Denis Stetskov

@GaryMarcus @nathanhamiel Let's add AI agent platforms to the list: techtrenches.dev/p/ai-agent-pla…

👇 @nathanhamiel and I totally called it.

Full analysis with my Claude Code session data, the complete incident timeline (Amazon, Cloudflare, McKinsey, terraform wipe), and the Zelensky parallel that ties it all together: techtrenches.dev/p/ais-announce…

@chao_mbogho Using Claude for a day and thinking professions are cooked is like test-driving a car in a parking lot and thinking you're ready for Formula 1.

I think we give AI too much credit & power. Everything AI can do was designed and programmed by the human brain. What AI does is offer possibilities & freeing up time we can engage our brain for more interesting work. I see AI as a possibility tool rather than a doomsday one.

@kepano And that edge is shrinking every quarter for anyone who stopped learning when they started prompting.

your edge is whatever you know that the models don't know

@_trish_xD Exactly the right question. 80% of candidates with 10 years experience can't explain how their own tools work anymore. Five years ago that number was 30%. AI isn't replacing engineers. It's eliminating the ones who understand. techtrenches.dev/p/the-comprehe…

interviewer: "AI can write 80% of the code now. why should we hire you?" me: "cool, so who's gonna debug that 80% when it breaks? who's gonna explain to the AI why your database is doing full table scans? who's gonna tell it that no, we can't just 'add more servers' because aws bills exist?" AI writes code. developers build systems that actually work.

@milesdeutscher Or you could just learn to build something people actually pay for. None of these tools matters if you have nothing to say and nothing to ship.

CANCEL your weekend plans. You NEED to: • Learn Claude Code • Learn Cowork (build 1-2 practical workflows) • Set up Perplexity Computer/Perplexity Finance • Optimise Cowork (plug-ins + skills) • Set up OpenClaw • Test Google AI products (Nano Banana 2, NotebookLM & more) • Experiment with basic agentic solutions (Manus) • Use AI to create a business plan/strategy/context files • Build an AI second-brain database (Notion) • Experiment with Notion Agents' *brand new* • Learn basic automation tools (MCPs, Zapier, n8n) • Learn prompt engineering - the better you can communicate with AI, the better your Outputs • Read AI articles • Dive into robotics • Research AI stocks/ETFs/investment arbitrages You have way too much to do...

@AdamRackis Seems YC does not build nothing more complicated than CRUD

lol is yc ok?

@LaconicAddress @GaryMarcus I'm pretty much sure that he tried with porn, but the result was shitty as usual.

@ftt_tech @GaryMarcus He did neither and did produce the most safetymaxxed, refusal-heavy, increasingly stupid chatbots on the market that sparked an enormous backlash that’s just snowballed with his military contract opportunism, leading to mass talent walk-out.

🔥

@adxtyahq Writing code was never the bottleneck. Understanding requirements, reviewing output, debugging production at 2am, explaining to clients why it broke. That's the job. AI writes 90% of the code and 0% of the understanding. The ratio keeps getting worse.

🚨 ANTHROPIC CEO: IN 3–6 MONTHS AI WILL WRITE 90% OF CODE this was 12 months ago btw.

@GaryMarcus @sama The apology won't come. Admitting scaling hit a wall means admitting the $1.4 trillion in infrastructure commitments was a bet, not a plan.

Dear @sama, You owe me an apology. You have relentlessly, publicly and privately, attacked my integrity and wisdom since my 2022 paper “Deep Learning is a Hitting a Wall”. But in your own way you have just come around to conceding *exactly* what I was arguing in that paper: that current architectures are not enough, and that we need something new, researchwise. beyond a scaling (a “megabreakthough” in your words below). That’s all I was trying to say. And I was right. And you should be man enough to admit it. Gary cc @_KarenHao

@qikipedia Can confirm. AI made my team produce in a day what used to take a week. Great, except now I'm reviewing 5x the code in the same hours. The output got faster. The human bottleneck didn't.

A recent study has found that in the tech industry, AI has increased the number of hours that people work and the intensity of their work, risking burnout and poorer decision making.

@AnishA_Moonka 74.5% exposure, 35% actual automation. That gap is the whole story. I broke down why the math on AI replacing developers doesn't work. The bottleneck is never the code. techtrenches.dev/p/the-country-…

94% of tasks in computer and math jobs are theoretically automatable by AI. The share that’s actually being automated right now: 35%. Karpathy just released a tool scoring 342 US jobs for AI exposure, and the treemap is genuinely useful for seeing which parts of the labor market are most vulnerable. What it can’t show you is the gap between “could be automated” and “is being automated.” Ten days ago Anthropic published the most detailed study yet on AI and jobs, built from actual usage data across millions of Claude conversations matched against 800 occupations. Computer programmers sit at 74.5% real-world AI exposure, the highest of any single job. Customer service reps: 70.1%. Data entry keyers: 67.1%. After that it drops fast. 30% of American workers have literally zero AI exposure in the data. Cooks, mechanics, bartenders, lifeguards. If your job requires showing up and doing something with your hands, AI hasn’t touched it yet. Mass layoffs haven’t happened. Anthropic found no increase in unemployment for highly exposed jobs since ChatGPT launched in late 2022. The real signal is quieter. Entry-level hiring for workers aged 22 to 25 is slowing in exposed fields. Every 10 percentage point increase in AI exposure tracks with 0.6 points less job growth projected through 2034 according to BLS. Jobs are slowly not being created (not vanishing as of yet) One detail on methodology: Karpathy scored jobs using an LLM, similar to the 2023 OpenAI/UPenn paper that first mapped AI job exposure. Anthropic tested that theoretical approach against real usage data and found that LLM-based scores alone don’t match what labor economists independently project. Only the real-world metric, built from what people actually use AI for at work, tracks actual employment trends. Gen Z is already pricing this in. A Jobber survey found 77% say it’s important their future job is hard to automate. 60% plan to pursue skilled trades this year per a Resume Templates survey. Plumbing and HVAC businesses pull median gross sales above $390,000. The generation reading Karpathy’s treemap is voting with their career choices.

@xAaron11x @TukiFromKL theguardian.com/world/2026/mar…

@ftt_tech @TukiFromKL Didn’t the Iranians blow the school up?

🚨 Do you understand what just happened? A guy in Florida asked ChatGPT to sell his house. Not help him sell it. SELL IT. Pricing, Marketing, Showings, Contract.. Everything. Sold in 5 days. The average real estate agent takes 6% commission. On a $400K house that's $24,000. For what? Putting it on Zillow and sending you emails? This man paid $20 a month for ChatGPT and did the whole thing from his couch. Real estate agents spent 20 years telling you the process is too complicated for normal people. That you need a professional. That the paperwork is too risky without them. It was gatekept. And a chatbot just kicked the gate open.. Lawyers are next, Accountants are next, Financial advisors are next.. Every profession that made you feel stupid so you'd pay them is about to get the same treatment. The middleman economy is dying. And it's not dying because AI is smarter than them.. It's dying because AI proved you were always smart enough to do it yourself. You just weren't allowed to..

@naval Software ate the world. AI ate the software. Now nobody understands what the software does anymore. techtrenches.dev/p/the-comprehe…

Software was eaten by AI.

@GaryMarcus Unlimited budgets, unlimited compute, unlimited hype. Still can't ship a product that works reliably. Turns out the bottleneck was never the money. techtrenches.dev/p/big-techs-36…

Scaling is not all you need, illustration #3761: two of the two richest guys in the world tried to build AGI with basically unlimited budgets … and failed.

@GergelyOrosz Coding is solved, btw

One thing that endlessly frustrates with Anthropic, a $300B+ dollar company, where most code is written with AI: Their landing page for paying customers, Claude .ai has been broken for weeks UX-wise, and no one notices or cares or fixes: It "loses" stuff I type while it loads:

@TechLayoffLover And she is not exists

A director of engineering at a major cloud platform just messaged me about something that's keeping her up at night Last September she got the call: "We need you to optimize your team for the AI era. Cut your 12-person backend team down to 3. Use Copilot, Claude, whatever it takes. Same deliverables expected" She did it. Laid off 9 engineers she'd hired herself. Kept her 3 strongest seniors and armed them with every AI tool in existence Her team's velocity actually increased 40%. Deploy frequency went from twice weekly to daily. Bug rates dropped 23% because the AI caught edge cases humans missed Got promoted to VP Engineering in February. Glowing performance review. "Exceptional leadership during digital transformation" Yesterday she got a calendar invite. Same conference room. Same two VPs who congratulated her 6 months ago "We've been analyzing our engineering leadership structure. With AI handling more architectural decisions and our offshore partnerships scaling up, we need fewer management layers" Her role is being "consolidated" with two other VP positions. One person will manage what used to require three The meeting ended with them asking her to document all her processes, decision-making frameworks, and team management strategies "for knowledge transfer" She's building the playbook for whoever replaces her. Probably someone offshore making 1/4 of her $280k salary The best part: they want her to train her replacement on "AI-enhanced leadership methodologies" She optimized herself right out of existence