Thoughts on AI, the tech stack, and the indie hacker journey.
OpenAI, Anthropic, and xAI run separate companies on separate infrastructure, and all three had outages within the same window on September 3. If your fallback plan is switching to a different model, this is the day that proved it isn't one.
A company that started out burning flared natural gas to mine crypto is now valued at $30 billion because it signed a five-year, roughly $13 billion GPU contract with a trading firm. That's the kind of deal keeping solo operators' cloud GPU bills where they are.
Antigravity's terms of service treat routing your Gemini subscription through a third-party agent as a breach of the agreement. Paying developers have already lost their whole Google account over it, not just their Antigravity access.
Polars 2.0's pre-release swaps the default execution engine under existing code. The dangerous part isn't the five breaking changes that throw errors, it's the one that silently stops guaranteeing row order.
A 27B open-weight model hitting roughly 1,500 tokens per second on Cerebras moves an entire category of feature from 'technically possible with a loading spinner' to actually instant. The catch is that wafer-scale inference isn't priced like a budget GPU endpoint.
A widely discussed Indie Hackers post logged roughly 5,200 agentic coding runs across Cursor, Claude, and Codex: median cost 42 cents, p99 $34.80, and a retry-loop bug that silently drained $30 in eight minutes. The fix is a hard spend cap, not a per-task cost comparison.
CVE-2026-33017 is a CVSS 9.8 unauthenticated remote code execution flaw in Langflow, and attackers weaponized it in 20 hours, before a public proof-of-concept even existed. Here's what the exploit does and the exact fix if you're running an exposed instance.
Astra is the first model OpenAI has placed at the Critical cybersecurity capability level in its Preparedness Framework, after it independently found and chained two zero-day vulnerabilities during evaluation. Access is gated behind a vetted-defender program, not a signup form.
Texas became the first major data center hub to halt new grid interconnections after speculative power requests hit 474 gigawatts, over five times peak demand. GPU rental pricing sits on top of exactly this kind of grid capacity.