AI doesn’t fix a team or break a budget on its own — it amplifies whatever discipline already exists; the 2025 DORA finding is that “when you increase speed without investing in quality, bottlenecks simply shift downstream,” and the same logic governs cost.

AI tooling just became the third biggest line item on our P&L — right after people and cloud. Tried negotiating proper discounts with the big providers for coding tools. Without committing $200K annually, you get almost nothing. So now we’re seriously looking at self-hosting open-source models for code review instead. How are other CTOs and engineering leaders dealing with this new cost reality?

I wanted to share my experience with fable. Last night I wanted to experiment to understand how powerful fable is. So I took an algorithm that I had developed during my PhD and asked specific upgrades from Gemini, grok, opus, manus, and finally fable. Fable was completely out of this world. The way it analyzed a super complex problem, and provided extra thoughts. I didn't expect this, honestly! For analyzing extremely big, complex problems, I would say there is no comparison with the rest of the models today.

ChatGPT just hit 1 billion monthly users faster than any product in human history. That record will probably be broken very soon again!

More and more agentic flows

Cloud costs rising isn't a bug. It's the receipt for AI productivity. We gave engineers the ability to spin up real experiments at scale with almost zero friction. Result: 10x more ideas actually running, more prototypes live, more shadow workloads. Your bill is the visible proof of leverage. Most still optimize for "efficiency" while velocity compounds demand. Architected, not assembled — or the invoice will keep getting louder.

I remember two months ago, everyone was going crazy about openclaw. Who really still uses an agent on their day-to-day? Just curious 🧐

Microsoft's Majorana 2 topological chip claims 1000x qubit stability—coherence jumping to ~20 seconds via AI-optimized lead superconductors. IBM targets verified quantum advantage by end-2026, fault-tolerance by 2029. Progress is accelerating, error correction shifting from physics experiment to engineering reality. But hype cycles repeat. Real leverage comes from hybrid architectures, not isolated chips. Architected, not assembled. Automate trivial noise; explore infinite classical-quantum orchestration. Sudo scale or get left behind. Who's building the production stack? λ(0) → ∞

First crop ever

Recursive self-improvement isn’t coming. It’s accelerating right now. Anthropic’s data shows Claude turning training code 52x faster than skilled humans. NVIDIA just dropped open agent models built exactly for this. Cool. But the no-BS truth for anyone scaling real engineering teams: You don’t win the recursive age by shipping 52x more tokens. You win by architecting systems where AI handles the trivial so humans can explore the infinite. Automate the trivial. Explore the infinite. λ(0) → ∞ The teams that compound will be the ones with integration discipline — not just better models. What’s the single biggest process in your org that AI still can’t touch… and why?

Grok Video Agent in Action -> Love it