unitedby.ai chennai • jan 2026
can ai actually
do your job?
spoiler: it's complicated. but yeah, probably.
sanju
founder, thisux
unitedby.ai chennai • jan 2026
spoiler: it's complicated. but yeah, probably.
sanju
founder, thisux
before we start
i'm not here with "ai won't replace you, someone using ai will" linkedin energy.
i'm here to show you the actual numbers. then you decide if you want to panic or pivot.
who's talking
thisux
design studio, 3 years
dunsocial
ai product, building in public
dun
ai-native suite of tools
i've automated parts of my own job. now doing 3x with fewer people. it's awkward to admit.
keep it raised if you used it for actual work, not just asking what to eat for dinner
that's the wrong question.
the right one: "how long until it can?"
from research paper to existential crisis in 8 years
timeline
where we are today
people now using ai tools globally
+64 million new users added in 2025 alone. largest year-on-year jump ever.
of large enterprises now using ai
$37B
genai investment 2025
(3x from 2024)
$3.70
ROI per $1 invested
(top performers: $10.30)
79%
of orgs experimenting
with ai agents
the fine print nobody reads
of ai projects still fail
so yes, the hype is real. so is the chaos.
5 minutes of technical stuff. i promise it's useful.
the embarrassingly simple version
input: "the capital of france is"
→ "paris" (99.7% probability)
→ "not london lol" (0.2%)
→ "definitely berlin" (0.001%)
predict the next word. billions of times. somehow becomes intelligent. we're still figuring out why.
the architecture
📝
tokenize
"hello" → [15496]
📊
embed
[0.1, -0.3, 0.8...]
👀
attention
what's relevant?
✨
predict
next token
the magic is in "attention" — it's how "it" knows to refer to "the cat" and not "the mat".
the scale is absurd
parameters in gpt-4 (rumored)
1.5B
gpt-2 (2019)
175B
gpt-3 (2020)
~86B
human neurons
honest answer: we don't fully know. which is slightly terrifying.
the scary part
abilities that appeared suddenly at scale. nobody programmed them.
we trained it to predict text. it learned to think. oops.
the not-so-fun part
may 2023: lawyer steven schwartz used chatgpt for legal research.
chatgpt hallucinated 6 court cases that didn't exist.
he submitted them to federal court. got sanctioned. became a meme.
77% of businesses worry about ai hallucinations. they should.
— everyone at openai, probably
2025-2026 numbers. fresh off the press.
q4 2025 - jan 2026
tech jobs lost
28.5% of these layoffs directly attributed to ai restructuring
the names you know
of microsoft's recent layoffs were developers
ai tools now do what junior programmers used to. that's not a prediction, that's a press release.
plot twist
klarna: replaced 700 customer service reps with ai
quality dropped. customers revolted.
they had to rehire humans.
forrester predicts half of ai-attributed layoffs will be quietly rehired — offshore, or at lower salaries.
or knows someone who has? be honest.
but here's what's real
klarna (before the drama)
ai handled 2/3 of customer chats
cursor
40% of their codebase is ai-generated
harvey ai
contract review: days → hours
midjourney
11 employees
even the lawyers are doing it
23%
legal ai adoption
2024
54%
legal ai adoption
2026
doubled in 2 years. and lawyers are supposed to be the careful ones.
what's coming (per the people who get paid to predict this)
genai adoption is already reducing entry-level hiring. the ladder is getting shorter.
imf, january 2026
not me saying it. the international monetary fund.
employee concerns about ai job loss
28%
2024
40%
2026
the anxiety is catching up with the reality.
this part might sting
the uncomfortable truth
global ai index 2024
#1
usa
#2
china
#3
uk
#10
india
we're good at adoption and talent supply. we're bad at research, infra, and building foundation models.
the bigger problem
tcs, infosys, wipro → business model is billing hours.
5.4 million it employees in india.
what happens when ai can do 40% of that work?
nasscom says 1 million it jobs at risk by 2028. i think that's optimistic.
you should be paying very close attention right now
ai is the ultimate arbitrage. it doesn't need a visa.
the disruption nobody's talking about
why this matters
phi-3 mini (microsoft)
3.8B params
runs on phone
llama 3.2 (meta)
1B-3B params
runs on laptop
gemma 2 (google)
2B-9B params
runs locally
qwen 2.5 (alibaba)
0.5B-72B params
multilingual beast
why slms change everything
🔒
privacy
data never leaves your device
💰
cost
no per-token pricing
⚡
speed
no network latency
🎯
fine-tuning
customize for your use case
not via api. locally. embedded. invisible.
2025 → 2026
<5%
apps with agent capabilities
2025
40%
apps with agent capabilities
2026
software is evolving from static to dynamic. adapt or get adapted.
this is where it gets real
the game changer
user: "book a cab to the airport for tomorrow 6am"
ai thinks:
1. need to call book_cab function
2. destination: airport
3. time: tomorrow 6:00 AM
ai calls: book_cab({
destination: "chennai airport",
pickup_time: "2026-02-01T06:00:00",
pickup_location: user.home_address
})
result: "booked! ola arriving at 5:45 AM"
function calling. ai decides what to do and does it.
the protocol
anthropic's open standard. one interface, any tool.
resources
files, databases, apis that ai can read
tools
functions ai can call
prompts
reusable templates
today: 1000+ mcp servers. slack, github, notion, gmail, figma, everything.
oct 2024
anthropic released "computer use" — ai that can:
it's clunky now. give it 18 months.
ai can learn that. it's just a matter of time.
my predictions (i could be wrong, but i'd bet money on it)
if your job is repeatable, documentable, and doesn't require physical presence...
yes. eventually. probably sooner than you think.
but here's the thing
because it's not all doom and gloom
ai job postings 2024 → 2025
ai-related job postings jumped
demand is growing faster than supply. for now.
the money talk
wage premium for workers with ai skills
same job. add ai skills. make 56% more. the math is simple.
fastest growing roles (yoy)
as the most in-demand skill in ai-related job postings
also in top 10: communication, collaboration, leadership. human skills aren't going away.
high risk roles
data entry & processing
basic customer support
junior copywriting
basic bookkeeping
translation services
qa testing (manual)
basic graphic design
report generation
lower risk (for now)
physical work
plumbing, construction, nursing
high-stakes decisions
surgery, legal judgment, crisis mgmt
relationship building
sales, leadership, therapy
novel problem solving
r&d, strategy, architecture
but "for now" is doing a lot of heavy lifting in that sentence.
the question is: are you transforming with them?
pwc 2025 global ai jobs barometer
faster skill change in ai-exposed roles vs non-ai roles
skills in ai-adjacent jobs are evolving faster. you either keep up or get left behind.
that money has to go somewhere. make sure some of it goes to you.
survival strategies that aren't bs
strategy #1
every team needs someone who actually knows how to use these tools. be that person before someone else is.
67% of jobs now require ai skills. this isn't optional anymore.
strategy #2
execution is getting commoditized.
strategy, judgment, relationships — these aren't. yet.
the higher you go, the longer you're safe. but the ladder is shrinking.
strategy #3
integrate it into everything. don't treat it as a separate tool you use sometimes.
the people who win are the ones who can't imagine working without it.
strategy #4
reading about ai doesn't help. watching youtube videos doesn't help. building with it does.
the gap between knowing and doing is where most people get stuck.
what i actually use daily
seriously. pick one thing. start tomorrow.
let me control this mac from my phone
dunmac
built today. hono + claude + sveltekit.
in 2 years, knowing ai will be like knowing excel. expected, not impressive.
start now or start catching up.
go build
anthropic prompt engineering
docs.anthropic.com
vercel ai sdk
sdk.vercel.ai
mcp servers
modelcontextprotocol.io
ollama (run models locally)
ollama.ai
questions? arguments? existential crisis?
sanju
sanju.sh — personal
thisux.com — my design studio
dunsocial.com — our saas product
unitedby.ai — chennai chapter
let's build the future together (before it replaces us)