you're scrolling through social at 11pm.
another post about "ai will replace designers." another midjourney screenshot. another "look what i made in 2 minutes" flex. and you're sitting there wondering if your 8 years of design experience just became worthless.
let me tell you something
using midjourney doesn't make you ai ready.
prototyping in lovable doesn't either.
trust me, i spent the last week
digging into this. spoke to heads of design at india’s best companies (a lot of them are growthx members). i asked how they hire. then i looked at how ai first companies like lovable, bolt, replit, cursor hire designers. everything from the job descriptions to the interview questions they are asking.
and here's my core realization
there are two types of proof of
work in ai. but it depends on
where you want to land.
ai fluency vs ai first
ai fluency
this means getting 10x faster at your current job. most of you work at internet first companies. you need to show you can leverage ai to crush your existing metrics.
ai first is different.
it's transitioning to an actual ai first company. the proof they want is completely different.
let's start with ai fluency?
designers: your math problem
no no, i am not taking a dig at your 12th std math marks (i was quite mid too). i am talking about the core job to be done for a designer.
the reality is that we are hired for impact. correct? but what does impact mean? we know this in abstract terms but let me articulate it for all of us.
let’s assume that you’re a communication or a visual designer. (product designers just follow along, you’ll be able to tweak this for yourself.)
i am obviously simplifying this ;)
but the point is to look at the input levers that impact how we create impact. are with me till here?
now, if you look at these levers.
you actually don’t own all of them. you own only quality and velocity, right? reach is completely a marketing or a product lever, and Δmetric is also a function of copy, price and users.
but quality and velocity?
they’re definitely a lot under our control. so any improvements here will directly have a effect on “impact.”
now…the real question
can ai help me with these 2 levers? because why the fuck would you want to learn random tools that don’t help here?
don't jump to random tools
because some 99rs ai course
told you to. tools are everywhere.
understanding which lever to pull
is what separates the employed
from the unemployed.
this is what we’re going to do.
breakdown quality and velocity more. then look how we can use ai to improve those two levers. that’s what we will call being ai fluent.
then we'll go deeper into
how to transition into an ai first company. that’s not the same as ai fluent.
let’s get back our original equation
impact = reach × Δ metric × quality × velocity
we will focus on quality & velocity
let’s breakdown quality first.
quality =
attention × clarity
attention breaks down to:
visual hierarchy × cognitive load × emotional hook
clarity breaks down to:
copy precision × visual metaphor × user context match
what you want to do is focus on any of these sub levers and use ai tools that help you with them. again tools DO NOT matter. you can legit just ask gpt what tools help you with “visual hierarchy”. it’ll tell you. that’s the best part about ai. if you ever get stuck, all the worlds content is in that chatbot. you can breakdown velocity the same way.
all you need is motivation.
for motivation, i cannot stress enough how important it is to surround ourselves with smart and ambitious people. legit the core reason why growthx even exists.
one designer at a d2c brand told me she went from 10 hero banner variations per week to 250+ (insane, i know). conversion lift? 23%. same design sense. 25x output.
i reccomend: create a system.
idea generation with tool XX (30 min)
rapid prototyping in tool XX & YY (2 hours)
documentation & handover with XX (30 min)
final polish and ship (1 hour)
4 hours total. down from 2 days?
and with a system you can swap out tool XX for another one if that performs better for you. it allows you to stay flexible and use the best tool for the job. don’t make the tool the job.
but udayan, isn't this just
making us push button faster?
fair question.
here's the thing though.
taste. judgment. understanding humans. making complex systems feel simple. your understanding of human behavior, your sense of what converts. these become your moat.
use ai to multiply your impact.
become the designer who ships
10x faster while thinking 10x deeper.
bad designers with ai
still make bad design, just faster.
good designers with ai?
i’ll let you answer that :)
the ai first designer
the above are all table stakes for an ai first designer. but now that you’re designing products that are ai first and NOT internet first. a lot of things change.
to understand this i’ve researched job descriptions, interview questions and spoken a couple of real humans who work in ai first companies. let’s dive in.
They care whether you can design for models that hallucinate, APIs that throttle, and users who expect magic.
what ai companies actually test for
forget your portfolio review prep.
ai first companies run a completely different playbook. i tracked interview patterns across elevenlabs, harvey, jasper, runway, and apple's ai teams. here's what they're really evaluating.
prompt and model fit
by round 2, they'll ask "how did you trim tokens to meet a 150ms target?" not because they expect you to code. because they need designers who understand the engineering constraints behind the interface. you can't design what you don't understand.
multimodal choreography
runway probes hand offs between text, image, and video during live critiques. they watch how you think about state transitions across modalities. when does voice become text? how does image generation failure cascade?
ethics and risk
rocket lawyer asks for a bias fix story in their timed assessment. before the panel even meets you. they want specifics. what broke? how did you catch it? what did you change? vague answers about "considering all users" get you rejected.
user centered metrics
jasper and harvey press for exact adoption or retention lifts. not vanity views. not engagement. real behavior change tied to your design decisions. "users loved it" isn't data. "edit rate dropped 23%" is.
system thinking
apple and meta test how you extend design systems to hold ai specific states. loading isn't just loading anymore. it's token streaming, confidence levels, fallback patterns.
cross functional literacy matters
know token limits and latency.
those are ml constraints that shape every interaction. understand fine tune vs rag choices. that's an ml plus pm decision that changes your entire ux. grasp dataset hygiene. that's research foundations determining if your design even works.
sounds like french?
think back to when you first learned web design. you had to understand lazy loading to design performant pages. you needed to know SVG versus JavaScript animations to make informed tradeoffs. you understood browser constraints to push creative boundaries. these weren't nice to have, na?. they were needed for being taken seriously in technical discussions.
ai literacy works the same way.
when you understand inference costs, you can advocate for design patterns that balance user delight with unit economics. when you grasp model limitations, you can design fallbacks that feel intentional rather than broken. when you speak the language, you shape the product. to understand foundations started i reccomend andrej karpathy’s videos on youtube
proof that gets you past screening
this where we get into proof of value territory.
this is stuff that gets you hired.
what works? live evidence of ai native thinking.
ship a live ai interaction
scope it tight. one feature marrying prompt logic, ui states, and success metrics. show prompt history. show metric deltas. show figma to production screenshots. this proves you can work within model constraints.
record a five minute demo
walk through a real prompt tweak. hit run. narrate the metric goal. hiring panels replay this during deliberations. they want to see how you think about uncertainty, not just outcomes.
document design system slots
add confidence badges. loading skeletons. fallback modals. check them into your library and link the diff commit. this shows you understand ai needs new patterns.
write a bias mitigation note
maybe a one pager. write potential harm. write guardrail copy. escalation path and sop. interviewers scan for this by round 3. ethics design is core design work btw.
a reddit thread i read really hit home. they said that great designers in ai companies design the experience of uncertainty itself. they know when a model might fail and what users see when it does. they measure success in milliseconds and confidence intervals, not just conversion rates. this quote cracked me up "i spend more time designing for when things go wrong than when they go right."
your figma skills still matter.
but at ai first companies,
they're table stakes.
the real test.
whether you can design for systems that think. systems that hallucinate. systems that surprise even their creators. designing interfaces is fine. but you’re supposed to think beyond that.
ai meets designer in three acts.
first, denial.
"it can't do what i do."
second, panic.
"wait, it just did my job in 4 seconds."
third, evolution.
"okay but can it defend a design in a meeting?"
the space between taste and execution.
ai cranks out 50 variations while you're still picking typefaces. but you know why helvetica neue feels wrong here. you sense the visual hierarchy breaking at 375px viewport. you catch the microinteraction that feels 20ms too slow.
designers will become curators.
system architects. the ones who know when the ai is lying about color theory. who can smell when the ai is just averaging pinterest boards from 2019.
whether you become ai fluent or ai first,
i think becoming ai something is inevitable.