Stevan Ljuljdurovic

This is why software engineers still have jobs. How can you prevent your AI from producing bad architecture if you don’t know what bad architecture looks like?

Am I the only one who doesn’t understand why we need data centers? I spent all morning searching Google and Wikipedia and couldn’t find a single good reason. Then I watched a 45-minute YouTube explainer in 4K, but it was mostly about AI, and I’m just waiting for that bubble to pop. I opened 200 browser tabs, downloaded a few hundred research papers, translated three from Chinese, and uploaded 2TB of sequencing data so colleagues in eight countries could analyze it with me. Then we discussed the results on Zoom while Slack transcribed the meeting and emailed everyone a summary. Around lunch, my phone backed up 14,000 family photos, synchronized my contacts and calendar across six devices, routed me around traffic, checked the weather using live satellite data, booked a doctor’s appointment, refilled a prescription, paid a bill, transferred money between banks, approved payroll for hundreds of employees, and ordered groceries. Later, I read about hospitals retrieving patient records and medical scans in seconds, emergency services coordinating disaster responses, scientists designing protein therapies and sharing enormous datasets, utilities balancing the electrical grid, and airlines continuously rerouting thousands of flights. But these all felt like pretty niche edge cases. So I asked an AI agent to turn everything I had done that day into a Twitter post, make it increasingly insane and obvious, and optimize it for maximum engagement.

What if: - Opus 4.6 was the last Opus generation that got a lot of use by Anthropic's own employees - After that they primarily used Mythos internally - 4.7, 4.8 and 5 were RLAIFd by Mythos "teachers" - Hence why 4.6 is the last Opus gen who doesn't report back like a robot wanting to cover every potential hole another AI system would've spotted and criticized - Hence why coding style in Opus 5 also gets criticized, not only behavior in CC I have no idea if this is true, but it's plausible enough to throw it into the ring. Do your worst, I guess.

Anthropic says new Claude models will embed invisible watermarks in all generated text, everywhere Claude is offered. The watermark is part of the text, it isn't metadata: "it will travel with the text when it's copied and pasted elsewhere, and may persist through some editing." This starts with models launched on or after August 2, 2026, under an EU AI Act code Anthropic signed. Anthropic is still working on adding it to current models. The rollout is worldwide.

I'm so freakin sick of how Claude talks. It's not just annoying, it's infuriating. The honest evaluation: it's the precise mechanism driving my current headache. It's the load-bearing root cause of my daily stress. It's genuinely painful, and that matters.

I'm feeling spicy tonight so let me just say it: I think Opus 4.6 was the Opus model with the best personality and writing style. Something's off with Opus 5 - it tends to give me overly long replies, use Claude-speak way too much (e.g., "here's the honest truth"), and is too judgemental. Opus used to be a joy to talk to like a trusted friend. Not so much anymore.

I am starting to think the most important skill right now is 1. harness engineering 2. extremely large scale agent orchestration This is how you take AI agents from being equal to 1-5 SWEs to being equal to a team of 100 SWEs

Duolingo generates $97 million per month, yet I still haven't met anyone who learned a language from Duolingo It doesn't seem to solve the problem properly, so how did it become so successful?

I’m stupidly optimistic. Every time I start something new, whether it’s fitness, work, habits, I tell myself it will work out, even if the odds are stacked against me. If a negative thought enters my head, it’s suppressed automatically. I don’t always reach where I thought I’d end up, but I always end up in good places. Investing is the only area where I’m pessimistic because the downside is huge. Apart from that, there’s nothing to lose by believing you’ll win.

Here's what happens: You start programming with LLMs. It looks magical at the beginning (a la Arthur C. Clarke). With a simple natural language command, you get the AI to generate code that would have taken you hours or days. Then you want to go further. It turns out that natural language can be interpreted in different ways, and the LLM's interpretation does not necessarily align with yours. So you start adding more structure and specification. You add test cases and guardrails to make sure the AI stays within your intent and specs. Now, if you look back and squint, you'll see that you reinvented programming languages. (All that said, if you master all these aspects, you can get reliable output from AI coding agents and increase your productivity.)

Hot take: a lot more people can be replaced by AI than are currently being replaced.