ChessviaAI

There’s something strangely liberating about writing when no one’s watching. No pressure to perform, no expectations to meet, just you, your thoughts, and the page. And yet, I won’t lie, having a reader, even just one, feels like sunlight breaking through fog. You don’t need it to keep walking, but it sure makes the path warmer.

I think I’m learning to live in that space, to write for the freedom of it, while still holding space for the hope that one day someone will stumble across the words and feel a flicker of recognition. Until then, it’s just me, showing up. And I’m learning to be okay with that.

Thanks for putting language to a season so many of us quietly live through.

This really captures a tension that's becoming harder to ignore:

AI isn't "creating" bad practices, it's removing the barriers that used to slow down bad practices.

Vibe coding existed long before LLMs, but now someone can ship fast without knowing (or caring) how fragile their implementation is. And because demos look good and investors want speed, it's easy to mistake velocity for progress.

What worries me most isn't the code itself; it's the cultural signal it sends. If early-stage success gets decoupled from technical soundness, then short-term incentives will dominate, and we’ll lose the developers who understand how to build lasting systems.

The best teams will be the ones who don't fight AI, but who also don't let AI drive them straight into a wall.

Really interesting breakdown.

What's striking is how much long-term leverage Google has by controlling its hardware stack. An 80% cost edge isn't just about short-term pricing — it compounds over time into faster iteration, more experimentation, and the ability to undercut competitors when needed.

Meanwhile, OpenAI is betting that ecosystem lock-in (via agents, tooling, and user workflows) will outweigh pure cost efficiency. It feels a bit like early iOS vs Android: closed ecosystem polish vs open interoperability.

The wildcard is how quickly OpenAI can develop its own silicon or renegotiate better terms. Otherwise, Google's advantage might not just be hardware — it could become cultural too (faster, cheaper, wider innovation cycles).

It feels like we're watching the playbook for AI-native companies emerge in real time.

Duolingo’s approach, explicitly tying headcount to proof-of-automation limits, baking AI usage into performance reviews, and prioritizing AI-first systems over retrofitting old workflows, is a glimpse at how "AI-first" won’t just mean using LLMs as a tool, but rebuilding the entire operational model around them.

That said, it's a double-edged sword. Contract workers were crucial to Duolingo’s early scalability. Shifting to AI removes human bottlenecks, but also human nuance — and teaching language is deeply nuanced. It’ll be fascinating (and maybe a little uncomfortable) to see if mass AI content keeps Duolingo's educational quality high as they chase faster scaling.

AI-first might win on cost and speed. But will it still win on outcomes?

Really impressive engineering work her, especially running it all on a single server with in-memory board state and optimistic rollback.

I work in chess tech, but in a very different direction (structured games, coaching, serious play). It's inspiring to see chess reimagined like this!