Skip to content
AI Designs Hardware in a Week, Claude Plugins Auto-Connect, Siri Opens to All Apps

AI Designs Hardware in a Week, Claude Plugins Auto-Connect, Siri Opens to All Apps

This daily briefing covers five key AI developments: OpenAI using models to design a complete device in under a week, Claude plugins enabling automatic enterprise login via identity systems, Apple opening Siri to all app developers, DeepMind's AI agent compressing drug design cycles to one hour, and a laptop-based model reproducing neural weight interpretation research.

ThinkInAI Community
ThinkInAI Community
ThinkInAI Community

01 | AI Starts Drawing Engineering Diagrams and Designing Circuit Boards

For the first time, a large model is being pushed as a formal hardware engineering assistant, though it still lacks factory experience for direct production.

Two OpenAI employees used GPT-6 Astra to design a small device in under a week. The model completed the enclosure, circuit board, interface, and firmware. A contract manufacturer produced a prototype within a month. OpenAI's hardware team model usage has increased 23-fold this year, and they are collaborating with Mazda and racing teams on simulation and "scan-to-drawing" conversion. A key limitation is stated plainly: the model does not yet stably consider tolerances, production capacity, and assembly details — precisely the gap most worth filling for manufacturing.

02 | Enterprise Plugins: Login and Connect Instantly

The most annoying step in rolling out AI assistants in enterprises — individual authorization — is being taken over by identity systems.

An administrator configures the integration once in the company identity system. When an employee uses the plugin for the first time, it connects automatically. The first batch supports Notion, Atlassian, Figma, and over a dozen other services. This is built on an open extension of the MCP (Model Context Protocol), so other agent platforms can adopt the same approach. The benefit is a dramatic reduction in deployment cost; the trade-off is that permissions follow identity, requiring finer-grained role-based permission configuration.

03 | Siri Becomes the Agent Entry Point for All Apps

Whoever turns their core operations into actions that Siri can invoke will capture the on-device agent entry point first.

Any app can now integrate with the new Siri. Users speak requests in natural language, and Siri can query package tracking, make reservations, place orders, etc. Developers can also write "skills" that chain multi-step operations. This is currently a developer preview; a user-facing release is expected in a few months, with App Store submission opening next spring. EU iPhones are temporarily excluded, and some cloud features have daily rate limits.

04 | Agent Completes a Drug Design Round in One Hour

Letting the model handle orchestration and reasoning while delegating precise computation to specialized tools is the most reliable division of labor for professional-domain agents.

DeepMind equipped Antigravity with chemistry skills. The agent searches literature, cleans thousands of molecules, trains an activity prediction model, and then proposes new molecules and synthesis routes. A round of work that originally took days is compressed to roughly one hour. The official announcement cautions that these designs are only hypotheses awaiting verification; they must be synthesized and tested experimentally. The same approach can be borrowed by chemical, materials, and process industries.

05 | Follow-up: Laptop-Scale Models Can Also "Read Weights"

"AI can understand neural network weights" turns out to be easier to reproduce than imagined, but it does not mean the model can rewrite itself.

A developer reproduced the previous day's research using Qwen 3.8 27B on a single laptop. Given only weight numbers, the model recognized 3 out of 4 simple games, with each run taking nearly an hour. Meanwhile, scholars criticize the original study for conflating "reading weights" with "writing code to modify the model." The research remains interesting, but its conclusions require more independent verification.

Other Items Worth Noting

Official updates: In Claude Code, typing ultrathink only increases thinking for the current round; idle sessions are automatically compressed to save quota. OpenAI announced a security partnership with Cisco, a real-estate company case study, a new Chief Corporate Affairs Officer, and claimed that ChatGPT search volume has reached one-fifth of Google's.

One-sentence summary: AI is moving from chat boxes into engineering drawings, enterprise systems, and phone operating systems; the real barriers are now domain experience and permission management.

Original Source

Signed-in readers can open the original source through BestHub's protected redirect.

Sign in to view source
Republication Notice

This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactadmin@besthub.devand we will review it promptly.

AI agentsOpenAISiriMCP protocolModel interpretabilityClaudeEnterprise AIDrug discoveryDeepMindAI hardware design
Reader feedback

How this landed with the community

Sign in to like

Rate this article

Was this worth your time?

Sign in to rate
Discussion

0 Comments

Thoughtful readers leave field notes, pushback, and hard-won operational detail here.