All of my long-form articles and LinkedIn posts on tech and anything else I find intriguing, collected in chronological order. Oh Also, I try to post every week!
Every platform eventually monetizes whoever actually pays the bills — advertisers for Google and Facebook, and developers for AI. ChatGPT has ~900M users but only ~5.6% pay, while a single developer on a coding agent burns $200+/month in tokens. The consumer chatbot was the demo; the coding agent is the product — and the April 2026 shift by Anthropic and OpenAI from per-seat to usage-based enterprise pricing is the tell.
A take on building with AI coding agents: the funniest failure isn't the mistakes — every tool makes mistakes — it's the absolute, misplaced confidence ("I've carefully analyzed your requirements..." right before it rewrites three files you didn't ask it to). Once models are all good enough, the real differentiator becomes context: precision, memory, constraints, and timing — exactly what the team obsesses over at Sparkonomy.
Spotify's new anniversary icon breaks readability and brand rules, and designers are cringing — but it got everyone talking. By changing the thing people tap ten times a day, every complaint became an impression pairing "Spotify" with "20 years." They didn't design a pretty icon; they designed a conversation.
Cheap storage gave us Spotify, Netflix, and YouTube; cheap bandwidth gave us real-time collaboration and video calls replacing flights. With Subquadratic shipping the first production LLM where attention scales linearly with context (12M tokens, 52x faster than FlashAttention, 95.6% on RULER 128K), context may be the next constraint to get cheap — and the interesting question is what gets built that couldn’t exist before.
Riffing on Rémi Louf’s essay, Naad argues coding agents are overestimated as a way to make individuals code faster and underestimated as a way to make organizations externalize what they know. The new bottleneck is producing specs precise enough for agents to execute; context is the real commodity, and coherence — getting teams aligned on what to build — was always the hard problem. A lesson he sees weekly as founding engineer at Sparkonomy.
Reflections from a week at the Canton Fair in Guangzhou: tens of thousands of suppliers and a humbling pace from idea to sample to shipment. Takeaways — logistics is the whole game, showing up in person still beats a hundred emails, and AI, automation, and customization are already on the showroom floor.
On International Creators Day, Naad argues the underrated story isn't AI making art but AI making it possible for creators to keep making art. Agentic systems that read, reason, call tools, and complete tasks end-to-end can finally take over the admin — chasing payments, reconciling invoices, parsing brand deals — which is the bet at Sparkonomy: build the agentic layer underneath the creator economy.
Everyone thinks they know Git, until they encounter a complex merge conflict or a “rebase” request. If you’ve ever felt overwhelmed by Git, you’re not alone. And that’s okay.
Python is often criticized for its performance, but dismissing it as “slow” ignores its unique strengths and the productivity it brings. Let’s break down why Python deserves more credit than it gets.
The meteoric rise of artificial intelligence has led to a surge in self-proclaimed “AI experts” on social media and corporate boards. But does anyone really know what they’re talking about?