at the end of the quarter.
00:13:15.843 --> 00:13:19.123
Jonny: you had a bunch of designers that
turned out hadn't opened Figma for 30 days
00:13:19.715 --> 00:13:21.855
Avi Ashkenazi: Yeah, 15, 15 of them by now
00:13:23.583 --> 00:13:28.923
Jonny: 15 of 130 is is a under
10%, but it's-- or around 10%, but
00:13:28.923 --> 00:13:34.133
it's, still significant when you
think about a tool that, was…
00:13:34.423 --> 00:13:39.963
has been, the very center of
everyone's process for many years.
00:13:40.583 --> 00:13:42.813
that and precursors to
it a decade or more.
00:13:44.083 --> 00:13:44.783
yeah, really interesting.
00:13:45.063 --> 00:13:49.533
so these designers that are not opening
Figma, they are opening their code editor.
00:13:49.883 --> 00:13:50.363
Is that right?
00:13:51.075 --> 00:13:51.245
Avi Ashkenazi: Yeah.
00:13:52.005 --> 00:13:52.255
just
00:13:52.323 --> 00:13:52.643
Jonny: Okay.
00:13:53.475 --> 00:13:57.025
Avi Ashkenazi: and I would say that also,
look, lot of the designers I've been
00:13:57.025 --> 00:14:02.935
reading about it also from other people's
posts, the usage for Figma has slightly
00:14:02.945 --> 00:14:04.775
changed in terms of, what people do.
00:14:04.875 --> 00:14:06.765
There's some stuff that you just can't do.
00:14:07.085 --> 00:14:09.655
if you wanna create an icon, you…
00:14:09.785 --> 00:14:12.775
Yeah, you might prompt it, but it's
not gonna be great, so you'll just
00:14:12.775 --> 00:14:17.865
go and, do some vector tooling and
pen tool and, move some points.
00:14:19.045 --> 00:14:22.475
some of the designers, that's what
their usage in Figma is nowadays.
00:14:22.505 --> 00:14:28.245
For the whole interface, they just do
Claude code and for really, small things.
00:14:28.245 --> 00:14:32.935
So I think there is, interesting
opportunity to make something that
00:14:32.935 --> 00:14:37.615
would replace just these, small
little processes of visual refinement
00:14:37.615 --> 00:14:43.595
and fidelity, and then they might
not need even Figma that, version.
00:14:44.405 --> 00:14:49.785
, Jonny: I'd love your opinion as a
s-- you're a d- you've been leading
00:14:49.785 --> 00:14:54.335
design teams for a long time and, seen,
design teams outside Deel as well.
00:14:54.705 --> 00:14:59.375
How Deel specific this behavior is
versus what a prediction might be of
00:14:59.375 --> 00:15:03.895
where the design industry might end
up tooling itself over the next, say,
00:15:03.895 --> 00:15:05.795
twelve, eighteen, twenty-four months.
00:15:06.205 --> 00:15:11.605
w- is this, heading towards everyone's
vibe coding or is this actually just
00:15:12.305 --> 00:15:16.775
something that Deel's a bit mad and really
early or, yeah, somewhere in between?
00:15:18.249 --> 00:15:19.429
Avi Ashkenazi: I think
it's somewhere in between.
00:15:20.059 --> 00:15:24.359
it's also hard for me, for example, to
contrast Shopify to Deal because when
00:15:24.359 --> 00:15:25.699
I was at Shopify, we were at like 4.5,
00:15:25.699 --> 00:15:27.219
4.6
00:15:27.229 --> 00:15:28.849
in terms of Opus, right?
00:15:28.849 --> 00:15:29.139
And now
00:15:29.241 --> 00:15:29.421
Jonny: Yes
00:15:29.459 --> 00:15:31.329
Avi Ashkenazi: like way
more advanced, sophisticated
00:15:31.349 --> 00:15:33.349
models, so it does get better.
00:15:33.769 --> 00:15:39.259
And Anthropic seems to be, unlike
OpenAI, willing to go into any territory
00:15:40.339 --> 00:15:40.419
Jonny: Good
00:15:40.529 --> 00:15:43.589
Avi Ashkenazi: and start
battling against Figma.
00:15:43.779 --> 00:15:47.059
And, Dylan was also asked about…
00:15:47.059 --> 00:15:49.739
The CEO of Figma was asked about it
in an interview, I don't remember
00:15:49.739 --> 00:15:52.379
with who, but he was like, "Yeah,
relationship is complicated."
00:15:53.539 --> 00:15:55.719
because again, they're
like coming for their cake.
00:15:56.069 --> 00:15:57.459
it was a very interesting thing.
00:15:57.459 --> 00:16:00.989
Like he was, "Hey, we're opening
MCP, so you can bring your Figma
00:16:00.989 --> 00:16:04.879
design, and then now you can take
it back, but we want you to stay,
00:16:05.457 --> 00:16:05.887
Jonny: Yeah
00:16:05.979 --> 00:16:08.389
Avi Ashkenazi: we suddenly made
like Cloud Design so you can stay."
00:16:09.719 --> 00:16:14.699
And I think like that, Cloud Design was
better at anyone else at digesting the
00:16:14.699 --> 00:16:20.119
design system and bearable, results it.
00:16:20.969 --> 00:16:23.939
So I think Deal is ahead, definitely.
00:16:23.989 --> 00:16:26.769
I can't tell exactly what's
happening right now in…
00:16:27.149 --> 00:16:31.049
There's a lot of companies that are quite
advanced, I believe, as us, like Shopify,
00:16:31.059 --> 00:16:35.309
Ramp, example, potentially Fin as well.
00:16:37.459 --> 00:16:39.419
yeah, I think there's like a…
00:16:39.919 --> 00:16:44.789
It's an interesting journey together
with all those aspects of design system
00:16:44.789 --> 00:16:50.389
and generative, UI to react rather than
you planning everything in advance.
00:16:50.709 --> 00:16:54.629
There's a lot of really, interesting
avenues there to explore, in my opinion.
00:16:55.789 --> 00:16:57.569
Tom: Can I ask one more tactical question?
00:16:57.569 --> 00:17:02.899
Zooming in on those 15 people who
had their Figma seats revoked, what
00:17:02.919 --> 00:17:07.399
is the like when these designers
come to you, requesting a seat?
00:17:07.449 --> 00:17:11.889
Is it like a, what, this is
ludicrous, why is this happening?
00:17:11.919 --> 00:17:16.609
Or is there an openness to maybe
if you are the sort of designer who
00:17:16.609 --> 00:17:22.979
is just needing Figma for the icon
level refinement, is there a, real
00:17:22.979 --> 00:17:26.009
possibility that these designers will
go off and build their own tools for the
00:17:26.009 --> 00:17:27.639
specific use cases they're solving for?
00:17:28.509 --> 00:17:31.219
us a flavor of that,
spectrum that you're seeing.
00:17:31.977 --> 00:17:35.477
Avi Ashkenazi: I think they're all very
surprised that it happened, of course,
00:17:35.737 --> 00:17:38.897
they take it for granted that they
will have it, and maybe in the future,
00:17:39.867 --> 00:17:43.037
based on usage and what we're seeing,
maybe it's not gonna be for granted
00:17:43.037 --> 00:17:44.707
that everyone's gonna have Figma seats.
00:17:46.917 --> 00:17:51.357
but also you have the other side of
the equation, which is token maxing and
00:17:51.357 --> 00:17:53.237
token budget and those type of things.
00:17:53.737 --> 00:18:00.227
Designers, literally in Slack are
relieved the 1st of July comes.
00:18:00.597 --> 00:18:02.367
I'm, I burned all my tokens.
00:18:03.797 --> 00:18:05.347
waiting for, tomorrow at 3:00 AM.
00:18:05.427 --> 00:18:06.647
I, I know the time.
00:18:07.337 --> 00:18:10.097
Like an addict, I know the
time that this thing is open.
00:18:10.097 --> 00:18:13.497
I'm on, 90, 98% of my…
00:18:13.547 --> 00:18:16.397
And I'm in a group that is called
super users, which has way more
00:18:16.397 --> 00:18:18.277
budget than the normal users
00:18:20.059 --> 00:18:20.269
Tom: Oh.
00:18:22.049 --> 00:18:23.919
that's, yeah, fascinating.
00:18:24.109 --> 00:18:26.669
A- and is that across within
design or across the whole
00:18:26.669 --> 00:18:28.579
company, the super users bucket?
00:18:28.941 --> 00:18:30.011
Avi Ashkenazi: It's
across the whole company.
00:18:30.011 --> 00:18:31.241
There's 100 of those.
00:18:33.211 --> 00:18:37.521
I think there are four people
from design in those 100.
00:18:37.881 --> 00:18:40.441
The company is 7,200 people
00:18:42.253 --> 00:18:42.383
Tom: Wow.
00:18:43.975 --> 00:18:50.175
Jonny: Avi, you mentioned four, four, four
of your design team, including you are
00:18:50.285 --> 00:18:52.795
the sort of token maxes, the, super users.
00:18:53.485 --> 00:18:58.095
Can you describe like a sort of an ad-
adoption curve across the team and are
00:18:58.095 --> 00:19:04.135
there people that are still unsure and,
how are you thinking about, as, as much
00:19:04.135 --> 00:19:07.805
as you can say, how are you thinking
about how to either help those people
00:19:08.275 --> 00:19:11.295
to come on board or actually that it's
okay for some people to be like that
00:19:11.295 --> 00:19:13.085
and there's work for them to do as well?
00:19:13.485 --> 00:19:14.455
yeah, how are you thinking about that?
00:19:15.607 --> 00:19:17.847
Avi Ashkenazi: Look, I think there are
still people that for them it's way
00:19:17.847 --> 00:19:22.547
faster to go straight to Figma, and
they produce better, results faster.
00:19:22.547 --> 00:19:26.687
And I think that there's, there still
is room for Figma from a collaboration
00:19:26.777 --> 00:19:31.317
perspective and, yeah, primarily that.
00:19:31.317 --> 00:19:34.647
Because, the workflow f-for
code is very uncollaborative.
00:19:35.327 --> 00:19:38.487
It's like you submit the PR, someone
reviews the PR, someone pulls the
00:19:38.559 --> 00:19:38.819
Jonny: Yeah
00:19:39.067 --> 00:19:39.537
Avi Ashkenazi: natural.
00:19:39.537 --> 00:19:40.427
It's not easy.
00:19:41.117 --> 00:19:42.437
although I have seen…
00:19:42.677 --> 00:19:43.347
if you think of,
00:19:45.477 --> 00:19:48.307
OpenAI group chats where you can
incorporate or, Claude agents
00:19:48.317 --> 00:19:52.187
inside Slack, you suddenly have
this world of collaboration.
00:19:52.977 --> 00:19:55.447
How can they bring this thing
into the world of design?
00:19:55.447 --> 00:19:57.877
Or who can be the first to bring
this thing into the world of design?
00:19:57.887 --> 00:19:59.267
It's gonna be really interesting.
00:19:59.827 --> 00:20:07.507
In terms of distribution, I would say
that definitely of the designers are on
00:20:07.507 --> 00:20:10.147
Claude Design, on Claude Code doing this.
00:20:10.467 --> 00:20:16.247
Some of them are utilizing it as the,
this is the high fi- Le- Maybe, 40%
00:20:16.257 --> 00:20:22.447
of them, maybe 30% of them utilizes
it as this is my, platform to make
00:20:22.467 --> 00:20:27.507
this super high-end result that I
can sell to, or, user research with,
00:20:27.517 --> 00:20:29.907
or sell to, stakeholders, et cetera.
00:20:30.587 --> 00:20:34.657
The other ones are using it the completely
other way and are like, "This is an
00:20:34.667 --> 00:20:38.677
artifact that would allow me to align
with the PM and engineer, that this
00:20:38.677 --> 00:20:40.447
is the actual flow that I need to do.
00:20:40.667 --> 00:20:44.907
Then I'm gonna take it to Figma,
and I'm gonna make it, and bring
00:20:44.907 --> 00:20:48.517
it the way it needs to be brought
into them implementing it."
00:20:48.937 --> 00:20:51.947
So you have, quite an
interesting skew between them.
00:20:52.127 --> 00:20:59.317
And of course, and then there's,
20% who not, Are worried, are, not
00:21:00.657 --> 00:21:02.637
getting on board with this thing.
00:21:02.637 --> 00:21:06.707
And I think they gradually they are
struggling more and more because people
00:21:06.707 --> 00:21:09.077
are getting way faster across the board.
00:21:10.397 --> 00:21:13.607
I think that they have a really
nice support system to try to
00:21:13.607 --> 00:21:18.017
help them gain that, knowledge,
and they have lots of really good
00:21:18.017 --> 00:21:19.357
examples of what good looks like.
00:21:20.191 --> 00:21:20.531
Jonny: Yeah
00:21:21.365 --> 00:21:23.465
Tom: And g- could you give
specific examples of that?
00:21:23.505 --> 00:21:29.275
Is that through the team around them,
or are there specific, examples of
00:21:29.275 --> 00:21:32.535
resources that you've built or the
team has built to help people onboard?
00:21:33.349 --> 00:21:35.599
Avi Ashkenazi: we, as I mentioned,
we have twice a week sessions
00:21:35.629 --> 00:21:39.689
with, showcase like different
implementations of what designers…
00:21:39.689 --> 00:21:41.469
And the designers are
running it for themselves.
00:21:41.479 --> 00:21:45.369
It's not that, I come or anyone
from leadership comes and does this.
00:21:45.379 --> 00:21:48.769
They-- You have some designers who
are way more advanced at this thing,
00:21:48.769 --> 00:21:53.399
and they "How I shipped feature X into
production," and they show that workflow.
00:21:53.399 --> 00:21:59.119
Or, "How d- how did I optimize,
agents to do research more fast with
00:21:59.279 --> 00:22:01.299
Akai," which is our agent platform.
00:22:03.169 --> 00:22:07.119
and this is also for example, there's a
designer called Mariana, she's in Brazil.
00:22:07.429 --> 00:22:12.299
She worked on this with Akai on a
project to optimize research funnels.
00:22:12.629 --> 00:22:15.479
So if you think of what you need
to do to resea- for research, you
00:22:15.489 --> 00:22:17.129
have some target audience in mind.
00:22:17.129 --> 00:22:21.049
"Hey, I wanna talk to that
have contractors in Argentina
00:22:21.049 --> 00:22:22.319
and Algeria," for example.
00:22:22.539 --> 00:22:24.369
And the company is 200 people.
00:22:26.869 --> 00:22:27.659
how do I find them?
00:22:28.559 --> 00:22:29.379
you can maybe search.
00:22:29.389 --> 00:22:34.719
So would go search for that, and then
you would find 20 companies like that.
00:22:34.719 --> 00:22:36.729
Okay, now we need to
organize their user research.
00:22:37.099 --> 00:22:37.309
Great.
00:22:37.429 --> 00:22:40.319
Akai would set a budget of how
much do we need to pay them.
00:22:40.759 --> 00:22:45.679
It will go and talk to the CSM, so the
customer success people that are usually
00:22:45.679 --> 00:22:50.939
talking to them, ask them to ask for
a meeting and would also chase them
00:22:50.939 --> 00:22:54.589
in Slack until they book that meeting
with a client, which is easier than
00:22:54.589 --> 00:22:56.819
just sending like a cold email, right?
00:22:57.599 --> 00:23:01.619
then when time to book the meeting, it
would look at the calendars and basically
00:23:01.649 --> 00:23:05.799
suggest the times that you should book,
and then you just come and do a session
00:23:07.433 --> 00:23:10.843
Jonny: How do the CSMs feel about,
being told to do stuff by an AI?
00:23:12.589 --> 00:23:13.159
Avi Ashkenazi: They're okay.
00:23:13.379 --> 00:23:19.259
They're, they, I think at this point of
time they're being told so much, stuff
00:23:19.269 --> 00:23:21.779
by AI that they're totally used to it.
00:23:23.819 --> 00:23:24.119
Yeah
00:23:24.539 --> 00:23:25.069
Jonny: wild
00:23:25.739 --> 00:23:29.309
Tom: Just slightly changing gears, I
think you've given us a few examples
00:23:29.329 --> 00:23:35.839
so far of building and tools or
workflows to help automate or improve
00:23:35.839 --> 00:23:37.379
ways of working and processes.
00:23:38.089 --> 00:23:43.389
I'm curious about how you think
about trying to replicate your own
00:23:43.619 --> 00:23:48.699
judgment or scale your own judgment
and taste versus, which bits you're
00:23:48.699 --> 00:23:52.479
trying to keep deliberately manual and
deliberately human, if there are any
00:23:53.915 --> 00:23:54.335
Avi Ashkenazi: Yeah.
00:23:54.825 --> 00:23:56.755
I don't have, the best definitive…
00:23:56.785 --> 00:24:01.955
I'm, I think, very much, if you look
in the world of AI, memory is one of
00:24:01.955 --> 00:24:03.305
the things that is being talked about.
00:24:03.305 --> 00:24:05.665
I'm not talking about
memory like RAM and stuff.
00:24:05.705 --> 00:24:08.845
I'm talking about, like, how to
manage memory, how to optimize memory.
00:24:11.435 --> 00:24:13.995
if you think about it, the
whole art of storytelling is
00:24:14.005 --> 00:24:16.245
for us to remember something.
00:24:16.945 --> 00:24:19.565
How we write blog posts, how
we write books, et cetera.
00:24:19.905 --> 00:24:21.205
It's all anchored at that.
00:24:22.175 --> 00:24:26.675
But memory of an AI is different,
and memory of you is different.
00:24:26.675 --> 00:24:30.955
So I-- There's two projects that
a couple of people are working on,
00:24:30.955 --> 00:24:33.615
and, one of them is called Links.
00:24:33.645 --> 00:24:36.185
It's also an internal
product that I built.
00:24:36.945 --> 00:24:38.045
Again, same reason.
00:24:38.085 --> 00:24:42.435
People wanted Obsidian or Logseq
or Roam Research or Notion, and
00:24:42.815 --> 00:24:44.785
we couldn't, they couldn't use it.
00:24:44.825 --> 00:24:45.575
It's not approved.
00:24:45.585 --> 00:24:48.235
So I basically built our
own version of Obsidian.
00:24:49.225 --> 00:24:51.805
I cannot show it here because
I have it on my work computer.
00:24:54.295 --> 00:24:59.915
this tool started as just implementing
and mimicking Obsidian You have the
00:24:59.915 --> 00:25:03.805
knowledge graph, you have links that
between subjects that links to each other.
00:25:03.805 --> 00:25:07.725
You have different types of primitives,
like a person, a project, a product, a
00:25:07.725 --> 00:25:10.795
vertical, a decision, an idea, et cetera.
00:25:11.495 --> 00:25:16.515
And I initially started it with,
a Claude code skill that goes and
00:25:16.525 --> 00:25:20.275
looks through all my meetings and
summarizes it based on those things.
00:25:20.375 --> 00:25:24.075
Then I ended up building an MCP
to Claude, so it could feed and
00:25:24.075 --> 00:25:27.365
read it in a more coherent way, in
addition to the, my personal notes.
00:25:27.765 --> 00:25:30.475
And now it's all about
trying to manage a brain.
00:25:31.565 --> 00:25:35.905
and then a different project that I'm
also working on, which is our Jira
00:25:35.905 --> 00:25:38.425
replacement to some extent, or Monday.com
00:25:38.425 --> 00:25:41.975
replacement, which is called Launchpad.
00:25:42.675 --> 00:25:46.345
And in that world, you need a
brain per vertical or per team.
00:25:46.355 --> 00:25:50.505
And it's different from your brain
because a brain per team is reading the
00:25:50.705 --> 00:25:52.685
Slack channels of that individual team.
00:25:52.685 --> 00:25:55.875
And it's not your opinion,
it's holistic opinion.
00:25:56.265 --> 00:26:00.005
Even if someone onboards the team or
someone wants to ask or know a thing
00:26:00.015 --> 00:26:04.375
about the team, they, that surface
should have all the jobs to be done
00:26:04.375 --> 00:26:07.145
and all the metrics and the decisions
that were made and what the impact
00:26:07.155 --> 00:26:08.645
was done based on those decisions.
00:26:08.965 --> 00:26:14.065
So building those brains is definitely
like an adventure I have, been
00:26:14.075 --> 00:26:19.045
on in the last couple of months,
reading Andrej Karpathy's, comments,
00:26:19.275 --> 00:26:20.565
and no one has figured it out.
00:26:20.565 --> 00:26:26.595
There's m-ma-many ways, and it's really
hard to make sure that it's not gonna be a
00:26:26.615 --> 00:26:28.785
thing that people just forget and not use.
00:26:28.805 --> 00:26:32.215
It's really hard to make people maintain
it, and it's also hard to make it
00:26:32.225 --> 00:26:37.985
maintain itself automatically and keep
the most up-to-date content keep it
00:26:37.985 --> 00:26:39.755
healthy in a way that is queryable.
00:26:43.033 --> 00:26:45.463
Jonny: People want all of these tools
and these capabilities, but then you
00:26:45.463 --> 00:26:49.753
give them to them and it's adds to
the chaos, the seeming chaos in 2025,
00:26:49.753 --> 00:26:55.323
'26 of I'm meant to know all of this
stuff suddenly and, the tool, as
00:26:55.323 --> 00:26:56.923
you say, the tool's moving too fast.
00:26:56.923 --> 00:26:58.083
I can't keep up with it all.
00:26:58.333 --> 00:26:59.193
Now I have a brain.
00:26:59.193 --> 00:26:59.963
What's a brain?
00:27:00.253 --> 00:27:02.943
Now I have to actually do more work.
00:27:02.943 --> 00:27:06.033
Feels like more work, not less work, even
though it's an efficiency and there's
00:27:06.033 --> 00:27:07.863
a new capability that you've given me.
00:27:08.363 --> 00:27:11.923
And, I remember running my last
company and trying to work out how to
00:27:11.993 --> 00:27:18.093
effect change internally within, our
customer, as we rolled out to, clients
00:27:18.533 --> 00:27:21.773
and working with internal HR teams on
change management and, like, how do
00:27:22.543 --> 00:27:25.163
fire people up around this thing that
they now have, which is really good
00:27:25.163 --> 00:27:28.433
for them and, it's gonna help, but
they've got so much else on their plate.
00:27:28.823 --> 00:27:33.793
The actual work is so important
that how do you make space for that?
00:27:34.003 --> 00:27:39.473
I'm just interested in, m- is
there anything you've done candid,
00:27:39.593 --> 00:27:42.683
candidly, is there stuff that
you're like, "This is amazing.
00:27:42.703 --> 00:27:44.373
Everyone's gonna love it,"
and then it fell flat?
00:27:44.603 --> 00:27:48.323
And it sounds like the, video tool has
d- has done pretty well because there
00:27:48.323 --> 00:27:50.223
was just such, this latent demand for it.
00:27:50.223 --> 00:27:53.243
But are there other things
that you've been like, "Oh,
00:27:53.243 --> 00:27:53.403
it's a bit of a damp squib"?
00:27:54.106 --> 00:27:57.916
Avi Ashkenazi: Yeah, I think, look, I
think L- Lynx, the kind of our version
00:27:57.916 --> 00:27:59.946
of Obsidian is, hasn't taken off yet.
00:28:00.146 --> 00:28:03.536
I don't have product market fit,
even though it's the only tool there.
00:28:04.266 --> 00:28:10.436
And I think a lot of it is because
am I working for the AI or am,
00:28:10.496 --> 00:28:11.936
or does the AI work for me?
00:28:12.906 --> 00:28:14.846
with a tool like this,
it slightly feels…
00:28:14.846 --> 00:28:18.136
And that's why I'm working on these brain
optimizations, because I want to make
00:28:18.136 --> 00:28:23.226
sure it feels more like AI is working
for me, and I do a little bit I'm just
00:28:23.226 --> 00:28:26.646
sitting there writing all the notes so
the AI can read it, et cetera, et cetera.
00:28:26.866 --> 00:28:30.976
and then, add to that the aspect of,
and the company's following me and
00:28:32.086 --> 00:28:33.976
of really documenting everything.
00:28:33.976 --> 00:28:34.406
And is it…
00:28:34.906 --> 00:28:35.456
In this case,
00:28:35.537 --> 00:28:35.777
Jonny: Yeah
00:28:36.156 --> 00:28:39.596
Avi Ashkenazi: offline and it's on your
computer, but even that, is potentially
00:28:39.636 --> 00:28:43.996
queryable in some jurisdictions, in
some conditions, et cetera, et cetera.
00:28:44.486 --> 00:28:49.406
So I think it does sums down
to am I working for the AI
00:28:49.406 --> 00:28:50.866
or does the AI work for me?
00:28:50.886 --> 00:28:58.226
And people think of, in warehouses in
China that are doing the dishes and
00:28:58.236 --> 00:29:02.326
having s- tons of cameras around them and
teaching the future robots of how to do
00:29:02.326 --> 00:29:06.646
your dishes and how to fold your laundry,
and they don't want to be that in the work
00:29:06.653 --> 00:29:06.923
Jonny: Yep
00:29:07.116 --> 00:29:08.806
Avi Ashkenazi: and they believe
that they are better than,
00:29:09.026 --> 00:29:10.536
They probably are better than.
00:29:10.616 --> 00:29:16.396
We just need to find, better systems that
would do all these things automatically.
00:29:16.396 --> 00:29:21.326
But, if you think of alignment, and
a lot of people are, in the world
00:29:21.326 --> 00:29:24.806
of design and also product, a lot
of people are very tired, especially
00:29:24.816 --> 00:29:26.106
in leadership or more senior roles.
00:29:26.106 --> 00:29:31.376
They're, like, talking to people,
politics, diplomacy, agreeing on stuff.
00:29:31.506 --> 00:29:33.146
AI cannot do that stuff for you.
00:29:33.306 --> 00:29:34.526
It's very, tiring.
00:29:34.596 --> 00:29:39.216
And, managing a brain is also,
like, that to some extent, right?
00:29:39.226 --> 00:29:41.246
It's what do you choose to read?
00:29:41.726 --> 00:29:42.356
what do you write?
00:29:42.356 --> 00:29:43.296
What is your intelligence?
00:29:43.296 --> 00:29:44.036
What do you query?
00:29:44.786 --> 00:29:50.646
cannot settle for boiling the ocean every
time and asking, "Look across all Google
00:29:50.646 --> 00:29:52.746
Drive and Slack channels for my answer."
00:29:53.586 --> 00:29:54.936
not gonna give you the best answer.
00:29:55.246 --> 00:29:58.446
The best answer would be based
on your brain, so i-i-it's
00:29:58.446 --> 00:29:59.376
worth making the effort.
00:29:59.376 --> 00:30:03.916
But these are the things that
sometimes make people tired.
00:30:04.589 --> 00:30:04.859
Jonny: Yeah
00:30:05.069 --> 00:30:08.929
I think I-- my question was about
the, thing that has always existed,
00:30:08.979 --> 00:30:12.179
which is like the value to me versus
the va-- the work I have to do.
00:30:12.199 --> 00:30:15.329
The value out must be at-- there's
some ratio, like ten times.
00:30:15.419 --> 00:30:19.339
Must get ten times the value out that I
put in, otherwise I'll ignore the thing.
00:30:19.739 --> 00:30:23.489
But I think you just touched on the
deeper and much more AI-specific
00:30:23.739 --> 00:30:28.119
question, which is why am I doing
this, and like who benefits in the end,
00:30:28.739 --> 00:30:30.059
of all of this stuff that I'm doing?
00:30:30.199 --> 00:30:31.649
And these stories, yeah, these stories.
00:30:31.789 --> 00:30:35.369
there's that cleaning company that set
up in the US and is now in the UK as
00:30:35.369 --> 00:30:39.829
well, where you get a free clean, but the
cleaner wears a hat with the cameras and,
00:30:41.859 --> 00:30:42.229
the…
00:30:42.379 --> 00:30:42.929
who's paying?
00:30:42.929 --> 00:30:44.129
If I'm not paying, who's paying?
00:30:44.129 --> 00:30:47.839
Turns out it's the, the companies
that are training the robots
00:30:47.839 --> 00:30:49.959
how to r-replace the cleaners.
00:30:50.959 --> 00:30:54.389
you have to take an ethical stance or
you just go for a free clean, and I'm
00:30:54.389 --> 00:30:55.929
sure a lot of people go for a free clean.
00:30:56.709 --> 00:30:57.979
But, if you are the…
00:30:58.329 --> 00:31:02.709
If you're the one wearing the
hat, that's, of course gonna be
00:31:02.709 --> 00:31:08.819
a consideration, that gets in the
way of all the cool stuff that, you
00:31:08.819 --> 00:31:10.289
know, that, you can give to them.
00:31:10.649 --> 00:31:13.289
But, yeah, I, completely
understand it for sure.
00:31:15.829 --> 00:31:16.209
great.
00:31:16.299 --> 00:31:16.609
Yeah.
00:31:16.639 --> 00:31:22.649
So but Deel is quite AI forward and
yet un-understandably across a company
00:31:22.649 --> 00:31:25.619
of thousands and a team of over a
hundred, there's gonna be a mixture
00:31:25.619 --> 00:31:31.879
of feelings about putting all of my
stuff in just for an AI to, train on it
00:31:33.168 --> 00:31:33.428
Avi Ashkenazi: Definitely.
00:31:35.168 --> 00:31:35.368
Yeah
00:31:36.334 --> 00:31:36.654
Tom: Yeah.
00:31:36.944 --> 00:31:37.524
Fascinating.
00:31:38.284 --> 00:31:43.114
have you found specifically with
this, brain-like tool, are there
00:31:43.114 --> 00:31:49.474
use cases that you're targeting
that are proving like there's more,
00:31:51.684 --> 00:31:54.234
structured value in a specific area?
00:31:54.234 --> 00:31:58.844
Or is it intended to be quite wide-ranging
and more sort of general purpose context?
00:31:59.810 --> 00:32:02.770
Avi Ashkenazi: Look, I thought at
the beginning when I made the initial
00:32:02.770 --> 00:32:06.060
pitch for this thing, I thought
that just saying, "Hey, something
00:32:06.060 --> 00:32:07.970
is the brain," is not enough.
00:32:08.760 --> 00:32:11.300
And then I thought, what
else can this thing do?
00:32:12.140 --> 00:32:15.400
And then I came up to,
it could save us tokens.
00:32:16.270 --> 00:32:18.770
could save us tokens because it
doesn't need to go and boil the ocean,
00:32:18.770 --> 00:32:21.810
and it's like a llama that filters
what you need to do and aims you at
00:32:21.810 --> 00:32:23.880
the right place at the right time.
00:32:24.830 --> 00:32:27.510
And people are not sure, like everyone
that I talk to, they're not sure.
00:32:27.510 --> 00:32:28.170
So I don't know.
00:32:28.220 --> 00:32:31.800
I don't know yet if that's actually
like the value that a thing like this
00:32:32.430 --> 00:32:33.000
would bring.
00:32:33.820 --> 00:32:36.480
I still think maybe,
but it's hard to tell.
00:32:36.480 --> 00:32:38.890
And how much do you save
locally and how is it…
00:32:38.960 --> 00:32:39.940
Is it manageable?
00:32:39.950 --> 00:32:41.530
When does it stop being manageable?
00:32:41.530 --> 00:32:46.210
It's a really interesting user
experience that you need, and
00:32:46.210 --> 00:32:47.410
technical question as well.
00:32:48.140 --> 00:32:51.020
And it-- do people really
care about this thing?
00:32:51.020 --> 00:32:52.010
Maybe today, no.
00:32:52.010 --> 00:32:57.260
Maybe later, going to be like a great
setup for the future where everyone
00:32:57.260 --> 00:33:00.080
is like hiking their cost stuff.
00:33:00.080 --> 00:33:05.620
And maybe in the future, we will end up in
a place where everyone has access to AI.
00:33:06.420 --> 00:33:10.180
and only people who we know that they
can actually get like really useful
00:33:10.190 --> 00:33:12.500
output out of it will have access, to AI
00:33:12.749 --> 00:33:13.459
Jonny: Yeah, interesting.
00:33:13.459 --> 00:33:15.949
I w-- I, I actually
had a related question.
00:33:16.059 --> 00:33:21.199
W-presumably Deel are exploring
open weight models and other,
00:33:21.229 --> 00:33:23.159
and non-Claude options.
00:33:23.389 --> 00:33:28.559
Claude, the Anthropic models
are very expensive, and GLM 5.2
00:33:28.559 --> 00:33:30.749
and all of these interesting
models coming out.
00:33:30.769 --> 00:33:35.579
Are you-- is there some exploration
into h- choose how to use your budget?
00:33:35.579 --> 00:33:41.379
You can have a worse model and, more,
swings at bat or anything like that?
00:33:41.838 --> 00:33:45.008
Avi Ashkenazi: at the moment,
it's primarily, Anthropic models,
00:33:45.008 --> 00:33:47.698
at least, for the normal dealer.
00:33:49.138 --> 00:33:52.348
but that, gives you the range
all the way from like Sonnet 4.6
00:33:52.358 --> 00:33:53.398
to Sonnet 5 and like
00:33:53.421 --> 00:33:53.641
Jonny: Yeah
00:33:53.838 --> 00:33:56.138
Avi Ashkenazi: But also Fable
is limited to its own budget.
00:33:56.528 --> 00:33:58.888
You cannot like, just burn
your whole budget and…
00:33:59.078 --> 00:34:02.798
Because if you think about it, you
don't wanna get to a point where you
00:34:02.798 --> 00:34:06.018
went, you burn all your Fable tokens,
which basically you burn all your
00:34:06.018 --> 00:34:09.508
monthly tokens because you didn't use
it appropriately, and now you're like,
00:34:09.988 --> 00:34:11.608
"Sorry guys, I can't do any work."
00:34:13.078 --> 00:34:13.688
a problem, right?
00:34:14.788 --> 00:34:18.028
and I don't know if it's like a deal
setting or like an Anthropic setting.
00:34:18.478 --> 00:34:19.358
maybe it's both.
00:34:19.478 --> 00:34:20.348
Hard, hard to tell.
00:34:21.168 --> 00:34:23.748
but that's a thing that like
I first burned through my…
00:34:23.778 --> 00:34:26.538
Because I got so excited and I was
like, "They're gonna take it away again.
00:34:26.668 --> 00:34:27.118
Oh, no."
00:34:27.548 --> 00:34:31.738
So I just like just went through all
my, tokens and then was like, "Okay,
00:34:31.958 --> 00:34:33.498
we can go back to four-- Opus 4.
00:34:33.498 --> 00:34:34.418
4.8
00:34:34.888 --> 00:34:36.048
and see what happens there."
00:34:36.238 --> 00:34:40.378
But I believe that in the future,
there's a stat, 70% of the startups
00:34:40.388 --> 00:34:42.528
in the US are on Chinese models.
00:34:42.568 --> 00:34:45.588
Kimi, the models from
Alibaba, Wan, et cetera.
00:34:46.898 --> 00:34:49.578
now no one says that, Trump
is not gonna come tomorrow
00:34:49.578 --> 00:34:50.548
and say, "Hey, you know what?
00:34:50.918 --> 00:34:52.198
you can only use American models.
00:34:52.248 --> 00:34:54.408
I don't want this thing
to like leak to this."
00:34:54.408 --> 00:35:00.288
And then, amazing for the OpenAI and these
guys who are lobbying very, hard, then
00:35:00.307 --> 00:35:00.417
Jonny: Yeah.
00:35:00.607 --> 00:35:01.867
Great regulatory capture.
00:35:01.877 --> 00:35:01.977
Yeah.
00:35:02.057 --> 00:35:02.497
Yeah
00:35:02.698 --> 00:35:04.038
Avi Ashkenazi: Yeah,
it could be incredible.
00:35:05.098 --> 00:35:08.298
I don't know if he will get
to it, they did it with cars.
00:35:08.558 --> 00:35:12.088
They haven't done it with products, but
they did it with cars, Or maybe they
00:35:12.088 --> 00:35:14.168
just put a slap a tax on it and then
00:35:14.201 --> 00:35:14.421
Jonny: Yep
00:35:14.568 --> 00:35:15.078
Avi Ashkenazi: the same price.
00:35:15.278 --> 00:35:19.128
So y-you don't know what's going to happen
there, and I think, yeah, it's definitely,
00:35:19.258 --> 00:35:22.448
an area to watch and see where it goes.
00:35:23.565 --> 00:35:26.620
Tom: Let's talk about human
to human interaction with AI
00:35:26.730 --> 00:35:28.570
in some part of that, loop.
00:35:29.220 --> 00:35:32.610
I guess there's two things I wanted
to set the scene with for this.
00:35:32.640 --> 00:35:38.380
The first is, posted on LinkedIn
about training an AI model or a
00:35:38.380 --> 00:35:42.880
tool to review designer portfolios
to help with the hiring process and
00:35:42.880 --> 00:35:46.490
help you screen candidates and, level
them at the start of the process.
00:35:47.490 --> 00:35:51.200
which I think is a very, compelling,
00:35:53.330 --> 00:35:56.490
compelling challenge for a
hiring manager to solve for.
00:35:56.700 --> 00:35:58.650
I'm curious to ask you some
questions about that in a moment.
00:35:59.290 --> 00:36:04.310
then you also mentioned a thread recently
about, AI as an en- enabler for human
00:36:04.310 --> 00:36:08.555
to human collaboration and how AI was
supposed to make things, it was supposed
00:36:08.555 --> 00:36:11.410
to make collaboration seamless, but
instead it has made us more isolated.
00:36:12.260 --> 00:36:17.230
talking about the loss of human friction,
associated with these workflows that now
00:36:17.230 --> 00:36:21.420
have you, building your own tool in AI
rather than, collaborating with other,
00:36:21.630 --> 00:36:23.920
disciplines or people to make them happen.
00:36:24.300 --> 00:36:27.620
So I think that is a space I'm really
interested to dig into a bit with you.
00:36:28.500 --> 00:36:33.550
perhaps we could start with the hiring
process aspects and, perhaps you could
00:36:33.550 --> 00:36:41.610
tell folks about the, the current state
of designer hiring at Deel and how AI
00:36:41.610 --> 00:36:43.460
is integrated into that, process, or
where you found success with AI and
00:36:44.260 --> 00:36:48.340
where, it's, where you'd like to solve
but maybe haven't quite captured yet.
00:36:49.428 --> 00:36:52.158
Avi Ashkenazi: Like one of the, one of
the things that I've done when I joined
00:36:52.158 --> 00:36:58.948
Deel was elevate our recruiting practices
and processes, introduce new interviews,
00:36:59.398 --> 00:37:03.068
provide better feedback, record videos
for people who I've heard from them,
00:37:03.378 --> 00:37:06.718
from candidates that were really,
helpful to explain what we're looking.
00:37:06.738 --> 00:37:10.648
Also, bringing a face to it makes it
nice and, personally, as someone who
00:37:10.648 --> 00:37:15.568
does not love just reading stuff, I
prefer to listen to stuff or watch stuff.
00:37:16.168 --> 00:37:20.178
It gives You I believe that as you
create content and you know someone who
00:37:20.178 --> 00:37:23.128
wants to work for you, you are creating
content for them because you want to also
00:37:23.128 --> 00:37:24.638
convince them to come and work with you.
00:37:25.328 --> 00:37:28.888
It is important to have those podcasts
as a part of my, the podcast that,
00:37:29.028 --> 00:37:31.308
I'm working on, a-as well as videos.
00:37:32.038 --> 00:37:36.648
I think that there's a lot of things
that people don't talk about, like how
00:37:36.658 --> 00:37:38.688
do we judge portfolios, how do we…
00:37:38.698 --> 00:37:42.588
Like some people are, which are more
consultants, would do these live
00:37:42.598 --> 00:37:47.678
YouTube or like LinkedIn portfolio
reviews or provide and give some time.
00:37:47.928 --> 00:37:48.978
I do usually,
00:37:51.218 --> 00:37:54.788
I get I don't know, four
messages a day on LinkedIn with,
00:37:54.848 --> 00:37:56.648
"Hey, check out my portfolio.
00:37:56.648 --> 00:37:57.318
What do you think?"
00:37:57.718 --> 00:38:00.418
And obviously, sometimes I have
time, but sometimes I don't have
00:38:00.418 --> 00:38:02.048
time, so I, the AI does help me.
00:38:02.048 --> 00:38:04.998
And if I do that with AI, I would
basically send them a screenshot
00:38:05.008 --> 00:38:06.858
of "Hey, I have this tool.
00:38:07.418 --> 00:38:10.258
Here is how it looks like in this
tool," and I'm gonna run it and
00:38:10.268 --> 00:38:11.768
walk them through what it says.
00:38:12.658 --> 00:38:14.258
The idea was to…
00:38:15.138 --> 00:38:18.928
We have 300 applications a day at
Deel for just for design, right?
00:38:19.228 --> 00:38:24.228
last year in total for Deel, there
were about 300, sorry, 3,000,300
00:38:24.228 --> 00:38:26.548
applications across the board year.
00:38:26.950 --> 00:38:28.390
Jonny: That's a bit more than 300, yeah.
00:38:28.698 --> 00:38:30.968
Avi Ashkenazi: that's a bit more
than 300 a day, but even 300 a
00:38:30.968 --> 00:38:33.658
day, times 365, that's a lot.
00:38:34.068 --> 00:38:34.468
Jonny: That's a lot.
00:38:34.548 --> 00:38:34.858
Yeah
00:38:34.918 --> 00:38:35.088
Avi Ashkenazi: I'm
00:38:35.188 --> 00:38:35.338
Tom: Wow.
00:38:35.488 --> 00:38:37.078
Avi Ashkenazi: at math, so I'm
not gonna try to attempt to
00:38:37.078 --> 00:38:39.258
answer that, but it is a lot.
00:38:40.878 --> 00:38:44.538
and a lot of it is, some of it is bot,
some of this is this, but like a lot of
00:38:44.538 --> 00:38:50.948
it is people who really try and really
want feedback, and if, they just get
00:38:50.958 --> 00:38:56.278
like a, "Sorry, not for us," type of
email, which I feel is a bummer, and
00:38:56.278 --> 00:39:00.778
it's not gonna help them improve in any
way, and I wanted to put an end to this.
00:39:01.148 --> 00:39:03.578
And so this is why I
started building this tool.
00:39:05.848 --> 00:39:08.218
it was just-- it wasn't
just, the problem for me.
00:39:08.218 --> 00:39:12.388
I literally felt the problem for
everyone who messaged me and went the
00:39:12.388 --> 00:39:14.668
extra mile and wanted to help them.
00:39:15.368 --> 00:39:18.808
And then I was like, okay, this
thing could be really powerful
00:39:18.808 --> 00:39:21.128
in multiple steps in the process.
00:39:21.158 --> 00:39:25.788
It could be really powerful when someone
is applying for a staff position, we
00:39:25.788 --> 00:39:29.758
can scan it as they paste the link and
tell them, "Actually, you're not staff,
00:39:29.778 --> 00:39:33.898
you're senior Deel's perspective, and
this is what's missing from staff.
00:39:34.058 --> 00:39:34.718
Do you have it?
00:39:34.738 --> 00:39:35.928
If you do have it, awesome.
00:39:35.928 --> 00:39:37.158
Bring it into the portfolio.
00:39:37.418 --> 00:39:40.778
We would love to look for--
look at it and, evaluate it.
00:39:41.268 --> 00:39:45.808
Or, you think we're wrong, it's also okay,
you proceed with application anyway."
00:39:46.448 --> 00:39:48.808
And also to bring it into, an automated…
00:39:48.818 --> 00:39:52.478
If we are rejecting someone or something
because it's a bot or something
00:39:52.478 --> 00:39:54.438
like that, we can provide feedback.
00:39:54.748 --> 00:39:58.738
Now, other-- some people would say,
"Yeah, but everyone then will improve
00:39:58.738 --> 00:40:00.278
based on what the AI is measuring."
00:40:01.088 --> 00:40:01.468
okay.
00:40:01.668 --> 00:40:02.238
That's okay.
00:40:02.258 --> 00:40:05.578
Then the average goes higher,
and the floor goes higher, and
00:40:05.578 --> 00:40:07.258
everyone gains from it, by the way.
00:40:07.708 --> 00:40:11.088
In addition to that, it's not that the
AI, you just build it and leave it.
00:40:11.798 --> 00:40:13.238
It improves over time, right?
00:40:13.358 --> 00:40:16.578
Either by itself or, you adopting
it based on what you're seeing,
00:40:16.578 --> 00:40:19.138
because at the end, you will look at
those portfolios and you will talk
00:40:19.608 --> 00:40:21.808
to those people, so everyone wins.
00:40:21.838 --> 00:40:23.458
Everyone can make a better portfolio.
00:40:23.458 --> 00:40:24.868
You will see better portfolios.
00:40:24.888 --> 00:40:28.688
The system gets better and,
the managers have less work.
00:40:30.728 --> 00:40:36.108
Tom: Yeah, Is this something that you
have scaled across all design hiring
00:40:36.108 --> 00:40:38.148
at Deel, or is it still Avi specific?
00:40:39.430 --> 00:40:41.370
Avi Ashkenazi: it's still
me specific, unfortunately.
00:40:41.720 --> 00:40:47.020
we have released our own ATS really
recently, I think like a month ago,
00:40:47.740 --> 00:40:49.340
we will see if we can incorporate it.
00:40:49.340 --> 00:40:52.270
But also, remember, this is very design…
00:40:52.470 --> 00:40:57.340
there is a tool for that for CV checking
ag-against kind of, the job requirements,
00:40:57.340 --> 00:40:59.810
but Portfolio is a very design-centric.
00:41:00.020 --> 00:41:04.530
I had to add stuff there, like
specifically read Figma and specifically
00:41:04.530 --> 00:41:08.910
understand when pages have password,
and if the person provided the password,
00:41:08.930 --> 00:41:10.640
you would be able to crawl through that.
00:41:11.780 --> 00:41:16.590
the ability to do OCR on images
and have some sort of analysis,
00:41:16.640 --> 00:41:21.300
like visually, the text is sizes
of text give some understanding.
00:41:21.300 --> 00:41:24.800
So there's lots of like very
design-specific things that are in this
00:41:24.800 --> 00:41:29.120
tool that are just not scalable for
something that is not like design hiring.
00:41:30.490 --> 00:41:36.290
but I did give the code for this thing to
some of my design managers, and I, can't
00:41:36.290 --> 00:41:40.310
tell for sure if they're using it because
it's not still in the form of code.
00:41:40.310 --> 00:41:41.880
It's not like a product that you can
00:41:44.730 --> 00:41:48.750
but the idea is to potentially keep
using it until we feel it's like
00:41:48.750 --> 00:41:50.430
the right time and, then implement.
00:41:50.430 --> 00:41:52.440
I don't think it's like the
top priority for the company,
00:41:53.220 --> 00:41:54.370
for us it's really helpful.
00:41:55.494 --> 00:41:58.164
Jonny: It's a sort of one of those,
hiring brand things where you--
00:41:59.234 --> 00:42:02.544
there's-- you can't point to a number
the-- it doesn't impact the bottom
00:42:02.544 --> 00:42:07.224
line directly or at least immediately,
so it has to be a sort of a passion
00:42:07.224 --> 00:42:10.274
project until it finds its correct home.
00:42:10.274 --> 00:42:13.744
But it sounds really interesting, and
I've always felt whenever I've been
00:42:13.754 --> 00:42:17.314
hiring, you get stuck in the situation
where you're like, "I want to give
00:42:17.314 --> 00:42:21.914
you really thoughtful feedback, but
I don't-- I have, no time to do it."
00:42:21.924 --> 00:42:25.024
And we generally don't do it because
then what if they argue their case
00:42:25.354 --> 00:42:26.974
and, that gets difficult as well.
00:42:29.744 --> 00:42:33.004
before interview stage, you are stuck
doing the thing that you hate, and you'd
00:42:33.004 --> 00:42:34.974
hate to receive the email that you send.
00:42:36.014 --> 00:42:37.874
but that's how the process works.
00:42:37.894 --> 00:42:41.754
Even worse with now AI and
so many more applications.
00:42:42.904 --> 00:42:43.824
yeah, kudos.
00:42:43.824 --> 00:42:47.204
It'll be interesting to see if you
ever manage to get it into Deel's ATS.
00:42:48.014 --> 00:42:52.434
and, to be clear, for listeners who
don't know Deel well, this is the ATS
00:42:52.434 --> 00:42:54.494
Deel sells to, to its customers, right?
00:42:55.156 --> 00:42:55.506
Avi Ashkenazi: Yes
00:42:55.564 --> 00:42:55.874
Jonny: this…
00:42:55.934 --> 00:42:56.214
Yeah.
00:42:56.284 --> 00:42:56.514
Yeah.
00:42:56.514 --> 00:43:01.424
So this is, this could be some-something
born out of Avi's head could, be
00:43:01.454 --> 00:43:04.634
deployed out to, thousands of companies.
00:43:04.634 --> 00:43:04.964
Who knows?
00:43:04.964 --> 00:43:05.974
But yeah.
00:43:06.408 --> 00:43:06.818
Tom: Yeah.
00:43:07.118 --> 00:43:07.458
Yeah.
00:43:08.228 --> 00:43:10.858
one more tactical question
on this specific example.
00:43:10.858 --> 00:43:15.268
I'm curious if there are any
takeaways you have on the ability
00:43:15.438 --> 00:43:19.708
for AI to evaluate a designer's
portfolio from having built this.
00:43:19.858 --> 00:43:24.058
Are there specific attributes that
are quite hard for it to evaluate
00:43:24.118 --> 00:43:26.028
and/or get wrong sometimes?
00:43:26.108 --> 00:43:27.048
And what are those?
00:43:28.078 --> 00:43:31.208
Avi Ashkenazi: look, would say there's,
a couple of things that I noticed,
00:43:31.208 --> 00:43:35.468
especially at the beginning when I was
building it, which is self-affirmation.
00:43:35.968 --> 00:43:41.238
for example, a designer would say, "An
expert of, UX design for 16 years," that
00:43:41.618 --> 00:43:43.668
They are like, "Yeah, the guy's amazing.
00:43:43.678 --> 00:43:44.378
He's an expert."
00:43:44.658 --> 00:43:46.558
No, he just wrote that
he's an expert, right?
00:43:46.558 --> 00:43:48.078
So you know,
00:43:48.332 --> 00:43:50.562
Jonny: Just believes
literally what they say.
00:43:50.662 --> 00:43:51.042
Yeah.
00:43:51.208 --> 00:43:51.488
Avi Ashkenazi: yeah.
00:43:52.098 --> 00:43:56.318
different thing that I think is really,
valued, and like it's a thing that
00:43:56.738 --> 00:44:00.798
everyone, and I'm sure you guys have
tested for this when you were interviewing
00:44:00.798 --> 00:44:05.328
designers, and I think like I've even
heard, like Johnny, in one of your
00:44:05.678 --> 00:44:10.188
previous podcasts that we were talking
about this as well, is, uniqueness, right?
00:44:10.198 --> 00:44:11.818
Like you want people to be unique.
00:44:11.828 --> 00:44:15.898
So the tool would look if you have
used like a stock Framer template or
00:44:15.898 --> 00:44:19.108
a stock Wix template or Squarespace
or anything like that, because
00:44:21.288 --> 00:44:21.688
should you do it?
00:44:21.708 --> 00:44:22.608
Let's change the color.
00:44:22.608 --> 00:44:23.438
Do something, right?
00:44:23.558 --> 00:44:27.518
so it would basically check if the
person actually made an effort make
00:44:27.518 --> 00:44:28.428
something a little bit different.
00:44:28.438 --> 00:44:36.138
Even if it's like in the world of Lovable
or, Cursor, if you use like stock, is
00:44:36.358 --> 00:44:39.038
completely visible right now to everyone.
00:44:39.038 --> 00:44:42.218
Like we all know when we see something
that was created with AI that no
00:44:42.218 --> 00:44:43.958
one has tinkered with a little
00:44:44.070 --> 00:44:44.500
Jonny: Yeah
00:44:46.048 --> 00:44:52.078
Avi Ashkenazi: so watching out for this
is, an important because it's lazy and
00:44:52.078 --> 00:44:57.538
you don't want lazy people and, yes, the
content controls and the storytelling,
00:44:57.548 --> 00:45:01.718
et cetera, but that's one aspect, but you
also don't want them to be lazy visually.
00:45:01.728 --> 00:45:05.198
If any s- especially if someone is
looking for a job, it's really critical.
00:45:05.198 --> 00:45:07.788
Now, one of the hardest thing, and
that's not the thing that AI can
00:45:07.788 --> 00:45:11.598
solve, is, imagine I wanna hire Johnny.
00:45:11.608 --> 00:45:15.528
Johnny is not looking for a job, so
Johnny does not have a portfolio, Johnny?
00:45:15.628 --> 00:45:17.238
I suspect you don't
have a portfolio ready.
00:45:17.648 --> 00:45:17.898
Yeah.
00:45:18.148 --> 00:45:20.708
So how do I assess that?
00:45:20.758 --> 00:45:23.638
And that's not just, for like
portfolio evaluation, right?
00:45:23.658 --> 00:45:27.478
Like he might have projects from like
long time ago and he's gonna come
00:45:27.478 --> 00:45:32.308
and like he will have to confirm to
the flow supposedly of like needs to
00:45:32.308 --> 00:45:36.758
do a case study, needs to do these
things and no, and he's, he, doesn't
00:45:36.768 --> 00:45:41.258
have it because he's so good that
he doesn't need, he doesn't need it.
00:45:41.588 --> 00:45:41.978
people just
00:45:42.062 --> 00:45:42.202
Jonny: And then
00:45:42.208 --> 00:45:43.318
Avi Ashkenazi: him because
they know he's good.
00:45:43.728 --> 00:45:44.828
So how do you go around that?
00:45:45.668 --> 00:45:47.068
these are some of the challenges.
00:45:47.528 --> 00:45:50.898
when you source people versus
when people apply to you, the
00:45:51.208 --> 00:45:51.488
Jonny: Right
00:45:52.098 --> 00:45:54.348
Avi Ashkenazi: and
differentiating is important.
00:45:54.918 --> 00:45:58.588
Jonny: Yeah, sourcing is so much
more gut and network and just,
00:46:00.948 --> 00:46:02.328
Yeah, which is risky as well, right?
00:46:02.328 --> 00:46:06.978
Because, someone who's set up some
good references but isn't great could
00:46:06.978 --> 00:46:09.528
slip through that net, so you still
have to have that level of rigor.
00:46:09.668 --> 00:46:14.508
One thing you just said I think is really
interesting because I think the last
00:46:14.508 --> 00:46:17.778
time I was hiring was really pre, pre-AI.
00:46:18.058 --> 00:46:21.818
last time I was hiring designers
was pre-AI, and you could get pretty
00:46:21.818 --> 00:46:23.588
far by just having built a website.
00:46:23.808 --> 00:46:29.218
the fact that it existed on the internet
and didn't look unusual or broken would
00:46:29.218 --> 00:46:34.838
get you some distance and, you'd made
some sensible typography decisions,
00:46:34.838 --> 00:46:36.418
and the hierarchy was okay, and,
00:46:39.388 --> 00:46:43.498
a- and now the cost of production
is so cheap that actually there's a
00:46:43.498 --> 00:46:50.168
premium on having a s- an opinion, and
that opinion being or feeling fresh.
00:46:50.988 --> 00:46:55.908
someone who I think we're gonna have
on the podcast, Innie, w- came to the
00:46:55.908 --> 00:47:02.688
first-- Tom's first, Mostly Working
morning and demoed her portfolio
00:47:02.688 --> 00:47:06.048
site, and it was, like, a room
that you could turn the lights on
00:47:06.048 --> 00:47:07.878
and off, and it was, like, really,
00:47:09.978 --> 00:47:14.278
w- exploratory and interesting and
unusual, and she'd, used image models
00:47:14.278 --> 00:47:21.018
in interesting ways, and that's so much
more-- it says so much more about the
00:47:21.018 --> 00:47:24.648
designer, even if that UI is not very
useful for really any other product.
00:47:24.948 --> 00:47:26.768
they're gonna go and design some B2B SaaS.
00:47:26.768 --> 00:47:29.598
It's not gonna look like that,
but, at least it has something
00:47:29.608 --> 00:47:31.688
about it, and you've gone beyond.
00:47:33.618 --> 00:47:38.558
yeah, no longer can you just design
linear in your portfolio and then hope
00:47:38.908 --> 00:47:42.498
that gets you over the, over the hill.
00:47:42.528 --> 00:47:46.808
But yeah, I'm not hiring
at the moment, Avi, yeah.
00:47:46.868 --> 00:47:53.418
But it, it feels a really interesting
time to decide as-- even as a hiring
00:47:53.418 --> 00:47:58.118
team, what your opinion is about what
good is, and you can have a unique
00:47:58.118 --> 00:48:01.568
opinion as a team about what you're
looking for in a much more meaningful way.
00:48:02.228 --> 00:48:06.438
Avi Ashkenazi: Yeah, and that's why
I would say this is gonna be very…
00:48:07.558 --> 00:48:09.208
it's gonna be controversial, I believe.
00:48:09.478 --> 00:48:10.018
that's where…
00:48:10.018 --> 00:48:14.618
And I think, Deel would allow
for this, which is good.
00:48:16.488 --> 00:48:21.988
if you hire for Google, probably
won't be able to someone like
00:48:21.998 --> 00:48:23.368
Ini and say, "You know what?
00:48:24.458 --> 00:48:25.608
I don't need anything else.
00:48:25.988 --> 00:48:29.838
I will interview her four times so I
understand everything," and it will…
00:48:30.038 --> 00:48:33.548
No, you will have to go through the
process because of many, reasons.
00:48:33.548 --> 00:48:36.118
Many very, by the way,
very important reasons.
00:48:37.908 --> 00:48:40.658
and like fairness and a lot of really
00:48:40.850 --> 00:48:41.040
Jonny: Yes
00:48:41.588 --> 00:48:46.508
Avi Ashkenazi: But you cannot, no
matter what you do, take like a complete
00:48:46.538 --> 00:48:48.418
bet on someone and bring them in.
00:48:48.678 --> 00:48:49.818
You just cannot do that.
00:48:50.838 --> 00:48:51.908
can do that.
00:48:52.038 --> 00:48:55.338
I made a post long time ago,
that it's like I saw someone,
00:48:56.058 --> 00:49:00.008
failed the interview, and I was
like, "We're hiring this person."
00:49:00.778 --> 00:49:05.558
And basically over, and it's a bet
and it su- like some point it sucks.
00:49:05.788 --> 00:49:09.338
and I we- I went and I took some
of the feedback from my team and
00:49:09.338 --> 00:49:11.668
I told them, "Guys, I decide.
00:49:13.838 --> 00:49:14.798
you're, wrong here."
00:49:14.988 --> 00:49:17.618
And I think, we made, there's a
bunch of other people, I cannot
00:49:17.618 --> 00:49:20.858
name them, but we were wrong.
00:49:21.608 --> 00:49:23.038
were wrong in not hiring them.
00:49:23.048 --> 00:49:24.938
And we had a interview.
00:49:25.058 --> 00:49:28.578
The interviewers maybe did not do
the job well or maybe we didn't
00:49:28.578 --> 00:49:33.768
communicate interview very well, and
we lost that candidate and now they're
00:49:33.768 --> 00:49:35.288
like principal designer somewhere
00:49:35.462 --> 00:49:37.702
Jonny: Yeah.
00:49:37.702 --> 00:49:37.952
Yeah
00:49:37.958 --> 00:49:43.378
Avi Ashkenazi: And so I think sometimes
it is important just be able to have
00:49:43.378 --> 00:49:47.398
this, wild card and some credits
with leadership to say, "I really
00:49:47.398 --> 00:49:49.218
wanna make a bet on this person.
00:49:49.378 --> 00:49:52.208
If it works, If it doesn't,
cool, I learned from it."
00:49:52.870 --> 00:49:56.300
Jonny: We should move, start, wrapping.
00:49:56.300 --> 00:49:57.710
We've got some closers.
00:49:59.570 --> 00:50:03.360
Tom: Yeah, I will pick out
this question for you, Avi.
00:50:04.790 --> 00:50:07.940
so the promise of AI is to give back time.
00:50:08.410 --> 00:50:13.620
So if AI and A- AI agents can do the
majority of your work in the background
00:50:13.630 --> 00:50:18.221
giving you all this time, my question
is: What would you spend that time on
00:50:18.623 --> 00:50:23.323
Avi Ashkenazi: I think that I would,
first of all, I, believe that my job and
00:50:23.333 --> 00:50:29.383
the job of a director as a whole, and
a design leader is to identify areas.
00:50:30.193 --> 00:50:31.433
basically being like water.
00:50:31.673 --> 00:50:34.013
you pour water to a bucket with holes.
00:50:34.943 --> 00:50:37.443
You see where there's, small
holes, and you just like…
00:50:37.483 --> 00:50:40.723
I just go in the container and feel it
with my hand and see where there's a
00:50:40.723 --> 00:50:44.413
hole, and I decide if I put a plaster
in it or I need, a complete rework of
00:50:44.413 --> 00:50:48.063
this area to make sure that the ship
is not gonna sink because of this.
00:50:49.003 --> 00:50:54.013
And so I need to find three, four
new things like this every quarter
00:50:54.183 --> 00:50:58.403
to tell the deal and ma- and make
those bets and, pitch them at least.
00:50:59.863 --> 00:51:03.703
if 50% of them passed and actually went
into the roadmap, then it's successful.
00:51:03.703 --> 00:51:06.473
And we're talking about, cross
vertical, really, big things.
00:51:06.483 --> 00:51:10.523
So completely changing the direction
of the company or making a massive bet
00:51:10.543 --> 00:51:14.873
with lots of resources associated with
it, then I've been successful, and if I
00:51:14.883 --> 00:51:16.613
haven't, then I have not been successful.
00:51:16.963 --> 00:51:19.143
The other thing that I need
to do is unblock my team.
00:51:19.533 --> 00:51:23.373
And so if I see a- all my team designing
with Lorem Ipsum, then I make a plugin
00:51:24.183 --> 00:51:28.753
in Figma that allows them to just, access
the database and fill real information
00:51:29.133 --> 00:51:33.633
from client information into Figma, so
we can actually judge the cases for real.
00:51:33.653 --> 00:51:34.973
That's, an example that happened.
00:51:36.353 --> 00:51:37.013
so I think…
00:51:37.383 --> 00:51:41.363
And, same, even with video,
they do best storytelling.
00:51:41.363 --> 00:51:45.183
Okay, here's the best tool, go
and, tell stories in a better way.
00:51:45.603 --> 00:51:47.653
Oh, you can't find stuff in your note?
00:51:47.683 --> 00:51:51.803
Okay, here's, Obsidian, like, all your
notes are connected, and you can query
00:51:51.803 --> 00:51:56.933
it with Claude I think these are, like,
optimizing the team from an operational
00:51:56.963 --> 00:52:00.513
perspective, getting them excited about
what's possible and inspire them and
00:52:00.513 --> 00:52:04.593
also making these big product bets
is what you need to do, and that's
00:52:04.593 --> 00:52:06.853
what I choose to spend my time on.
00:52:06.923 --> 00:52:10.723
And then the extra time that I have fun
while I watch football or, whatever I
00:52:10.723 --> 00:52:15.173
watch, is just like vibe coding 'cause,
you make a prompt, and then you need to
00:52:15.173 --> 00:52:19.193
wait for five, ten, maybe more So why not?
00:52:19.223 --> 00:52:20.953
and you can do it throughout, I
do it throughout the whole day.
00:52:20.953 --> 00:52:25.213
I do it, I didn't do it when we were
talking, in the majority of my meetings,
00:52:25.213 --> 00:52:29.213
I will for a second go make another
prompt and just, let it run because
00:52:29.223 --> 00:52:30.303
it's a waste of time if I don't.
00:52:31.189 --> 00:52:31.389
Jonny: Yeah.
00:52:31.699 --> 00:52:33.469
So you don't, you're not
picking up the guitar behind
00:52:33.469 --> 00:52:34.869
you, Avi, and, strumming that?
00:52:35.239 --> 00:52:35.909
Maybe occasionally?
00:52:36.657 --> 00:52:37.757
Avi Ashkenazi: Occasionally, yeah.
00:52:37.947 --> 00:52:38.297
Occasionally.
00:52:39.837 --> 00:52:40.217
Not enough
00:52:43.187 --> 00:52:43.887
Jonny: I'm gonna ask one.
00:52:43.887 --> 00:52:48.537
I'm not-- Actually, it might, be
off menu, but, Avi, I think we've
00:52:48.547 --> 00:52:52.317
had a few spicy opinions from you
or like spicy, like level of spice,
00:52:52.317 --> 00:52:55.417
like Nando's medium maybe, so far.
00:52:55.647 --> 00:53:00.927
But I'm interested, what's your
spiciest take on design and AI?
00:53:01.257 --> 00:53:03.107
where it's going, what's gonna happen?
00:53:03.737 --> 00:53:04.547
Predict the future
00:53:06.492 --> 00:53:07.302
Avi Ashkenazi: I don't know if it's spicy.
00:53:07.502 --> 00:53:09.102
I, will give you a, non-design, okay?
00:53:09.892 --> 00:53:12.462
and then I will think about it all day and
00:53:12.509 --> 00:53:12.859
Jonny: Yeah
00:53:12.942 --> 00:53:14.302
Avi Ashkenazi: you later and tell you
00:53:14.749 --> 00:53:14.919
Jonny: Yeah.
00:53:14.919 --> 00:53:16.389
leave a voice, leave a voicemail.
00:53:16.749 --> 00:53:17.859
we can pop that in.
00:53:18.019 --> 00:53:18.299
Yeah
00:53:18.732 --> 00:53:19.582
Avi Ashkenazi: actually, you know what?
00:53:19.582 --> 00:53:23.512
I actually think that, there was a
spicy opinion that got I think like
00:53:23.512 --> 00:53:31.462
my, most commented on post on LinkedIn
was like a spicy opinion, which was no
00:53:31.462 --> 00:53:33.102
one should be, at least at Deel, right?
00:53:33.432 --> 00:53:37.352
or in the majority of, I would
say, American tech companies, no
00:53:37.352 --> 00:53:40.882
one can be a manager without being
at least a staff level designer
00:53:43.124 --> 00:53:43.624
Craft-wise.
00:53:44.694 --> 00:53:48.144
that is because it is
expected for you as a manager.
00:53:48.354 --> 00:53:52.134
Y- no one can basically be pure strategic,
pure operations, pure alignment.
00:53:52.844 --> 00:53:57.674
know that there are organizations that are
more bureauc- bureaucratical and require
00:53:58.014 --> 00:54:01.894
and has room for this, but I don't believe
that in the tech industry there is room
00:54:01.894 --> 00:54:07.194
for managers who are just managing That's
one, and that's not I said, you know it.
00:54:07.204 --> 00:54:10.654
Like Brian, Chesky said
that and many others.
00:54:11.514 --> 00:54:12.444
And I…
00:54:12.494 --> 00:54:16.814
My, my add to this is that a manager
should at least be at a staff level,
00:54:16.814 --> 00:54:20.404
and in order to be at least staff
level, it means you need to practice
00:54:20.444 --> 00:54:25.094
design regularly, like I do on my
kind of, side projects and sometimes,
00:54:26.234 --> 00:54:27.894
in design with the designers.
00:54:28.964 --> 00:54:31.784
And especially if you
would bring in new talent.
00:54:32.024 --> 00:54:35.664
If, you actually want to make the
bet and bring new talent, junior
00:54:35.664 --> 00:54:37.964
talent, even more required for you.
00:54:37.964 --> 00:54:42.664
Because if you can't teach them, how
the hell would they look up to you?
00:54:43.434 --> 00:54:46.194
how the hell would they learn
from you how to do craft?
00:54:46.204 --> 00:54:49.214
It's not-- Like AI is not gonna teach
them how to do craft because AI does
00:54:49.214 --> 00:54:54.374
not know how to do craft, regardless of
all the seven million new, generative,
00:54:55.574 --> 00:54:56.774
design systems that are coming.
00:54:59.504 --> 00:54:59.764
Jonny: Amen.
00:55:01.234 --> 00:55:02.834
Tom, we gotta keep practicing.
00:55:03.840 --> 00:55:04.340
Tom: Oh, yeah.
00:55:04.370 --> 00:55:04.720
Yeah.
00:55:04.950 --> 00:55:05.310
Yes.
00:55:05.690 --> 00:55:06.130
We, do.
00:55:06.250 --> 00:55:06.620
We do.
00:55:07.800 --> 00:55:12.930
I'm curious sort of a, slight evolution
of that last question, but what's a belief
00:55:12.980 --> 00:55:17.490
about AI and craft that you held a year
ago that you've changed your mind on?
00:55:17.800 --> 00:55:22.550
Or is there anything you've changed your
mind on in the last year based on and
00:55:22.550 --> 00:55:23.330
explorations you've been doing with AI?
00:55:26.460 --> 00:55:31.290
Avi Ashkenazi: I think historically
there was always the people with a hammer
00:55:31.290 --> 00:55:34.430
always looked first for, a nail, right?
00:55:34.550 --> 00:55:37.660
And I think that it's
happening also with AI.
00:55:38.430 --> 00:55:43.480
I think where AI interjected and
ruined some of the, that metaphor is
00:55:43.490 --> 00:55:49.470
with process, because now you c- you
cannot really create process with AI.
00:55:49.710 --> 00:55:53.040
you can, but it basically compresses the
process to a point where it's not really
00:55:53.040 --> 00:55:58.030
a process is no longer important, so you
can't talk about process in the context
00:55:58.030 --> 00:56:00.040
of AI, really, which is interesting.
00:56:00.190 --> 00:56:04.230
and the areas where you can, which
is more alike, akin to what I was
00:56:04.230 --> 00:56:09.010
talking about with brain, are so
unknown and unclear, and by the way,
00:56:09.010 --> 00:56:14.180
so nerdy, that the normal person does
not know enough to talk about them.
00:56:17.500 --> 00:56:20.440
the, thing that has changed, like in
the last year, is that everyone is
00:56:20.460 --> 00:56:25.280
able to get to such good high level,
I would say, content, and sound like
00:56:25.280 --> 00:56:26.520
they know what they're talking about
00:56:28.892 --> 00:56:33.642
And, in the past, I believed
everything that I read or more, i-
00:56:33.652 --> 00:56:37.322
in the form of that if it sounds
convincing, and now I, believe almost
00:56:37.322 --> 00:56:40.192
like nothing, which is really hard
because it's a lot of cognitive load
00:56:40.202 --> 00:56:44.162
to everything and, want to check
00:56:46.582 --> 00:56:48.482
But that's the world that we're living in
00:56:49.188 --> 00:56:49.508
Jonny: Yeah.
00:56:50.848 --> 00:56:53.588
Avi, I think we should, we should wrap.
00:56:54.608 --> 00:56:58.642
Thank you so much for being our first
ever guest, what are the links that
00:56:58.642 --> 00:57:00.432
you'd like to drop in the show notes?
00:57:00.432 --> 00:57:02.622
is there anything that,
that you wanna shout out?
00:57:04.741 --> 00:57:07.071
Avi Ashkenazi: yeah, we
can drop vidit.video.
00:57:08.091 --> 00:57:10.261
We can drop, superavi.com
00:57:11.211 --> 00:57:13.711
and deel.com/careers
00:57:14.478 --> 00:57:16.398
Jonny: You're a good follow on
LinkedIn, I feel like as well.
00:57:16.398 --> 00:57:17.998
You've got some good spicy takes there
00:57:19.059 --> 00:57:20.209
Avi Ashkenazi: I am trying, yes.
00:57:21.069 --> 00:57:21.529
Hopefully.
00:57:23.127 --> 00:57:23.537
Tom: Yes.
00:57:23.589 --> 00:57:23.989
Avi Ashkenazi: I am, like,
00:57:24.057 --> 00:57:24.467
Tom: plus one
00:57:24.699 --> 00:57:28.469
Avi Ashkenazi: share every day, what's
happening inside, and, there's lots of
00:57:28.879 --> 00:57:30.209
really, interesting things that we do.
00:57:30.229 --> 00:57:35.719
I can't every day, but I am trying,
every week or two to, bring some
00:57:35.719 --> 00:57:38.479
video from the internals sh- showcase
00:57:39.518 --> 00:57:40.098
Jonny: Absolutely.
00:57:40.793 --> 00:57:41.233
Tom: Amazing.
00:57:41.243 --> 00:57:41.463
I feel like
00:57:41.548 --> 00:57:41.998
Jonny: Great
00:57:42.163 --> 00:57:44.373
Tom: the surface, but thank
you so much for your time, Avi.
00:57:44.383 --> 00:57:46.443
It was, chatting with you.
00:57:47.023 --> 00:57:49.773
great to have you here,
and we'll wrap up, there
00:57:54.504 --> 00:57:55.624
Hope you enjoyed our chat with Avi.
00:57:56.004 --> 00:57:59.284
If you have any feedback on the
podcast, good or bad, please drop us
00:57:59.284 --> 00:58:02.414
a note at podcast@nearfuture.works.
00:58:02.414 --> 00:58:03.224
We'd love to hear it.
00:58:04.044 --> 00:58:05.034
Oh, and one final thing.
00:58:05.074 --> 00:58:07.034
Last week, we announced a new
workshop we're opening up to
00:58:07.034 --> 00:58:10.214
individuals to build your own
design tools on the 13th of August,
00:58:10.274 --> 00:58:11.974
find out more and grab your spot
over at nearfuture.works/build.
00:58:14.474 --> 00:58:15.244
Catch you next week.