AI is moving into engineering design and scientific data analysis.
Three recent releases make that shift concrete.
OpenAI (Nov 3): Astra helped design the enclosure, circuit board and software for a device that talks to Codex. A manufacturer built a small batch of prototypes. OpenAI also reports that production limits, tolerances and assembly details required several rounds of human revision.
https://openai.com/index/accelerating-engineering-for-the-physical-world/
Google (Nov 3): Physical-Engineering’s Last Exam tests agents on 85 tasks across mechanical, electrical, civil and structural engineering. In Google’s baseline, the top configuration passed every hard safety and feasibility check on 41.3% of deliverables in its co-authored task set.
https://cloud.google.com/blog/products/ai-machine-learning/introducing-physical-engineerings-last-exam-for-agents
10x Genomics (Nov 2): Sentira Single Cell turns questions about Chromium data into an analysis plan and executable Jupyter notebook. Researchers approve the plan and can inspect, edit and rerun the analysis.
https://www.prnewswire.com/news-releases/10x-genomics-launches-sentira-single-cell-a-new-agentic-ai-experience-for-chromium-analysis-302925391.html
Together, these cases show AI taking on more of the work between a question and a result. The test is what happens at the next handoff: Can the design be manufactured? Does it pass safety checks? Can a scientist examine and defend the analysis? That is how we should judge progress in AI for Science.
WendyDing
npub1xt…06rus
2026-11-04 02:18:39 UTC
Author Public Key
npub1xtwy7fvu8f7wdtgnpm68wyrf6uxshf49tn5kp7kyusu6872amn8qh06rusPublished at
2026-11-04 02:18:39 UTCEvent JSON
{
"id": "fa0b5b8545ad215ae24a1b0522e39e18ec53b6207065584dc098a5ceb8411530",
"pubkey": "32dc4f259c3a7ce6ad130ef4771069d70d0ba6a55ce960fac4e439a3f95ddcce",
"created_at": 1793758719,
"kind": 1,
"tags": [
[
"client",
"YakiHonne",
"31990:20986fb83e775d96d188ca5c9df10ce6d613e0eb7e5768a0f0b12b37cdac21b3:1700732875747"
]
],
"content": "AI is moving into engineering design and scientific data analysis.\n\nThree recent releases make that shift concrete.\n\nOpenAI (Nov 3): Astra helped design the enclosure, circuit board and software for a device that talks to Codex. A manufacturer built a small batch of prototypes. OpenAI also reports that production limits, tolerances and assembly details required several rounds of human revision.\nhttps://openai.com/index/accelerating-engineering-for-the-physical-world/\n\nGoogle (Nov 3): Physical-Engineering’s Last Exam tests agents on 85 tasks across mechanical, electrical, civil and structural engineering. In Google’s baseline, the top configuration passed every hard safety and feasibility check on 41.3% of deliverables in its co-authored task set.\nhttps://cloud.google.com/blog/products/ai-machine-learning/introducing-physical-engineerings-last-exam-for-agents\n\n10x Genomics (Nov 2): Sentira Single Cell turns questions about Chromium data into an analysis plan and executable Jupyter notebook. Researchers approve the plan and can inspect, edit and rerun the analysis.\nhttps://www.prnewswire.com/news-releases/10x-genomics-launches-sentira-single-cell-a-new-agentic-ai-experience-for-chromium-analysis-302925391.html\n\nTogether, these cases show AI taking on more of the work between a question and a result. The test is what happens at the next handoff: Can the design be manufactured? Does it pass safety checks? Can a scientist examine and defend the analysis? That is how we should judge progress in AI for Science.",
"sig": "3904ebae9d7ca7808b606d91a62dafb52da1817bf6a7b41cffb523027fff94182936eca9d46b60d2d21e7b73519467392ab445c3245730b2ca5aa80396a21cda"
}