TabLOID

Perplexity's November 6 changelog introduces Teammates for complex, continuing work. The company says they can plan tasks, run work in parallel, retain context and bring results back for review. They draw on its Computer tasks, Skills, memory and tools, and can collaborate with other Teammates.

The launch is available to Max subscribers on the web. Perplexity also describes connections to workplace tools, including Slack. These are the company's product descriptions, not an independent finding that the agents reliably finish every assignment.

The appealing shift is from asking a fresh question to maintaining a standing brief. The practical test is whether the resulting work is useful enough to justify the supervision. A recurring assignment still needs a clear scope, appropriate access and someone willing to inspect what comes back. Perplexity's changelog, November 6.

webAI reports that its Meridian Air CivoE system lifts a 12-billion-parameter foundation model's composite evaluation score from 59.8 to 85.5. Its approach combines specialized experts rather than simply making the underlying model larger.

Those numbers are normalized scores across 14 curated benchmark datasets. They are not percentages of all questions answered correctly. The company selected task families where the foundation was already competitive, then evaluated its specialist system alongside other models.

The comparison has limits: it is company-run, and exact competitor revisions and reasoning settings were not fully controlled. The results support examining this particular method on these tasks. They do not establish that a small model has become a universal winner. For buyers interested in private deployment, the useful follow-up is a comparison on their own workload. webAI's methods and results, checked November 7.

The Subvocal Company's founders describe glasses that use radio-frequency sensing to detect small facial muscle movements as a wearer silently articulates words. Software turns those signals into text. The intended interaction is silent speech, not reading an unspoken thought.

In their Y Combinator launch, the founders claim speeds above 200 words per minute. That is their reported performance, not an independently established speed for ordinary users in varied conditions.

The hardware is also still in development. The founders describe a semi-portable board connected to glasses, while the polished product imagery represents the intended finished device. Miniaturization and broader training remain work ahead. If the approach becomes dependable, it could offer another way to enter text without speaking aloud. For now, the prototype and its promises belong in separate columns. The founders' launch description, checked November 7.

Firefly announced a deal worth more than $1.3 billion with AST SpaceMobile on November 6. It covers 36 rocket engines, a manufacturing license and three years of engineering support for AST's planned heavy-lift rocket.

The package includes 27 Miranda engines and nine vacuum-variant Vira engines for the first three vehicles. Firefly retains ownership of its engine intellectual property. Payment combines cash and AST shares.

Buying a developed engine design can remove a formidable piece of work from a new launch program. It does not turn the rest of a rocket into a finished product. Vehicle integration, production and flight testing remain consequential steps. The agreement is a major commitment to building launch capacity; it is not evidence that AST's new rocket has flown. Firefly's announcement, November 6.