OpenAI and Anthropic may soon compete against their customers’ own data loops
bigdata · x · 2026-07-28
Ben Lorica argues that the harder challenge for OpenAI and Anthropic is not another open model beating them on a benchmark, but customers turning proprietary data and workflows into compounding in-house capabilities.
The article says the post-training stack is filling in: supervised fine-tuning, reinforcement fine-tuning, environments for practice, graders and verifiers, data generation and curation, evaluation, deployment, and monitoring. More than 25 startups are already building around reinforcement fine-tuning, but the real product is often the surrounding machinery that lets teams iterate from task definition to failure analysis to training and redeployment.
The broader thesis is that specialized AI is becoming easier to build, and frontier labs may end up competing with the very data and feedback loops their customers control.
More from Companies & People
- Cohere Labs goes live on Twitch with a lecture series on transformers and NLP — nickfrosst · 2026-07-28
- OpenAI Exec Unpacks the Agent Architecture Behind 10M Codex and ChatGPT Work Users — Latent Space · 2026-07-28
- Critics say Meta’s performance review culture is hurting engineering — mgill25 · 2026-07-28
- Ambient co-founder says open source will win and calls closed AI a bad infrastructure bet — IridiumEagle · 2026-07-28
- Conviction launches Embed, betting that small AI startups can move faster than big companies — igarciacamargo · 2026-07-28
- Embed opens applications for founders, with cash, compute and a $1B alumni record — igarciacamargo · 2026-07-28