Microsoft and Hugging Face use database validation as RL reward to train agents
What happened
On 23 October 2026, Microsoft and Hugging Face jointly released ThinkingBox-Training, which uses ThinkingBox's deterministic validation of final database states directly as a reinforcement-learning reward to train stateful business agents, and fine-tuned the Qwen3.8-27B model on that basis. The approach turns database validation results into RL reward signals, according to reports. Public information so far covers only the release and the training method itself; there are no reports yet of subsequent evaluation or application progress.
Written by AI from the reports · updated 3 h ago
Timeline
Follow the reports to see the event from each side.
- Hugging Face BlogSelectedTurning Database Grading into RL Rewards: Post-Training Agents with ThinkingBox
This is the follow up to our first blog post, The Agent Said It Was Done. The Database Disagreed. ThinkingBox-Training is now available on GitHub . We built ThinkingBox to evaluate an agent's work by comparing the final database state with the expected outcome.
Heat of this event
Current heat 9·Peak in the comparable range 9（Oct 23 10:00）·Change over 24 hours in the comparable range –
The trend compares only the same accounts observed without a break, so its range can be narrower than the current heat count. Move the pointer or tap the chart to see each hour; with a keyboard, use the left and right arrow keys.