A guy DM'd me 2 months back. Laid off 8 months ago, still nothing. "Should I fudge my dates? Everyone says a gap is a death sentence." I told him not to lie. Then I asked what he'd actually done in those 8 months. Long pause. "Nothing. Applied. Got rejected. Applied more." That's when I told him the thing nobody tells you. The gap was never the problem. Recruiters don't reject you for HAVING a gap. They reject you for having a gap you can't explain. A gap with a story is context. A gap with silence is a red flag. Same 8 months, completely different interview. Right now his answer was "I was looking." That answer tells a recruiter one thing: nobody wanted him for 8 months. So I gave him three weeks. Build one small thing. Ship it. Write 300 words about what broke and how you fixed it. Put the dates on it. He said it felt fake. Like a cover-up. It isn't. You're not hiding the gap. You're FILLING it. The three weeks are real. The project is real. The debugging is real. And it changes the answer from "I was looking" to "I was rejected a lot, so I spent the time building. Here's what I shipped." One of those makes an interviewer nervous. The other makes them curious. He shipped a scrappy little tool. Nothing impressive. But he could talk about it for ten minutes without pausing. Two interviews in the next month. Both asked about the project. Neither asked about the gap. Here's what I want you to take from this. A gap isn't a hole in your resume. It's a blank page. The only question is whether you let someone else write on it, or you do.

Your salary in Indian tech is decided by a band. Level, years, company grade. HR picks a number inside it. You negotiate 5% and call it a win. Then I read one line in a 2026 GCC compensation report and haven't stopped thinking about it: For frontier skills, compensation stops being a band and becomes a negotiation. Not a BETTER band. No band. Look at the two sides of that line. Senior backend engineer in Hyderabad: ₹35 to ₹50L at a GCC, ₹18 to ₹24L at a mid-size services firm. Same title, same city. AI specialists earn 20 to 40% more than general engineers at the SAME level, purely because the pool is thinner. But here's what almost everyone gets wrong. People read this and conclude "learn AI." That's the easy half. The GCC reports keep flagging a different weak point. Mid-level people with deep technical skill who can ALSO own delivery end to end. Not one or the other. Both, in the same person. India has 2.36 million GCC professionals. We are not short of engineers. We are short of engineers who can own an outcome. AI skills get commoditised every model release. Ownership doesn't. So stop collecting skills. Start collecting outcomes you owned end to end. The scarcity isn't in what you know. It's in what you've been trusted to run

Two of the best engineers alive looked at AI coding agents this year and reached opposite conclusions. Five days apart. George Hotz, the guy who jailbroke the iPhone and built Tinygrad, spent six months running agents on real work. Firmware reverse engineering. His own deep learning framework. His verdict: the agent frontloads all the progress, then hands you a "slot machine lever." You pull it and hope it finishes. It never quite does. Five days earlier, Andrej Karpathy joined Anthropic's pretraining team. Same guy who was skeptical of agents a year ago. He reversed. Says agents crossed from unreliable to functional and the era of humans typing code into editors is ending. He described handing an agent one dense prompt. SSH setup, benchmark a model, build a dashboard, wire up services, write the report. 30 minutes. He didn't touch anything. Both of these people have earned the right to an opinion. So who's wrong? Neither. And that's the uncomfortable part. Look at the data sitting between them. MIT Tech Review found AI-generated code runs 2.2x larger than the human-written equivalent. Verbose. Copy-paste heavy. PR sizes grew 33%. Lines per developer went from 4,450 to 7,839. METR ran a controlled study. Developers FELT 20% faster with AI. They were measurably 19% slower. And production engineers keep reporting the same thing. Debugging AI-generated code takes roughly 3x longer. So the output went up and the throughput didn't. Here's the reconciliation nobody's saying. The variable was never the tool. It's whether you can break a problem into pieces small enough to verify. Karpathy said it himself. The people who decompose work well for junior engineers decompose it well for agents. The ones who micromanage humans micromanage machines and get about as far. Agents don't flatten the skill curve. They STEEPEN it. If you can specify precisely and catch a wrong answer fast, you got a 30 minute superpower. If you can't, you got a slot machine. And you'll pull the lever all day while your PRs get bigger and your ship rate stays flat. Same tool. Opposite outcomes. The difference is entirely you. So stop asking whether AI coding agents work. Start asking whether you can decompose a problem and verify an answer. That's the skill that decides which side of this you land on. And it's the one thing the agent cannot do for you.

An Anthropic researcher resigned today and said both OpenAI and Anthropic are "gambling with our lives." 46 million views in a few hours. That's not the shocking part. The shocking part is what happened next. Anthropic's own alignment lead replied in public and AGREED with him. Said he personally puts the odds of AI killing everyone at over 10% this decade. And that they still don't have a plan to solve alignment for superintelligence. Read that again. Nobody denied it. A senior person inside the company confirmed it and added a number. Normally when someone quits and calls their employer reckless, you get a rebuttal. A statement. A "we disagree with his characterization." Here you got agreement. THAT is the story. Both sides agree the risk is real. They only disagree on what to do about it. The guy who quit says stop. The people staying say if we stop, someone less careful builds it anyway. That's not a safety debate. That's a prisoner's dilemma with everyone's life in the pot. But here's the part that should bother us specifically. This entire argument is happening between maybe a thousand people in San Francisco. India has 5 million+ people in tech. Our jobs, our industry, our next twenty years, all of it gets decided by the outcome of a conversation we are not in. We are the largest downstream consumer of this technology on earth with almost zero voice upstream. One line from his thread stuck with me. He said a gamble this size should not be getting launched from a private company's Slack. He's right. And it definitely isn't being launched from ours. I don't know who's correct about the timelines. Nobody does. But when the man sounding the alarm and the company he's warning about both say the danger is real, "this is all just hype" stops being an available position. Stop treating this like someone else's argument. It's the one that decides what your career looks like in 2030.

I don’t know how people don’t realize this yet, BUT Astra can be a game changer for job seekers. Astra doesn’t just chat. It operates a computer. It browses, fills forms, builds decks, and runs multi-step workflows across real apps. Think about what that does to a job hunt. Tailoring your resume to 30 different JDs? Astra reads each straight from the opening. posting your resume and rewrites for the exact keywords, in your style, in one pass. No projects to show? Point it at a real problem and PAIR with it to actually build and deploy something this week. Not vibe-coded slop. A working project you understand, shipped 5x faster. It deletes the 80% of job hunting that was pure friction, so you can spend your energy on the 20% that actually gets you hired: real skill, real projects, real conversations.

OpenAI launched GPT-6 Astra. EVERY dev in my DM is panicking. They say: “Welcome to the AGI era.” “Best model for software engineering to date.” Here’s what the benchmarks actually say: On agentic coding (DeepSWE v1.1): Meta’s Muse: 75.4% ASTRA: 74.1% Gemini 73.8% Opus 5 73.7% The “generational leap” couldn’t beat Meta, and sits in decimal points range to Opus/Gemini. The labs are CONVERGING, not diverging. And a 74% score means it fails 1 in 4 real tasks. That failure rate is why you still have a job. Astra isn’t your replacement. It’s your leverage. Learn to drive it.

A fresher DM’d me few months ago, almost in tears. “I’ve solved 600 LeetCode problems in 4 months. Still zero interview calls. Is the market just dead?” I asked him one question that changed his whole approach. “Show me one thing you’ve BUILT.” “I don’t have projects. I’ve just been grinding DSA. Everyone said that’s how you crack tech.” There it was. The misconception quietly wrecking freshers in 2026: “If I solve enough problems, the jobs will come.” That was true in 2019. It’s a trap now. Here’s what I told him. In 2026, AI can solve 600 LeetCode problems in an afternoon. Your ability to reverse a binary tree is no longer enough. It stopped being a signal. What companies CAN’T get from AI is proof that YOU can take a problem and ship a working thing. That’s the scarce skill now. Then he hit me with the excuse I hear every single day: “But I don’t have a unique idea. Everything’s already been built.” Wrong. And this notion kills more freshers than bad DSA ever will. You don’t need a unique idea. You need a working project. Build a Twitter clone. A URL shortener. A todo app with auth. A weather dashboard. Nobody cares that it’s been done a thousand times. They care that YOU did it, end to end, and CAN explain and defend every decision amd every part of edge case you solved An interviewer has never once said “this is unoriginal, rejected.” They say “walk me through how you built this.” A clone answers that perfectly. Originality is a founder’s problem. Proof-of-skill is a your job. So I gave him a 3-week reset. Pick any common project. Build the ugliest version that actually works. Deploy it. Break it. Fix it. Understand every line. Then put THAT at the top of your resume, above the DSA. He built a basic expense splitter for his hostel. Been built a million times. Didn’t matter. He deployed it, got his friends using it, and hit one nasty concurrency bug he actually had to understand to fix. One line on his resume: “Built, deployed, and debugged in production.” He got 3 interview calls in the next 10 days. Same person. Same DSA. The only thing that changed is he finally had proof he could BUILD, not just solve. Freshers, hear me: LeetCode gets you THROUGH the door. It was never the thing that gets you TO the door. And your project doesn’t need to be original. It needs to be REAL. In an AI world, nobody’s impressed that you can solve. They want proof you can ship.

People don’t realize how much the taste matters in software development now. The taste in usability, ui, design, data. You have all the answers - just need to know which buttons to press.

ONE number changes how every fresher should India job-hunt: Freshers are now just 13% of India’s active tech openings. Meanwhile AI-specific roles went from 6.5% of vacancies (March 2025) to 16% (July 2026). The market has moved. Learn and apply for AI specific tech & roles.