Trendora Radar — Volume 11
This week’s radar is led by Frontier AI moving from trial to adopt, alongside fresh adoption of C and CUDA, while model distillation and chain of thought both moved into trial. The broader story is one of momentum and caution: Gemini 3.5, TypeScript, and Go moved back to trial, Apache Spark moved to assess, and headlines ranged from OpenAI’s reported Navier–Stokes breakthrough and global-scale ChatGPT infrastructure work to critical security issues and new geopolitical scrutiny of frontier models.
What moved on the radar
Top stories
- OpenAI Claims a Navier–Stokes Millennium Prize Problem Breakthrough
OpenAI published an article describing progress on the Navier–Stokes Millennium Prize Problem, a major open problem in mathematics and fluid dynamics. The post sparked substantial discussion because the result, if validated, would represent a significant scientific breakthrough.
Hacker News - OpenAI scales Habitat to support ChatGPT at global scale
OpenAI describes how it evolved Habitat from a Python library into a globally distributed storage platform. The system is now used to support ChatGPT at massive scale, handling 1 billion users and 22 million requests per second.
OpenAI Blog - Google to buy half of a Finnish nuclear plant’s output
Google announced a record €13 billion investment in Finland to expand AI infrastructure, including three new data centres and upgrades to an existing site. As part of the plan, Google signed a 22-year deal with Fortum to buy up to 50% of the output from the Loviisa nuclear power plant to help power its growing data centre needs.
Hacker News - Forgejo Versions Up to 16.0.3 Exposed to Critical RCE
A critical remote code execution vulnerability has been reported in Forgejo versions up to 16.0.3. The posted text is heavily corrupted, but the headline indicates a high-severity security issue affecting the self-hosted Git platform and its users.
Hacker News - Calif Research Demonstrates Zero-Click WeChat Worm
Calif Research says it has released a demo of WeWorm, a zero-click worm that can spread through WeChat calls on iOS and Android without the victim answering or interacting with the device. The team says AI helped them find the bug, build a remote code execution exploit, and develop the worm in a short timeframe.
Simon Willison - U.S. Accuses Six Chinese AI Firms of Copying Frontier Models
U.S. agencies say DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI have been conducting industrial-scale distillation attacks against frontier AI models from U.S. companies since at least late 2024. The agencies allege the firms likely operated with Chinese government awareness, citing use of capabilities extracted from models such as Claude, GPT, Gemini, and Grok.
Ars Technica AI - Apple unveils iPhone Duo, its first foldable iPhone
Apple announced iPhone Duo, its first foldable iPhone, featuring a 7.6-inch inner display, a 5.4-inch outer display, and a redesigned iOS 27 experience. The company says it uses an A20 Pro chip, a custom vapor chamber, dual batteries, Touch ID in the side button, and Apple Intelligence integration, with pre-orders starting October 16 and availability on October 23.
Hacker News - Microsoft’s September Patch Tuesday hits a record
Microsoft’s September security update fixes about 972 vulnerabilities, including 112 rated critical, setting a new monthly record. The surge follows similarly large patch releases from recent months and comes amid warnings that AI-assisted attacks may compress the window for defending against active exploitation.
Ars Technica AI
Implications at publication
The downgrade shows the market is not ready for broad deployment, but signals such as a hardware standard for agents, enterprise movement from assistance to execution, and safer cost optimization suggest the ecosystem is taking shape.
For business development, agentic AI is opening new conversations around integration, standards, and enterprise deployment, but it is not mature enough to sell as a general-purpose automation capability. BD will need to position around narrow problems with provable value and be ready to discuss safety, cost, and operability rather than pitching “agents” as a generic feature.
Agentic AI has been demoted from Adopt to Trial, and the articles show it being used in narrow flows like coding support, retrieval, and data preparation for agents, but without proving broader output gains.
For software engineers, agentic AI is moving from “worth trying” to “use only with tight controls” in narrow workflows such as retrieval, coding assistance, or task automation. The news suggests coding may get faster without a matching rise in engineering output, so the focus shifts to setting boundaries, validating outputs, and improving data/context integration rather than handing broad autonomy to agents.
The radar moved from Adopt to Trial, reflecting that enterprises are testing agentic AI in specific workflows rather than rolling it out broadly. The sources emphasize a shift from assistance to execution, but also show that data readiness, fluency, and control are needed before scaling.
For product managers, agentic AI is no longer just something to “try”; it is a question of choosing the right workflow to move from assistance to execution, but only where impact can be measured and risk controlled. PMs will need to prioritize narrow use cases, define success criteria clearly, and avoid assuming autonomous agents will automatically improve product outcomes or team productivity.
Even with the downgrade, the articles show agentic AI moving into automated testing, security investigations, and data preparation—areas that are more measurable and bounded than broad autonomous work.
For QA/testing, agentic AI is becoming more practically useful because it fits repetitive, well-defined steps such as generating, running, and scaling automated tests. But moving from demo to production still requires tight control over data, evaluation criteria, and risk, so testers will spend more effort designing guardrails and verifying agent outputs.
The move from Trial to Adopt shows market signals are strong enough to support broader commercialization. The accompanying articles show OpenAI rolling it out, publishing a system card, and already having concrete uses in finance, legal work, and other real-world tasks.
BD can now position GPT-6 Astra as a capability ready for deployment rather than just a technical preview. But conversations will need to emphasize usage conditions, supervision level, and fit-for-purpose context, because the current news shows the model being used in environments with high trust requirements.
GPT-6 Astra has been promoted from Trial to Adopt, showing it has moved beyond experimentation into broader use. The supporting news shows OpenAI has rolled it out, posted a system card, and reports point to strong benchmark performance, robot-arm task results, and real-world uses such as document review.
GPT-6 Astra is moving from a trial tool to something teams can put into the main development flow, so developers will use it more for code generation, bug analysis, and feature support. But with a system card already posted and signals around supervisory work, developers still need tight control over outputs, especially in critical logic and production integrations.
Get it in your inbox
A weekly digest of radar movement + top stories. One email a week, unsubscribe anytime.