GitHub turned its developer conference into an agent orchestration keynote. NVIDIA and AMD wrote checks to their own biggest customer. Google put a reasoning brain for robots on a public cloud endpoint, and Apple quietly opened Keynote and Pages to outside agents. Here are the seven stories that mattered in the past 24 hours.
1. GitHub Universe 2026: Copilot Becomes an Agent Control Plane
At GitHub Universe in San Francisco on Wednesday, GitHub laid out its vision as a complete software delivery system built around agents, anchored by a new GitHub Copilot app that connects agents and tools across projects in one surface. Jay Parikh, EVP of CoreAI at Microsoft, wrote in the company blog that the measure of the system is no longer code written but completed work shipped: a feature, a bug fix, a security fix. The app ships with Copilot projects for parallel agent sessions across multiple repositories, session continuity that moves one cloud agent session between the GitHub app, VS Code, the CLI and the web, and Copilot automations that learn from their own run results and propose improvements for human approval.
— GitHub
The technical centerpiece is Copilot Fusion, now generally available after its preview under the HydraFusion name. Fusion evaluates what a task needs across reasoning, code generation, debugging and tool use, then routes the work across one or more models. GitHub says Fusion Max scored within two points of Claude Opus 5.5 Max on the Artificial Analysis Coding Agent Index at half the cost and 40 percent less time per task. Copilot code review Max posted the highest F1 score on ReviewBench for high-severity issues, and GitHub Enterprise customers can now use all Copilot features with no separate subscription, paying only for the AI credits they consume.
— GitHub · The Register
The demand numbers behind the keynote explain the infrastructure spending. GitHub added 4 million CPU cores, 160 petabytes of high-speed storage and 50 terabits per second of network capacity this year, and is migrating to a multi-region Azure deployment expected to complete early next year. The Register reports 7.4 billion code commits hit GitHub in September alone, five times the level a year earlier, after agent-driven traffic caused repeated outages. The same-day Octoverse report counts more than 40 percent of pull requests starting from an agent draft in September, up from 17 percent in March, across 225 million developers.
— GitHub · The Register
2. NVIDIA and AMD Put $4 Billion Into Meta's Record Equity Raise
NVIDIA announced on Tuesday it will invest $3 billion in Meta's equity offering, with AMD committing $1 billion of its own, turning Monday's non-binding interest from Meta's $95 billion raise into signed checks. Meta simultaneously upsized the offering to $97.5 billion: $25 billion in common stock priced at $625 per share, $12.5 billion in convertible preferred depositary shares at $50, and a separate program to sell up to $60 billion at market prices beginning after its nonvoting Class C shares start trading in the first quarter of 2027.
— NVIDIA · AMD · Reuters
The investment is wrapped in expanded commercial commitments. NVIDIA said Meta will widen its deployment of Vera Rubin GPUs and Vera CPUs, and the two companies will deepen work on Muse, Meta's consumer AI agent launched last month, which runs on NVIDIA infrastructure. AMD said its Helios racks are installed and testing at Meta, that EPYC processors power Muse, and that Meta has deployed millions of EPYC chips across generations as a first customer for sixth-generation EPYC.
— NVIDIA · AMD
The circular flow of capital is the part analysts keep circling. Chip suppliers are now financing the expansion of the customers who buy their silicon, on top of NVIDIA stakes in OpenAI, Anthropic, Intel and xAI this year and AMD's pledged investment of up to $5 billion in Anthropic in July. Meta, for its part, burned $2.12 billion in free cash flow last quarter, its first negative quarter in over a decade, while committing roughly $825 billion in future spending. William Blair told clients the raise reinforces both Meta's confidence in AI returns and the increasingly capital-intensive nature of the race.
— NVIDIA · Reuters
3. Google Puts a Robot's High-Level Brain on Google Cloud
Google Cloud opened Gemini Robotics ER 2, the embodied reasoning model, in public preview through Model Garden on the Gemini platform on Monday. ER 2 acts as a high-level brain for robots: it plans multi-step tasks, watches continuous video to judge whether each step actually succeeded, and hands physical execution to vision-language-action models or navigation APIs that developers declare as tools. The model is built on Gemini 3.5 Flash and priced at $1 per million input tokens during the preview period, moving to $2 from January 2027.
— Google Cloud
The claims center on verification, which is where most robot deployments stall. Google says ER 2 locates the exact frame where a key event happens with 91.3 percent accuracy and a 0.96 second mean absolute frame distance, reads more than ten categories of instruments from dials to liquid thermometers, and lets multiple robots share one spatial understanding to split and hand off work. Boston Dynamics is applying it to Spot's video inspection feature, Singapore's ASTAR IMRE is using it to give lab robots the flexibility to operate legacy equipment with no API, and Arm is exploring the model for future robotics compute platforms.
— Google Cloud
Enterprise plumbing is the real news. The model ran through Gemini API and AI Studio since July, but enterprise customers can now call it inside their own Google Cloud projects under existing security, access and billing systems, with simulator integration for MuJoCo-Warp and NVIDIA Isaac Sim, video labeling in Cloud Storage and BigQuery, and time-critical tasks like mapping running locally on Google Distributed Cloud. The move positions Google to sell the full robot development loop, from simulation to data labeling to VLA training, as a managed cloud workflow.
— Google Cloud
4. Apple Adds AI Assistants and MCP to Keynote, Pages and Numbers
Apple announced on Monday a major Apple Creator Studio update that puts beta intelligent assistants inside Keynote, Pages, Numbers and Logic Pro, built on Apple models and OpenAI models. In Keynote the assistant helps design, animate and refine slides; in Pages it drafts and restructures writing; in Numbers it calculates, visualizes and organizes data, recommending chart types and building custom tables. The Logic Pro assistant can start a song, set up a vocal track from a single prompt, or refine an active project.
— Apple Newsroom · The Verge
The quieter but bigger move is interoperability. Keynote, Pages and Numbers now support Model Context Protocol, so external AI agents and tools can read content, perform edits and run automated actions directly inside the three apps, a door Apple has kept closed until now. Keynote also adds an open JSON-based interchange format, .keyx, designed to be readable by outside tools and agent workflows. MCP support requires macOS 27 or later, while the built-in assistants run on Apple Intelligence-capable devices on iOS 27, iPadOS 27 or macOS 27, in US English only.
— Apple Newsroom
The update lands alongside Apple's M6 MacBook Pro and iMac announcement pitched at on-device AI, and arrives in November as a free update for existing Creator Studio subscribers at $12.99 per month. Apple adopting an open agent standard rather than building a private bridge is the detail to watch: the company that resisted outside app ecosystems for decades is now letting third-party agents drive its productivity suite.
— Apple Newsroom · The Verge
5. Suhail Doshi Launches Particle, a Model Company Attacking Token Waste
Suhail Doshi, the co-founder of Mixpanel and founder of Playground, launched Particle on Tuesday with a first model, Particle-1, aimed at developers whose agent bills are growing faster than their output. The company postures as an inference infrastructure play: Particle-1 is a post-trained version of DeepSeek V4.1 Flash, served through an API, and the pitch is that it completes coding and data-engineering tasks with fewer output tokens than the model it is built on, measured across four software and data-engineering benchmarks.
— Particle
The thesis lands on a real pain point. Agent workflows burn tokens on retries, verbose reasoning and repeated context, and inference cost has become a line item teams now optimize as deliberately as compute. A post-training pass that trims output tokens without losing task completion is an efficiency wedge into a market where the underlying model weights are open and cheap, and the differentiation is entirely in serving economics.
— Particle
Doshi's move follows the broader pattern of 2026: founders with product analytics backgrounds building companies that sell measured outcomes rather than model access. Mixpanel made product usage measurable; Particle wants to make agent token usage measurable and then shrink it. Whether independent benchmarks confirm the claims will decide if this becomes a real category or another thin wrapper on open weights.
— Particle
🔗 Particle
6. Rime Launches Prism, a Speech-to-Speech Model Built for Enterprise Phone Calls
Rime, a San Francisco voice AI company handling nearly 100 million phone calls a month for customers including the Mayo Clinic and Dialpad, launched Prism on Tuesday, a fine-tunable speech-to-speech model for enterprise voice agents. Prism takes audio in and produces audio out without the separate transcription, reasoning and speech chains most voice stacks still run, and it generates a verifiable text checkpoint of what it intends to say before it speaks, giving companies an auditable record of every response.
— Rime
The numbers are the sales pitch. Prism responds in roughly 500 to 570 milliseconds to first audio input, scored 56.6 on the τ-Voice customer service benchmark ahead of real-time models from OpenAI and Google, and in a trial with Revmo completed order placement 94 percent of the time after caller consent, against 57 percent for an OpenAI model in the same comparison. Companies can fine-tune it on their own data and deploy it in a private cloud or fully on-premises, which addresses the compliance walls that keep regulated industries off public voice APIs.
— Rime
The launch follows a raise the company puts at $35 million, and it defines a clearer enterprise wedge than general-purpose voice assistants: phone calls where a wrong answer costs money, in healthcare, finance, airlines and food service. Speech-to-speech with a written checkpoint is an architecture designed for auditors first and users second, which is exactly where enterprise voice procurement decisions are actually made.
— Rime
🔗 Rime
7. Disruptive Closes a $10 Billion Fund to Write $1 Billion Checks
Disruptive, the Dallas investment firm founded by Alex Davis, announced on Monday a $10 billion flagship fund to back roughly ten growth and late-stage private technology companies with checks of up to $1 billion each, pointing to AI, compute infrastructure and defense as the sectors driving the capital need. The firm's portfolio includes Airbnb, Databricks, ElevenLabs, Groq, Palantir, Reflection AI, Spotify and Shield AI, and Davis told a Venture Dallas audience he expects to deploy the full $10 billion into about ten companies over the next year and a half.
— Disruptive · TechCrunch
Structurally, this is the firm's first blind-pool fund, meaning limited partners commit without knowing the targets, a departure from the deal-by-deal special purpose vehicles Disruptive used until now. The Wall Street Journal reported earlier this month that the firm had secured at least $7.5 billion in commitments while fundraising, and Bloomberg reported Disruptive told LPs it has returned more than $6 billion on roughly $1.9 billion invested in startups. The Dallas Business Journal calls it the largest venture fund ever raised in Texas.
— Disruptive · TechCrunch
The fund is a bet on the barbell forming in private markets. Mega-funds of $500 million or more account for only 6 percent of funds closed this year but 78 percent of capital raised, and Gartner warned this month that over $3.5 trillion of data center spending could face impairment by 2030. Concentrated $1 billion checks into ten category leaders is the venture answer to that concentration: fewer, larger, later bets, with AI infrastructure as the centerpiece.
— Disruptive · TechCrunch
🔗 Disruptive · TechCrunch
KD Agentic · AI Daily Digest
Top comments (0)