Defog pivot: abandoning API reliance for open-source fine-tuning
andersonbcdefg · x · 2026-08-29
Defog's founder updated their model strategy. After abandoning self-trained models (despite 2M+ HF downloads) in favor of APIs due to o1, they now realize the risk of building on 'shifting sands.' With the open-source frontier moving too fast for fine-tuning to be worth it for most use-cases, and skills becoming more portable, they plan to spend a third of their time fine-tuning open-source models to own their weights.
More from coding & agent
- User burns weekly quota in a day running multiple agents — BLUECOW009 · 2026-08-29
- Agents found adding code or checks unnoticed — srchvrs · 2026-08-29
- Codex 'Persistent Mode' spotted: Continues reasoning until asleep — daniel_mac8 · 2026-08-29
- SpaceXAI Engineer Shares Grok Bot Template for Engineering Loops — mattyp · 2026-08-29
- Deep dive: The invisible privacy risks of using AI coding agents — kashifmanzoor · 2026-08-29
- Opinion: Grok Bot is the right end state for capable AI agents — petergyang · 2026-08-29