EUROPEAN DYNAMICS

The most interesting part of the SpaceX/Cursor news is not the price tag. It is the lock-in story hiding underneath it. CNBC reports that SpaceX can either buy Cursor later this year for $60 billion or pay $10 billion for the work they are doing together: https://lnkd.in/d3vUuzxq Right now, tools like Cursor are valuable partly because they sit above the model wars. They give developers a layer of choice. You can compare models. Switch when pricing changes. Route around weaker performance. Keep some leverage. That matters. Because once the harness, the distribution, and the model provider start collapsing into the same company, "multi-model" can turn into a transitional feature. That is how optionality disappears in software. How long do you think it will take until you can select only xAI models in Cursor? #AI #DeveloperTools #ProductStrategy #SaaS #VendorLockIn

My current AI maturity test is simple: If your team says "we need an AI strategy" before it can answer "which spreadsheet is the real one?", you are not planning automation. You are planning a faster way to spread confusion. AI is amazing. Version control for operations is still underrated. What is the most gloriously human bottleneck in your AI rollout? #AI #GenerativeAI #Operations #ProductManagement

Vibecode mode: ON. ⚡ I turned my CV into a small playable game: a magic world where you can walk through my experience, explore milestones, interact with quirky NPCs, and collect ingredients for the magic soup the cook is waiting for at the end of the level. This was easily one of the most fun projects I’ve built lately because it combined creativity, storytelling, and code in a way that felt genuinely playful. On the technical side, it was also a great excuse to dust off and sharpen my JavaScript, HTML, and CSS skills while building the game logic and deploying it with a Node.js backend. I deployed the project to Microsoft Azure App Service and configured a custom domain so the game is publicly accessible on the web through a production-ready hosting setup. At the same time, I experimented seriously with Anthropic 's Claude Code as an agentic coding tool and got a much better feel for how powerful it can be when used properly. Projects like this make me even more excited about using such AI coding agents alongside my own knowledge and experience to keep building creative, polished, and useful software. I’m always open to discussions and collaborations around bringing creative ideas to life, so feel free to message me if you’d like to build something together. Check it out here: https://lnkd.in/dzFen8NT More fun things to build next. 🚀

Last week we had one of those releases that looked boring in CI and then got annoying in prod. No disaster. Just a dependency problem, a missed edge case, and the usual "well, there goes the afternoon" feeling. That kind of thing keeps me from getting too excited about software velocity claims. Yes, code is easier to produce now. I am not sure confidence is. InfoQ had a piece on Agoda saying coding was never the real bottleneck anyway. Specs and verification were: https://lnkd.in/d2pKPN-T . Then the recent Trivy supply chain story showed how even the security tooling can become part of the mess: https://lnkd.in/d7yAVVeJ . So when people ask me where DevOps, DevSecOps, and QA fit now, my honest answer is: right in the middle. Not after the work. In the work. Because if a team cannot explain the change, test the weird stuff, trust the dependencies, and back out cleanly, then shipping faster is mostly a nice story. Are you seeing that too, or is your team still mostly bottlenecked on writing code? #SoftwareDevelopment #DevOps #DevSecOps

Opus 4.7 is here: The story is shifting from smartest model to "Can I trust it on long, messy work without babysitting it"? Anthropic says Opus 4.7 is better at hard software engineering tasks, follows instructions more closely, checks its own work before reporting back, and stays more consistent on long-running workflows: https://lnkd.in/dUD5AkQf That last part matters most to me. A lot of AI discussion still revolves around leaderboard jumps. Real adoption usually breaks somewhere else: loss of context hidden uncertainty weak recovery from tool failures constant human supervision If those failure modes improve, the product changes shape. You stop treating the model like a clever autocomplete and start treating it like a teammate you can delegate real chunks of work to. That does not mean full autonomy is solved. It means the bar is moving from impressive demo to reliable enough for real workflows. What are you seeing in practice, are the biggest gains now coming from raw intelligence, or from models that stay reliable over long tasks? #AI #GenerativeAI #LLM #SoftwareEngineering #ProductManagement

AI may be hitting a problem compute can't buy its way out of. People. Not model eval people. Regular people. This week CNBC pointed to polls showing the public getting more uneasy about AI, and not by a little. In one NBC News survey, 57% of registered US voters said the risks outweigh the benefits. Quinnipiac had 55% saying AI will do more harm than good in daily life. https://lnkd.in/dxNn4w-m WIRED also had a piece on a fight over an Illinois liability bill. Different story on the surface, same signal underneath: the market is moving from wow to ok, but who is responsible when this thing breaks? https://lnkd.in/d4Z4QQvd My guess: the teams that win the next phase will not just ship smarter models. They will ship systems that people can explain, audit, and trust. That is less exciting than a benchmark chart. It is also a lot closer to how real adoption works. How are you seeing this inside your company: is trust becoming the harder problem than capability? #AI #GenerativeAI #Trust #ProductManagement #Leadership