Progress Verified. Completion Pending.
The week’s trending paper, Apodex 1.1, claims to scale agentic intelligence for complex work. The twist: the brains didn’t get bigger; the executable environments the agents run in did. The system trains agents to coordinate long-horizon jobs across stages, keeping a running state log so a derailed task can be recovered and picked back up. Our four-year-old chief took the paper’s advice literally, scaled up the environment, and deployed the whole operation to what she considers the gold standard of complex work: a galaxy made of blocks.
Source: Apodex 1.1: Scaling Agentic Intelligence for Complex Work · hf-trending
▶ Animated edition: the characters speak for themselves (Korean audio, English subtitles included)
What this means for ThakiCloud
Apodex 1.1 points at the same lever ThakiCloud pulls daily: the gains come from how well the environment is designed, not from a bigger model. Paxis ships that harness as a product. Agents get released, the orchestration holds the work state, and a derailed stage is recovered and picked back up. Metis layers learning and inference cost on top, so you can check which lever actually produced the progress. And one more difference: where the environment runs. On-prem, scaling up means our electricity bill moves, not somebody else’s meter. Half self-deprecation, half ownership.
An auto-generated comic riffing on this week’s industry news.