Every paper cited in the newsletter, the one line that mattered, and my verdict where I gave one. The abstract is free. The opinion took a fortnight.
Hyungjin Lukas Kim
An estimated 82% of migrations off deprecated models happened after the shutdown date, once the app was already failing. 94% of the apps had the model ID hard-coded.
AI apps age like milk, not wine, and most teams don't even keep the expiry date on the fridge.
Gruenbaum, Porat, Natanzon, Zavida, Dinachi, Itzahary
When a question states the rules, the top four models score 22 to 25 out of 27. When the answer hinges on a hidden fact buried in contradictory records, four of six models get 6 of 24 or fewer.
Jeremy Canale
Strict gate success: 94.98% for Gemini 3.8 Flash, 83.29% for GPT-5.6 Luna, 74.18% for DeepSeek v4.1 Flash. A deterministic baseline passed all 1,700 gates.
"Automate 95%" sounds like savings until someone has to hunt for the 5%, forever.
Villamizar, Fischbach, Şahin, Frattini, Ferrari, Korn, Vogelsang, Mendez
More smells went with lower functional correctness, measured by test suites. Cleaning them up helped but didn't guarantee correct code.
Jiapeng Li
When the agent could read back whether a write had happened, instructed frontier models duplicated a lost-ack write only 0.5% of the time. Without a read-back: 56 to 74%. Idempotency keys on every write cut duplication from 28% to 4%.
Zhang, Xia, Wu, Yue, Zhang, Cheng, Tu
Of 111 approval/trace pairs, in 40 the fields a human sees left effects uncovered. The authors prove no record-only policy can be both safe and permissive.
Gloaguen, Mündler, Müller, Raychev, Vechev (ETH Zurich)
Providing context files "does not generally improve task success rates, while increasing inference cost by over 20% on average." Repository overviews, the thing model providers keep recommending, were "not helpful."
Last updated . Reuse with credit under CC BY 4.0: “The Dabbawala Protocol, thedabbawalaprotocol.com”.