Post type: Stream of consciousness.

How sure am I? Reasonably sure, until the next model upgrade.

Note: I dumped these as voice notes into Claude and had it clean them up.

I wouldn’t want to be a young graduate or postgraduate looking for a job right now. If my LinkedIn inbox is any indication of the state of India’s job market, things must be really, really bad.

Before you jump down my neck that the plural of anecdote is not “data” and my LinkedIn inbox is not an indicator of anything, calm the fuck down. I know. I know my inbox isn’t a proper gauge of the health of India’s job market.

But the problem is that labor statistics in India suck. The complexity of this economy, especially its informal and semi-formal sprawl, makes pretty much all labor data feel useless, or at least inadequate. And I think people discount the value of anecdotes in a country like ours. Yes, there’s a real risk in extrapolating too far. But anecdotes still give you a signal. They help you make sense of what’s happening. Used alongside other data points, proxy or otherwise, they’re a reasonable input.

So here’s the anecdote: the number of young people messaging me for jobs has been going up. A lot.

What recruiters are actually saying

Whenever I meet someone building a company or someone senior on an org chart, one of the first things I ask is how they’re using AI and what it’s doing to their hiring calculus. Almost uniformly, the answer is the same: AI, and by AI I mean large language models, has either reduced the need for new hires altogether or unlocked enough efficiency that they no longer feel the same urgency to hire. In some cases, it has emboldened them to fire.

Anecdotal, yes. But the pattern has been consistent enough to notice.

It’s easy to extrapolate this to total doom. I don’t think we’re there yet. But things are changing very fast. With each new model release, priors get invalidated. If I think about the number of assumptions I held that have become false since November 2023, when ChatGPT entered the mainstream, the progress is almost absurd.

Humans have always been bad at perceiving change while they’re living through it. One of the curses of the human condition is that what seems magical one moment becomes utterly banal the next. Even in the last year alone, we’ve become numb to what these tools can do. We take them for granted. We forget how much has been compressed into a very short span of time.

My mental image for AI progress is an ominous shadow of obsolescence, growing with each model upgrade and swallowing more and more things in its wake. Tasks with clear verifiability, defined processes, and little need for human judgment are being subsumed or threatened fast. And this is still early days. Diffusion is low by most measures. Less than 15% of the world uses these tools, and the data we have is shaky to begin with.

The distinction that matters

Here’s where the conversation gets messy, because you can’t treat the job market in the US the same way you treat the job market in India. Going into the age of large language models, the labor markets in the US and Europe were in reasonably solid shape. Hiring was strong post-pandemic. There was resilience.

India was already shaky. We were producing graduates on an industrial scale, most of them with paper certificates to wave in your face but not much real employability. MBAs by the truckload. Engineers by the truckload. And the one sector that was a meaningful driver of white-collar hiring was IT: the great outsourcing machine of TCS, Infosys, Wipro, and the rest.

That is exactly the sector sitting squarely in the shadow.

Will AI kill IT outsourcing as we know it? I don’t know. It’s easy to argue yes. Maybe body-shopping and labor-cost arbitrage have a longer runway than the doomers assume. But it would be foolish to discount the threat.

So AI tools are arriving in India at a moment when we have a gigantic pool of people chasing employable work and far too few employable opportunities. And here I can point to more than vibes. The International Labour Organization’s India Employment Report. Azim Premji University’s work on youth employment. Various labor studies. They all paint a grim picture.

Some hard numbers from the 2026 State of Working India:

  • Graduate unemployment is close to 40% for young people aged 15 to 25, and 20% for those aged 25 to 29.
  • Between 2004-05 and 2023, India added roughly 5 million graduates to the labour force each year. Only about 2.8 million found employment.
  • Only 7% of young male graduates secure permanent salaried work within a year of being unemployed.

The degradation of the labor market, especially for young people, has been getting worse through the post-pandemic period. That brief post-pandemic hiring spike may have hidden the underlying rot that anyone willing to spend five minutes looking could see.

Uncertainty on both sides

Right now, AI is generating uncertainty at both ends of the market.

On the recruitment side, recruiters seem to have either frozen hiring or sharply reduced it because they’re running experiments with LLMs in their workflows. They’re seeing the efficiency gains. They’re seeing that in many cases, they don’t need new hires to do what used to require new hires. These models can either automate a process that would have fallen to a fresher, or they can supercharge the output of existing employees.

In a lot of ways, an LLM is what happens when an employee gets cloned. Press a few buttons a few times a day, and the thing does the work faster, cheaper, and often well enough. That invalidates the need for new humans and, in some cases, existing ones.

On the job seeker side, you can feel the anxiety.

The traditional playbook is dead. Cold emails, cold messages, resume blasting: none of it works the way it used to. Ironically, AI tools now let job seekers broadcast resumes at a scale that was previously impossible. Recruiters are being swamped. Many have essentially stopped looking.

I keep putting myself in the shoes of a young graduate today and asking, “What would I do to make myself seen?” What would I do if I were entering the job market right now? And I get depressed thinking about it, because I don’t have a clean answer.

The frame matters

When talking about AI, it’s tempting to reach for the old platitudes. Past technological shifts created more jobs than they destroyed. Humanity has always adapted. We’ll be fine.

Those platitudes miss the mark.

For one, it’s hard to foresee the true impact AI will have on work. Two, this technology is moving fast enough to invalidate mental models that were load-bearing just months ago. Look at coding. Senior programmers are in awe of these systems. Simon Willison, who has twenty-five years of experience, says he is barely writing code in the old sense anymore. A huge share of it is being written for him by LLMs. If Simon is in awe, that should tell you something.

I keep asking my technical coworkers about these capabilities at work, and everybody, often begrudgingly, accepts that the systems have gotten dramatically good. People are no longer writing code line by line. They’re moving up a layer of abstraction. They’re becoming orchestrators, conductors, waving a baton and coaxing the models into doing what they want.

Dror Poleg has a tweet I keep coming back to, paraphrased: both Yann LeCun and Dario Amodei are wrong. AI will create more jobs, but the transition will be unpleasant, and job security and income distribution will change dramatically, with political consequences. He had another one about how new tools redefine what it takes to be the biggest and most productive person and change both the distribution of income and the longevity of relevant skills.

https://x.com/drorpoleg/status/2045860257931366649?s=20”anecdote”

Both feel right to me.

Even if AI creates more jobs than it destroys, the transition could still be deeply unpleasant. It could still create huge amounts of social damage, dislocation, and second- and third-order effects we cannot foresee from where we stand today.

The bifurcation

AI doesn’t affect everyone the same way. Some jobs have a gooey, unsaid human element that these models may never fully internalize. But people vastly underestimate how many bullshit jobs exist that are ripe for automation. Pushing pencils, pressing buttons, clicking around random systems, and performing highly repetitive process work. Especially in India.

Bullshit jobs have always been a feature of human civilization, and they’ll continue after AI. My conjecture is that the current crop of them will almost all be automated. Then, of course, new bullshit jobs will emerge.

My broader suspicion is that human workers may eventually become a bit like gourmet food. It may become fashionable to say something was done by a human. Human assistants, human secretaries, human butlers, human-made goods, and human-crafted services: status markers for the rich. You can easily imagine an anti-technology reaction where human-created things carry a premium.

But all of that ignores the destructive effects of the transition from here to there. Right now we’re in the uncanny valley, the penumbra of uncertainty, where too many things are happening at once and we can only see a few inches ahead. That produces anxiety.

Meanwhile, at the two ends of the market, something interesting is happening. Younger workers with some baseline competence are now supercharged by these tools. If they have even a basic understanding of something, LLMs act as force multipliers, which further reduces the need to hire fresh graduates at the bottom. At the top, senior and exceptionally skilled workers are cloning themselves. They can run multiple versions of themselves and do more than they ever could before.

What’s happening in the middle, I honestly don’t know. Simon Willison gestures at the idea that the middle is especially threatened. I hope he’s right, because at least that gives us a framework to think with.

So what do you actually do?

I’ve been racking my brain for useful advice for a young person entering this market. Here’s what I have.

First, know your enemy. At this point, large language models and AI systems are your enemy in the sense that you have to understand them intimately to have a chance. Use them heavily. Understand their capabilities. Think about how to use them to solve problems for the recruiter you’re trying to reach. This is non-negotiable. You cannot afford to sit around reading dumb takes about stochastic parrots vomiting the median mediocrity of humanity. That may have been a comforting belief once. In 2026, it’s an expensive one.

Second, expand the repertoire of skills you have. Don’t be a one-trick pony.

Third, reading still commands a premium over almost any other way you could spend your time. Having an empty brain is a mistake. You can no longer lean on a prestigious degree and fancy papers, because the signal value of those is collapsing.

Fourth, let your work do the talking instead of your words. The resume is basically dead. Anyone can generate a gorgeous resume from a single prompt in Claude or ChatGPT, and anyone can now automate the sending of those resumes. The signal value has collapsed. What still cuts through is evidence of thinking and building. Study what a company is doing, pick out one or two real problems it has, and send the recruiter not just a statement of intent but proof of work. A thesis. A small solution. Something tangible.

Human networks are more valuable than ever. One of the few ways you still have a reasonable chance of getting in front of a recruiter is through an actual connection. I can say this from the other side of the table: I get multiple unsolicited resumes a day, and I have almost never hired anyone through LinkedIn. The same is true for most people I know in hiring positions. A friend was telling me just a few weeks ago about the horrendous time he’d had with people hired straight off LinkedIn. The quality was that bad.

And I think young people who can demonstrate what I can only fuzzily call thoughtful smarts will command a premium, at least until the next model upgrade. I know the phrase is vague. I don’t have a better one. I mean the ability to go above and beyond the brief, to learn relentlessly, to keep adding mental models, to fine-tune your brain so it doesn’t go dumb, and to keep seeking new knowledge and new ways of thinking. Not hustle in the cringey LinkedIn-inspirational-post sense. Real hustle: disciplined curiosity, the capacity to go beyond what’s asked, the ability to make yourself useful in unpredictable environments.

If I look at people in my extended circle who are switching jobs successfully right now, they almost all look like that.

The bigger picture

Labor statistics come with a lag. Diffusion comes with a lag. Much of this will take time to show up in hard data. Right now we’re dealing with noise.

But to my mind there are already undeniable signs that the entry rung of the job ladder is being kicked out. Whatever news story you pick up, you see more and more young graduates struggling. Even people from elite institutions are finding it hard to land work, either because the need has disappeared or because the uncertainty itself has frozen hiring at the biggest employers in a sector.

I keep telling friends and colleagues that in my thirty years of existence, there has never been a single moment in my lifetime that threatened to change my worldview about education, jobs, careers, and the building of a financial future.

Since November 2023, every mental model my generation relied on has been thrown into question, and every assumption we’ve had is up for renegotiation.

At the very least, it’s both an exciting world and a terrifying one at the same time. The ability to navigate those two poles, on any given day, when the gap between optimism and pessimism swings so wildly, might itself be the real skill.