독서

Microsoft is turning Copilot from an AI add-on inside Office into the entry point that carries Office itself.

Last week, at a closed-door event for large enterprise customers, Satya Nadella positioned Copilot as the "OS for work." Coding, agent capabilities and the full power of Office will go directly into Copilot, where users can create and edit documents in an interface that resembles the web version of Office. The Verge reported the event's contents on October 1.

The product moves have already happened: the desktop Scout agent announced at Build in June has been put into maintenance mode, with the cloud version rebranded to Autopilot. The Verge, citing sources, reports the rebrand was driven in part by trademark concerns around Yahoo Scout, and that the Copilot reorganization involved internal tensions; Microsoft has not confirmed either point.

Copilot chief Jacob Andreou admitted that agents have been "almost impossible to bring into the enterprise" because of the autonomy they need; Microsoft says the new architecture keeps organizations in control. On the consumer side, roughly 90 million people pay for Microsoft 365 out of their own pocket, which is why Microsoft is not giving up on consumers.

Google released Gemini 4 Argon on September 30, with introductory pricing matching OpenAI's latest discount — but the model is not broadly available yet.

According to CNBC, Argon ties OpenAI on a key cybersecurity benchmark and posts leading software engineering results; these are industry-benchmark citations, not independent retests. Introductory pricing is $2 per million input tokens and $10 per million output tokens, matching OpenAI's newly discounted GPT-6.1 Sol.

Google said its published rates will eventually double to $4 and $20 per million tokens. Access is initially restricted to select cybersecurity customers while the model goes through the voluntary U.S. government safety-testing process.

Market attention has shifted to personal agents: Meta's Muse is free and passed 5 million downloads as of September 30, per Sensor Tower, while Google's Spark remains limited to paying subscribers. JPMorgan analysts wrote Thursday that Google needs a significant advance in personal agents to generate consumer enthusiasm.

Cybersecurity startup Armadin announced on October 1 that it has raised $255.5 million in Series B funding at a valuation exceeding $2.5 billion, co-led by Andreessen Horowitz and Accel.

Founder Kevin Mandia previously built breach-investigation firm Mandiant, which Alphabet acquired for $5.4 billion in 2022. The new company uses swarms of AI agents to simulate attacks, find weak points in customer infrastructure, and generate remediation advice.

The company says one customer project ran 26,000 agents that scanned more than 25,000 assets and found 38 verified attack paths; those figures are the company's own. The new funds will go toward research, AI training and go-to-market programs.

A US federal judge has dismissed publishers' antitrust suits over Google's AI search features, holding that free content in exchange for traffic is not an agreement but simply how a search engine works.

Per The Verge's October 1 report, US District Judge Amit Mehta, who ruled in 2024 that Google illegally monopolized search, dismissed the two suits on Wednesday (September 30), filed last year by online education company Chegg and Rolling Stone parent Penske Media; Reuters reported the ruling earlier.

The suits accused Google of coercing publishers into supplying content for AI Overviews, the AI summaries at the top of search results, for free or risk disappearing from search. Per the ruling as quoted by The Verge: "an expectation is not an agreement. It is simply how a general search engine works." Mehta wrote he was not unsympathetic to publishers, but antitrust rules cannot replace a legislative decision on the economic impact of new innovation.

Publishers' traffic declines continue. This week The Information reported that Google is paying around 100 publishers for their contributions to AI Overviews, AI Mode and Gemini as part of a pilot program. The Verge itself is part of plaintiff Penske Media; the report does not say whether the plaintiffs will appeal.

Anthropic announced on September 30 that Claude for Government is now generally available to US federal and state agencies, ending the public beta that began in July.

The platform runs in a FedRAMP High environment, the strictest US cloud security authorization level. Agencies get capabilities comparable to commercial customers, including desktop file work, skills and plugins, and Claude Code for building and modernizing public-sector software.

There are no seat fees. Agencies pay for usage in fixed increments with a hard not-to-exceed cap; administrators allocate budgets by department, set model access per group, sensitive operations require two-person approval, and conversation history stays on agency-managed devices.

The boundary is equally clear: per The Decoder, the Pentagon still designated Anthropic a supply chain risk in March, so the military market remains out of scope. How many agencies will sign and at what spend, the announcement does not say.

Alkaid Computing (九章云极), a Beijing-based AI computing infrastructure provider, filed its listing application with the Hong Kong Stock Exchange on September 29. The offering size and amount to be raised have not been disclosed.

According to Zhidx reporting and the prospectus, revenue reached 1.098 billion yuan in 2025, up 177.7% year on year, and 588 million yuan in the first half of 2026, up 92.0%. Adjusted net loss narrowed from 206 million yuan in 2024 to 91.3 million yuan in 2025, and 14.88 million yuan in the first half of 2026.

The risks are equally concrete: the top five customers accounted for 88.0% of 2025 revenue, the top five suppliers for 84.8% of purchases, operating cash flow remains negative, and net liabilities stood at 2.644 billion yuan as of June 30, 2026. The company sources computing capacity mainly through long-term arrangements rather than building its own facilities, to limit upfront capital spending.

The company has completed 14 private funding rounds, with Beijing state-owned investors including Beijing State-owned Capital Management and Zhongguancun Capital. Whether the listing proceeds, and at what valuation, depends on the exchange review and market appetite for AI computing assets.

Ai2 released the open-source training framework Olmo-core 3 on October 1, with self-tested throughput 2.7× its previous implementation.

In the official benchmark, a 47-billion-parameter mixture-of-experts model processed 52,000 tokens per second per GPU on eight NVIDIA B300s, versus 19,400 on the earlier stack. The trillion-parameter tests used random routing and measure system performance, not training quality.

Code and a technical report are public. Academic and smaller labs gain another option for training very large MoEs, and the numbers come from Ai2's own tests.

Ai2 open-sourced AstaBrief 8B on October 2, a model that turns a research question and retrieved literature excerpts into a cited report, releasing the weights and training data together.

Built on Qwen3-8B and trained with supervised fine-tuning plus direct preference optimization, it writes the full report in one pass rather than section by section. Ai2 says Fast mode averages 51.1 seconds per report versus 178.5 seconds for its Claude-powered Thinking mode, about 3.5× faster.

Open weights let institutions run it on their own infrastructure, which matters for sensitive or unpublished research. Note that most training and evaluation was completed in 2025 against the proprietary models of that time; Ai2 states it has not rerun the evaluation against today's frontier models.

On October 2, Elon Musk announced on X that Tesla is cutting the AI5 chip's memory from 144GB to 72GB and the AI6 from 216GB to 144GB, reserving the saved memory for Optimus robot production. He said memory bandwidth stays unchanged, so the impact on Optimus performance should be "negligible." The move follows Micron's CFO predicting on an earnings call that humanoid robots will need over 200GB of DRAM per unit, with memory prices unlikely to fall even if the AI bubble bursts. The spec change is so far only Musk's personal announcement; the chips have not shipped, and the real impact will show in production.

Microsoft CEO Satya Nadella announced in a Thursday memo, dated October 1, that Ryan Roslansky, who led LinkedIn for six years, will leave the company at the end of 2026.

Roslansky became LinkedIn CEO in 2020 and grew membership from 700 million to 1.3 billion; in 2025 he took on additional responsibility in the Office group, and earlier this year was named alongside two other executives to lead Microsoft 365 applications and the Copilot AI platform. LinkedIn's COO Dan Shapero already took over as its CEO.

Roslansky wrote that leading the next phase would require being in Redmond full time, while his family lives in the Bay Area. He follows gaming chief Phil Spencer and Office leader Rajesh Jha, who both left this year.

Cloud platform Modal introduced Sidecars on October 1: isolated containers that run on the same host as the main Sandbox, built to hold trusted code such as credentials, proxies and tool logic.

Splitting an agent's "brain" from its "hands" previously meant every tool call crossed a network round trip, trading latency for security. Sidecars connect over an internal bridge network so the two talk locally; Modal says communication across the trust boundary is 3x faster than using separate Sandboxes, a figure from its own internal benchmarking.

Isolation matches separate Sandboxes, using gVisor or VMs depending on the runtime. The feature is exposed through an experimental API, and real-world gains remain to be seen.

Shopify released Canvas on October 1, letting merchants build an entire store by chatting with its AI assistant Sidekick, with changes visible in real time.

Shopify's previous no-code editor worked by rearranging modules and blocks within a chosen theme, and deeper customization still required editing code or hiring a developer. Canvas renders the store's actual theme code rather than a static preview, so merchants can test full interactivity and animation. Shopify simplified the theme architecture so Sidekick can read and write theme files directly.

Wix, Squarespace, Webflow and others already offer similar AI site builders; Canvas brings that approach to a leading e-commerce platform. The product is early days by Shopify's own account, and real-world results remain to be seen from merchant use.