The Stages of Judgment-Driven Development Most of the pain I’ve seen in software wasn’t caused by bad code. It was caused by bad decisions that were never treated explicitly as such. In the last two posts, I argued that execution is no longer the bottleneck. AI made building cheap. Judgment is now the scarce resource. If that’s true, the way we develop software must change, not in terminology or ceremonies, but in where and how we place judgment. ...
What “Human Judgment” Actually Means in the Age of AI We often talk about judgment as if it were intuition, taste, or seniority, something vague that people either have or don’t. That framing is wrong. Judgment is not intuition. It’s accountability. In real systems, judgment isn’t about gut feeling or instinct. It’s about being accountable for decisions made under uncertainty. Judgment shows up in moments like deciding something is good enough to ship, deciding not to ship even though it technically works, deciding to stop a direction after weeks of investment, or deciding that a shortcut today will become an unacceptable liability six months from now. ...
When Vibe Coding Meets Reality I didn’t start thinking about this because I’m excited about AI writing code. I started thinking about it because I kept seeing the same patterns repeat over the years. This time, they come under a new name: vibe coding, the habit of describing what you want in natural language and letting an AI scaffold the system for you, often without deeply understanding the code it produces. ...
I heard Satya Nadella describe this week on the All-In podcast how LinkedIn has been collapsing product management, UX, and engineering roles into a single “full-stack builder” role. The framing was optimistic: AI reduces construction costs and speeds up delivery. At a surface level, this makes sense. AI dramatically accelerates prototyping, implementation, and iteration. In many environments, the traditional separation between PM, design, and engineering really does introduce friction that slows learning and execution. ...
Playing around with Dragon 32 code this week pulled me straight back into the 80s. That reminded me that some time ago I implemented SpaceMonger, a tiny retro space game where you fight spiders and collect coins. So I revived it and, at the same time, tested Google’s new Antigravity. You can see the result on my Projects page (Warning: it is addictive.) As for Antigravity? A disappointment. It works if you want something free while learning vibecoding, but it is not close to Cursor and very far from Claude. The ultimate humiliation was opening Claude inside Antigravity to fix bugs Antigravity could not handle. ...
From CLS to Agentic AI I was born in 1977, right at the intersection of two worlds. Old enough to remember analog. Young enough to grow with digital. My first program was written on a Dragon 32, a home computer with 32 kilobytes of RAM and a keyboard that felt like it came from a typewriter factory. My oldest brother brought it home one afternoon, and it was the first computer I ever saw in real life. ...
Metadata Then & Now: A Decade of Machine Understanding Ten years ago, I wrote a blog post for Dalet about metadata and online video advertising. In 2015, digital video was accelerating fast, budgets were moving from TV to online, ad-blocking was spiking, and programmatic was scaling, so we focused on the unglamorous layer that made it all work: metadata. Back then, my main argument was that metadata shouldn’t be entirely human or entirely automatic. Machines could process at scale, but humans understood nuance. That idea sounds self-evident now, but in 2015 it bordered on heresy. ...
Introduction For years, I wanted a place to share my thoughts on product management, operations, AI, software development, and the small “weekend projects” I build from time to time. I always wanted to, but never actually did. Why? No good reason. At first, I tried Twitter. Too short, not a good fit for full ideas. Then I moved to LinkedIn — better, but quickly chaotic. Posts disappear into the feed, long-form posts feel clumsy, and it’s hard to revisit or organize old ideas. The more I wrote, the more I felt the need for a home, a place where my work, thoughts, and experiments could live together. ...
Summary CVMatch matches candidate CVs to role descriptions using minimal infrastructure and pragmatic AI. Focus: signal over noise, reproducibility, and low operational overhead. Problem People waste time on manually checking if their CV matches job descriptions or are busy copying and pasting from/their AI chat. LinkedIn’s “AI” tool, thought to be a great promise, is cumbersome and not a good fit. Approach Embedding-based similarity for CVs ↔ role text. Lightweight prompt ranking with guardrails. Deterministic pipelines, versioned artifacts. Privacy first. Strict token budget management to reduce costs. BYOA - Bring Your Own AI Architecture Ingestion: parse PDFs/Docs to normalized text. Vectorization: embeddings store. Scoring: hybrid rank (semantic + keyword boosts). Review UI: simple shortlist with rationales. MLOps Reproducible runs (seeded). Config-as-code (YAML). Eval set and regression checks. CI for lint + smoke tests. Results Faster shortlist creation. Clear rationales for ranking. Easy to extend with domain rules. Try It Install CVMatch directly from the Chrome Web Store: 👉 CVMatch on Chrome Web Store
Summary image2sound is an artistic experiment that converts visual information into audio. Each pixel’s brightness, hue, and position are mapped to pitch, volume, and rhythm — allowing you to hear the hidden patterns within an image. Concept This project explores synesthesia through data — bridging vision and hearing using simple yet expressive Python-based tools. Features Converts PNG or JPG images into WAV files. Maps RGB values to frequency ranges. Customizable tone generation and envelope shaping. Supports batch conversions for entire folders. Try It 📦 PyPI Package: View on PyPI ...