simstack

Insights on how AI engineers actually get made & HIRED: This interview that Gergely Orosz did with Janvi Kalra (now at OpenAI) is a true masterclass on frameworks, career bets, tactical growth & learning in public. If you’re a junior or mid-level engineer trying to break into AI—or just grow in this brutally competitive market—you must watch this interview! (& take a lot of notes!!) It’s not another “inspiring journey” post. It’s a full blown playbook. Janvi breaks down real-world frameworks on how to: • Evaluate startups like an investor • Navigate between big tech and startups strategically • Upskill into AI without waiting to be picked • Combine infra, product, and user thinking like a full-stack founder Here are my 5️⃣ favorite takeaways: 1️⃣ Startup due diligence is your responsibility > “You’re paid in equity. That makes you an investor.” Janvi uses a 4-part rubric to evaluate companies: Revenue trajectory, Market size, Customer obsession, Competitive edge She doesn’t just trust the pitch—she checks Reddit, talks to real users, and builds her own conviction. 📌 Takeaway: Don’t join a startup for vibes. Join with a thesis. 2️⃣ Didn’t get picked? Build anyway. > “After a polite no from the AI team, I spent 5 months learning LLMs, building projects, and blogging.” Janvi studied how ChatGPT works, joined hackathons, fine-tuned models on Replicate, and wrote about her learnings. 📌 Takeaway: Real interest shows up in your free time, not your job title. 3️⃣ AI engineering is still software engineering! (This is probably my favorite insight!) > “Prompting, fine-tuning, hosting… all built on top of core engineering practices.” Being an AI product engineer isn’t about model research—it’s about using models to build real, valuable products. And yes, writing reliable code still matters. 📌 Takeaway: Don’t confuse hype with fundamentals. AI fluency is built, not bought. 4️⃣ Big tech vs. startup? Optimize for your learning curve. Big tech = moonshot projects, credibility, faster green cards. Startups = breadth, chaos, and shipping code that matters. 📌 Takeaway: There’s no one right path. Just make sure you’re learning fast. 5️⃣ Hackathons are underrated accelerators > “In Buildspace, we didn’t win by building cool stuff. We won by getting users.” Janvi learned more from testing AI products in the wild than any course or tutorial. 📌 Takeaway: Shipping + feedback loops beat passive learning every time. (Link to the full interview in the comments) P.S. Not as good as Janvi, but 5 years ago I shared my own career framework in this long-form article based on a 2019 talk. Still one of my most-read pieces (also in the comments)

Last week we teamed up with EDEM Escuela de Empresarios and simstack for one of my favourite ever Valencia Codes events, and even better this time I got to be in the audience! You wouldn't let a pilot land a plane without hours of practice in a simulator. In fact, if you're landing in rough weather, you would hope they would have practiced a bunch of times how to deal with crosswinds long before they actually attempt it with you in the back. So why do we throw engineers into complex situations without the chance to practice? Sure, you can have Designing Data-Intensive Applications on your bedside table, you can follow a tutorial that basically ends up as a copy-pasting exercise, or you could also just wing it hope Gippity comes through for you when it matters, but this is the equivalent of practising in simulator with blue skis and a light breeze. What does it feel like to SSH into a server that is really getting pounded with massive network traffic, no documentation and no clue where to start? That's what the team at simstack have created. It's like an escape room for SREs where you SSH into real, live, overloaded infrastructure with just your wits and a vague memory of some unix command syntax to try to get it back online again. It's the driving rain, gale force wind flight simulator for engineers that you didn't know you needed and I HIGHLY recommend you check it out. I had so much fun at this event (especially since I was able to just sit back and participate) and the feedback from the community was just amazing. Thanks so much to the simstack team for running the session and Frances Morales Velilla for hosting ❤️ Next up from Valencia Codes will be our July summer party 🏖️ 🥳 which you won't want to miss so watch this space!!!

This didn’t age well: "If I look at coding, programming, which is one area where AI is making the most progress. What we are finding is that we're 3 to 6 months from a world where AI is writing 90% of the code. And then in 12 months, we may be in a world where AI is writing essentially all of the code." - Anthropic CEO, Dario Amodei, just over 3 months ago. Why do we keep listening to these expert beginners? Check out Dario Amodei 's LinkedIn profile, his entire experience as a software engineer is 20 months part time, writing just 14k lines of code! I originally wrote about what he said 3 months ago, noting: “And if he really believed this he wouldn't be hiring software engineers for Anthropic. By the time they were hired, onboarded and up to speed in 3-6 months they'd be replaced with AI.” So 3 months on from his claim that “we're 3 to 6 months from a world where AI is writing 90% of the code”, where are we? Well Anthropic continue to have many open roles for software engineers and still doesn't want you to use AI in your application. Sundar Pichai claims AI is writing about 30% of code and making Google engineers 10% more productive. Microsoft CEO Satya Nadella said that 20% to 30% of their code is written by AI. But a post on Reddit three weeks ago drew our attention to Microsoft's AI agents' public work on the .NET runtime repo. Three weeks on from that all the PRs are closed and unmerged. The comments make interesting reading. And Builder.ai went bust, most likely due to inflated revenue figures but with claims that the AI was not as good as advertised / not AI. I’ll be revisiting this post again in 3 months and 6 months (to hit 6 and 12 months after Amodei’s claim). I think it won’t age well….