Good enough is ample for most coding tasks. Understand that the global software evolution is now speeding up and we still need human operators making human choices.
somehow, i cannot imagine a computer addicted to playing minecraft. I can certainly imagine a computer helping to pick out twenty acres of raspberries during the right time of day. there are plenty of human tasks just like that best set to robots if possible. This task will take 100 humans, none of whom want to be there. Try is some time.
The principle task of the Mark One human will always be optimizing his small section of local biome. This is no small task. It means recognition, and point biome optionality. Then ordering up improvements.
You can spend a lifetime on a single acre because you get good at it. This is why we need a population of 100,000,000,000. Better still, you will get great pleasure from doing this.
Discussing the Future of AI With Warren Redlich
July 1, 2026 by Brian Wang
https://www.nextbigfuture.com/2026/07/discussing-the-future-of-ai-with-warren-redlich.html#more-210737
Anthropic AI Revenue
I, Brian, describe the explosive revenue and usage growth of frontier AI companies. Anthropic’s revenue reportedly jumped dramatically (from ~$1B to $10B ARR, then to a $50B annual run-rate) thanks to new models like Opus 4.5 and now Fable. Demand is massively outstripping supply (reportedly 5x more demand than available compute), which is driving very high pricing for AI compute.
AI ProductivityB
Huge real-world productivity gains are already happening. Developers report massive efficiency boosts (some claiming 10x–100x improvements in certain tasks). Enterprises are seeing faster software development and significant cost reductions.
Is Claude Code Better?B
Comparison of coding-focused AI tools. Claude is strong for complex, high-stakes coding tasks but expensive. Cheaper models like Grok can handle ~95% of work effectively. The recommended approach is often hybrid use cheaper/faster models for routine work and premium models for review or hard problems.
AI Compute vs AI Model
The real bottleneck and value driver is intelligence demand, not just raw model size. Inference (running the models) now dominates compute usage (90-95%). Training is still important for frontier models, but most economic value comes from widespread use. Free or cheaper models are “good enough” for many tasks.
