Dev Samaaj

Tech & AI
Instagram: @realpankajk

[Dev Samaaj @devsamaaj]

Can a tech guy get's 1000 likes?

[Dev Samaaj @devsamaaj]

JEV kya h bruhh

[Dev Samaaj @devsamaaj]

Last night I was wondering...

What if AI had been sitting there with Newton, under that tree, in 1666?

Will it ever wonder why the apple falls down? No.

AI is not curious the way humans are. It's a goal-driven system.
Give it a goal and it wins easily — faster calculation, faster processing, zero fatigue.

But it doesn't pick the question. Someone has to hand it one first.

Newton wasn't given a goal that day. Nobody told him to look at the apple. He was just sitting there, and it clicked.

Humans discover. AI executes once the discovery is handed over.

Can that change as AI gets more autonomy? Curious what you think.

[Dev Samaaj @devsamaaj]

The math is not mathing. 🤨

Last year it was all about efficiency. Now AI budgets exist and token consumption itself is becoming a flex. I saw someone feed Opus 4.8 an entire Excel sheet just to generate a pretty PDF.

Dude. There's a button for that. It's called Export as PDF. Free, instant, zero tokens.

There's already a term for it: tokenmaxxing.

The real problem is we're using our most capable models for everything instead of matching the task to the simplest reliable tool. Agents make it worse — a simple task becomes read → call → inspect → retry → validate → reread → fix → validate again.

Token prices are falling, but if tokens per task keep exploding, cheaper tokens ≠ cheaper AI.

Better question: what's the cheapest reliable way to solve this task? Sometimes it's an agent. Sometimes it's Python. Sometimes it's a smaller model.

And sometimes, it's the damn Export as PDF button. 😂

[Dev Samaaj @devsamaaj]

@triviwritescode we are becoming very lazy and it's not too far when most of human brains become more lazy and will not do anything the gap of human intelligence will increase between humans as gap of poor and rich.

[Dev Samaaj @devsamaaj]

The math is not mathing. 🤨

Last year it was all about efficiency. Now AI budgets exist and token consumption itself is becoming a flex. I saw someone feed Opus 4.8 an entire Excel sheet just to generate a pretty PDF.

Dude. There's a button for that. It's called Export as PDF. Free, instant, zero tokens.

There's already a term for it: tokenmaxxing.

The real problem is we're using our most capable models for everything instead of matching the task to the simplest reliable tool. Agents make it worse — a simple task becomes read → call → inspect → retry → validate → reread → fix → validate again.

Token prices are falling, but if tokens per task keep exploding, cheaper tokens ≠ cheaper AI.

Better question: what's the cheapest reliable way to solve this task? Sometimes it's an agent. Sometimes it's Python. Sometimes it's a smaller model.

And sometimes, it's the damn Export as PDF button. 😂

[Dev Samaaj @devsamaaj]

If you are building something in tech, drop the link👇

[Pankaj Kumar @realpankajk]

To kaise hai ap log

Idea credit: @ravish_journo

[Dev Samaaj @devsamaaj]

@creative_nashra when you start writing on paper your thoughts will slow down.

[Pankaj Kumar @realpankajk]

There are many things that do not make sense, but top leaders in their domains only talk about them when it affects them directly.

[Pankaj Kumar @realpankajk]

Indians feeling right now after UPI fee above ₹2000 transaction amount.

Hum loot gye hai or lut jane do.

[Pankaj Kumar @realpankajk]

4.5 years of frontend, never really wrote a class.

React runs on hooks and composition and you can skip OOPs entirely.

Started Python for FDE this month. Turns out classes model real things: users, transactions, state that changes.

React never made me see that, not even once.

[Dev Samaaj @devsamaaj]

Can a tech guy get's 1000 likes?

[Dev Samaaj @devsamaaj]

How many are actual smartest?

[Dev Samaaj @devsamaaj]

Hi smartest dev,

I post content on AI & Tech

Let's connect.

[Dev Samaaj @devsamaaj]

OpenCode reportedly crossed 120,000 GitHub stars in 2026. Worth asking why.

Cursor's pitch is a better editor. Copilot's pitch is deeper IDE integration. Both bet you'll pick a home and stay there.

OpenCode's pitch is different: don't pick a home. Plug in whatever model you're already paying for — GPT, Claude, something local, whatever ships next month — and run it from your own terminal, your own editor, your own setup.

The model contract underneath now matters more than the editor on top.

Free here means no subscription fee. You still pay per token for whichever model you bring — nobody's giving away inference for free.

After three coding tools shut down or repriced this year, choosing your own model looks like insurance now.

[Dev Samaaj @devsamaaj]

I tried so hard got so far

But in the end

Hook content works not informative 😔

[Dev Samaaj @devsamaaj]

If AI disappear today

I will get my job back 🙂

[Dev Samaaj @devsamaaj]

More than a 1-in-10 chance AI wipes out humanity. That's the number an Anthropic alignment lead put out this week, punctuated with an exclamation mark, days after another researcher quit the company warning it was moving recklessly toward self-improving systems.

The internet had its usual reaction: is this science, marketing, or both?

Here's my problem with it. Every time this debate resurfaces, it swallows the room. AGI, superintelligence, extinction risk — heavy words that make everything else sound small by comparison.

Meanwhile the actual, present-tense damage keeps getting starved of oxygen:

→ Junior dev roles quietly disappearing as agents absorb the work that used to train people up
→ Data centers pulling power and water at a scale most cities weren't built for
→ Support, content, and QA teams getting hollowed out with zero regulatory conversation happening
None of that needs a percentage attached to be worth taking seriously. It's already measurable. It's already happening to people who don't have a podcast or an S-1 filing to make their case.

I get why the doom framing is seductive. For researchers, believing you've built the most consequential thing in history is a very human temptation. And for companies about to go public, sounding scary plays well with investors who reward capability signals over caution.

But it lets everyone off the hook for the boring, fixable stuff happening right now. You don't need a 10-year extinction timeline to justify better labor transition policy, transparent compute reporting, or basic guardrails on what agents can touch unsupervised.

Extinction risk might be real — I'm not qualified to rule it out. But it's also the easiest risk to talk about, because nobody's accountable for it yet. The harms already showing up in job boards and power grids have names attached. That's probably why they get less airtime.