Jacob Meltzer

Excited to share what I’ve been working on since Microsoft AI: autonomous science at Altara. If we want AI to keep scaling, it has to learn to hill-climb compute and energy through real-world, experimentally verifiable discovery

30

5 Comments

Waymo collected ~20M miles over 15 years before launching L4 in SF. That’s years of data collection, billions in capital, and a scale most companies simply can’t afford. And in many industries — especially defense — you don’t even have the luxury of commercial-scale deployment in the first place. The reality? Autonomy doesn’t fail because of models. It fails because of missing edge cases, weak coverage, and slow data iteration cycles. That’s exactly the problem we’re solving at DiffuseDrive. In this Forbes piece, Sabbir Rangwala breaks this topic down. Synthetic data isn’t about replacing reality. It’s about completing it. https://lnkd.in/dqNSA4t7 Roland PinterJordan KretchmerOutlander VCPresto Tech Horizonse2vc #PhysicalAI #Autonomy #DefenseTech #SyntheticData #Robotics #GenerativeAI

91

2 Comments

🧠 Software engineers are building the AI that could take their own jobs At avua, we’re seeing a fascinating shift from our position as a hiring platform. Software engineers are now creating AI systems that can write, debug, and deploy code, the same work they were hired to do. Every new AI agent brings us closer to a future where software builds software. 📊 What we’re seeing (as of Oct 2025): ⚡ 80% of large enterprises are now using AI across core operations (McKinsey) ⚡35% have deployed AI agents in production (PwC) ⚡ Up to 30% of software engineering roles are expected to evolve into AI oversight and orchestration rather than direct coding The irony is clear: Engineers are writing the code that could automate their own work. In doing so, they’re also designing the next generation of engineering itself, where humans and AI collaborate to build faster, smarter, and at scale.

11

5 Comments

rqlite -- and the way I develop it -- is fundamentally changing. LLMs are in the loop -- from ideation to Pull Request. https://lnkd.in/ejwXGddR

27

1 Comment

SWEs have LeetCode. Quants have QuantGuide. PMs? They now have the PMGrind.com, offering all-in-one PM prep for 24/7 practice. Two months ago, I made a post about the pains of PM recruiting that went viral. It got~1M hits, sparked 4.5K+ comments, 1K+ reactions, and filled my inbox with hundreds of DMs (and counting!). The message clearly struck a nerve: PM recruiting is broken. 🛠️ Aspiring PMs are stuck: ❌ Wasting 2 hours for every 1 hour of mock practice due to mandatory role-switching with peers. ❌ Receiving inconsistent feedback from other inexperienced candidates. ❌ Struggling to practice on their own schedule. So Aneesh Patnaik and I founded the PMGrind.com: ✅ AI-powered mock interviewing agent — simulates real PM interviews with dynamic back-and-forth (no clunky “press enter to continue”). ✅ Scoring engine trained on top company rubrics like Google, Amazon, and Meta — custom-tailored to your question type. ✅ Free — no waitlists, no awkward favors. 🚀 To celebrate the launch, I’m doing free resume reviews! Comment what company you want to see next on the PMGrind.com, and you’ll be entered to win a free review from me! 📄🔍 #productmanagement #pm #recruiting #leetcode #ai #agent #mock #interview

190

18 Comments

#AI is changing how we build and how we communicate. When a Product Owner can prototype in a few hours, the conversation shifts from “explaining the idea” to “reviewing the idea in motion.” No endless docs, no guesswork, just something you can click, test, and react to. The first stages of the software development life cycle (Agile, Waterfall, whatever you follow) exist to bridge the gap between concept and product. With AI-assisted coding, that bridge becomes much shorter: show the idea fast, then iterate, refine, and scale together. I loved hearing Pierpaolo Di Panfilo at Musixmatch talk about their experiments and what they achieved. A great laboratory for exploring how #AI can amplify creativity and strengthen communication in software companies.

3

I’ve been going deeper into machine learning lately through UC Berkeley’s professional certificate program. What’s struck me most is how quickly the "magic" of AI disappears once you start doing the actual work. It isn't just about passing assignments; it’s about the trade-offs you have to make between accuracy and interpretability, and understanding why a model works—or why it doesn't. With everything moving so fast, it's easy to feel like you're just trying to keep up. But getting into the weeds with things like classification, regression, NLP, recommendation systems, ensemble learning, model evaluation, and model tuning has changed how I look at the tools we use every day. It’s less about the hype and more about knowing how to ask the right questions when a system fails or behaves unexpectedly. The hands-on part has been the most valuable. There’s a big gap between the theory and actually spending your afternoon debugging data or feature engineering. It's been a challenge, but I'm really enjoying seeing how it all fits together. #AI #MachineLearning #GenerativeAI #BerkeleyExecEd #ContinuousLearning

12

"Claude and Cursor work just fine from Boston." That's what I told someone earlier today when they asked if the Harvard iLab was 'as good as YC.' Asking if anything is as good as YC is just the wrong question. It doesn't have to be. All you need now are some YouTube videos, tokens, and room to cook. If a space like the iLab helps someone get in that zone, find a collaborator, or put up that initial commit... that's a W

56

7 Comments

There are two types of vibe coding: "Full vibe coding" and "Semi-vibe coding". Recently had a thought-provoking conversation with Jenny Wong about #VibeCoding. Many engineers are against it, but as a serial CTO from Series A/B startups and someone who has run my own startups, I am NOT against vibe coding at all. 1️⃣ Full vibe coding, with tools like Lovable/Replit/Bolt, is a game-changer for non-technical founders aiming to build MVPs quickly and validate ideas. While critics highlight issues like AI-generated bugs, security holes, and the risk of sky-high cloud bills, these are manageable risks for small-scale, early deployment. The key is to recognize that vibe-coded prototypes aren't meant to scale; when validation is achieved, expect "big bang" refactoring or complete revamp. 2️⃣ Semi-vibe coding, which is already mainstream for experienced teams like ours at Aizzie: we've been leveraging GitHub Copilot since 2021, then Cursor last year, and now Claude Code and Kiro. Here, AI assists us in code generation, but everything remains under expert supervision. This hybrid model enhances productivity while keeping true engineering discipline intact. Embracing both forms of vibe coding can accelerate time-to-market and empower both engineers and non-technical founders, as long as we remain mindful of the trade-offs and transition at the right time.

25

1 Comment