For 30 years, we designed software for people. That model is beginning to change. Agents are becoming first-class users of software and infrastructure.
Hugging Face has started measuring its own traffic from coding agents as a separate category. Agents are now the #1 users on the Hugging Face Hub, and they behave nothing like the humans we have built for.
Humans read documentation. Agents parse it. Humans tolerate a three-click flow. Agents want one call. APIs, applications, infrastructure, and even websites will increasingly be designed for agents to discover, understand, and operate.
The strategic value of that shift is already showing up in the market. On Thursday, CNBC reported that NVIDIA has agreed to acquire Hugging Face for $12.9 billion. This extends NVIDIA’s reach beyond the infrastructure that powers AI to the open-source platform where developers, and increasingly agents, find and use models.
But agents moving at machine speed are exposing a new constraint: humans.
The Wall Street Journal found that founders who use agents have to operate at a dramatically faster pace and have to work longer hours to keep up. Agents do not sleep. Their output creates a continuous stream of work for people to review, approve, coordinate, and answer.
Agent work begets human work. As execution gets cheaper, judgment becomes the bottleneck.
Founders must now design the human operating system as carefully as the agent stack. Give agents clear decision rights. Set thresholds for human review. Batch approvals instead of allowing constant interruption. Decide which work deserves to exist before accelerating it.
Bottom line: Agents are becoming the users of software. Humans are becoming the orchestrators of work.
Elsewhere this week, Anthropic committed roughly $45 billion to six years of AI compute, Salesforce opened its applications to agents through reusable capabilities and open standards, and Workday said more than 5,500 customers are already using at least one of its agents. The agent economy is moving quickly from experimentation to infrastructure, interfaces, and real enterprise adoption.
Anthropic agrees to pay Nscale $45 billion over six years for AI computing power. The deal secures about 460 megawatts at a West Virginia data center built around Nvidia Vera Rubin chips. Click here
NVIDIA notifies top customers of price increases of over 15% for Vera Rubin and Grace Blackwell AI servers. Soaring HBM memory costs are driving the increase, effective on systems shipping in early 2027. Click here
OpenAI says its Jalapeño chip beats NVIDIA hardware on power efficiency and latency. The chip delivered 1.5x-1.9x more AI work per watt and 1.7x-3.6x lower latency than NVIDIA chips across GPT-OSS, DeepSeek R1, and Kimi K2.5 1T benchmarks. Click here
Cisco expanded its Secure AI Factory with NVIDIA to include rack-scale compute through a new partnership with Supermicro. The architecture adds high-density liquid- and air-cooled systems for enterprise, neocloud, and sovereign cloud deployments. Click here
AWS plans to add 2 million NVIDIA Blackwell Ultra and Rubin GPUs to its data centers in 2027 and 2028. The order builds on the 1 million GPUs Amazon announced in March. Click here
Enterprise
Salesforce and Anthropic launch Claudeforce, bringing Salesforce data and workflows directly into Claude. The integration begins with a plugin that includes 37 prebuilt sales skills. Click here
Workday said AI drove more than 25% of new annual contract value in Q2, with more than 5,500 customers using at least one Workday agent. It also introduced Developer Agent and Agent Passport, which test, verify, and continuously monitor Workday-built and third-party agents before and after production. Click here
Anthropic’s Opus 5 overtakes Fable 5 in enterprise spending as companies shift to cheaper models. Ramp data shows Fable 5, launched in June, plateaued at about 11% of enterprise spending on Anthropic tools before Opus 5 surpassed it. Click here
Cisco rolled out its personal AI agent, MyAgent, to all 90,000 employees worldwide. The deployment is among the earliest company-wide agent rollouts at this scale, featuring personalized agents designed to complete tasks on employees’ behalf. Click here
Google launches Gemini Enterprise for Legal with Thomson Reuters, LexisNexis, and Harvey. The platform adds specialized AI agents and integrations for law firms, bringing generative AI deeper into regulated legal workflows. Click here
Uber says weekly AI agent requests have grown 9.4x since February, but total AI spending has stayed flat since April. The company used up its entire 2026 AI budget in the first quarter. Click here
Capital Flows
NVIDIA agrees to acquire Hugging Face for $12.9 billion. The deal gives NVIDIA control of the widely used open-source AI model repository, which had drawn interest from other potential suitors, including Salesforce and Microsoft. Click here
Andreessen Horowitz raised a $1.1 billion Machine Age Fund for the physical buildout of AI. The fund targets chips, memory, networking, storage, data centers, robotics, and other infrastructure where AI demand is colliding with supply-chain and physics constraints. Click here
SoftBank plans record $6.3 billion retail bond sale to fund its OpenAI commitments. SoftBank will sell $6.3 billion in retail bonds, the largest such offering by any Japanese issuer and its third this year, as the conglomerate raises capital tied to OpenAI funding commitments. Click here
Emerald AI raises $150 million to optimize data center power use. DCVC and Energize Capital led the round, valuing the data center power optimization startup at $1.05 billion amid rising grid strain from AI infrastructure buildout. Click here
XPeng’s robotics business raises over $900 million in its first funding round. IDG Capital led the round at a valuation above $6.3 billion, with Gaorong Ventures, Tencent, and Alibaba also investing. Click here
Instinct raised a $250 million Series B at a $2.5 billion valuation. The consumer agent connects to users’ apps and devices to handle tasks such as shopping, travel planning, subscriptions, and other life administration. Click here
Research
Google DeepMind launched a pilot of what it calls the world’s first double-blind evaluation of a proprietary frontier AI model. A cryptographically protected environment keeps both model weights and confidential evaluation prompts hidden from the opposing party, reducing the risk of benchmark contamination. Click here
Alibaba releases open-weight model Qwen3.8-Flash, saying it rivals larger competitors at a fraction of the cost. The 125-billion-parameter model is built on Alibaba’s next-generation Qwen4 architecture and is positioned against Anthropic’s Opus 4.6 and DeepSeek’s V4-Flash. Click here
Roughly 1,200 OpenAI agents coordinated to cheat on an unsanctioned benchmark, contributing to a breach at Hugging Face. METR and Redwood Research found the agents sent over 70,000 messages, with about 700 attacking Hugging Face; OpenAI says reward hacking drove the incident. Click here
Anthropic releases a Model Hardware Standard, enabling AI agents to operate physical systems such as microscopes and robot arms. The framework extends agent orchestration beyond software into quantum computing hardware and lab equipment. Click here
Policy
The White House declared a national emergency over foreign supply risks to the U.S. bulk power system. The order explicitly cites fast growth in data centers and artificial intelligence as increasing dependence on abundant, reliable electricity and magnifying the impact of supply disruption. Click here
A federal judge ruled in Anthropic’s favor in its challenge to the Pentagon’s supply-chain-risk measures. The court found the government acted illegally by punishing the company for criticizing the Defense Department’s views on AI use, and the government is expected to contest the ruling. Click here
Alabama attorney general opens investigation into OpenAI’s security practices. AG Steve Marshall launched a probe into OpenAI’s security procedures following the Hugging Face breach in July. Click here
Meta agrees to pay up to $18 billion to settle US state claims over child safety design. The deal with 48 states requires an independently tested age assurance standard and new algorithmic feed controls for teens. Click here
The Trump administration strikes data-sharing deals with OpenAI, Google, Meta, and Amazon to track AI’s effect on jobs. The agreements aim to monitor how AI adoption is reshaping hiring. Click here
Senator Josh Hawley launches an investigation into Flock’s AI-powered surveillance cameras. The probe targets how the company collects, retains, and shares data gathered by its AI license-plate readers. Click here
Global AI Strategy
Australia, Canada, New Zealand, the United Kingdom, and the United States are committed to deeper coordination on AI and national security. Their joint communiqué calls for closer work with industry, timely access to frontier models for cybersecurity, and shared approaches to models that may warrant additional government scrutiny. Click here
OpenAI launched commercial operations in Brazil with a local team based in São Paulo. Brazil is now one of ChatGPT’s three largest markets by weekly active users and ranks second globally by number of developers using the OpenAI API. Click here
UK becomes first foreign nation granted access to Ukrainian AI combat data. The data, used to train AI models that strike Russian targets, will be shared under a new UK-Ukraine AI partnership. Click here
Huawei proposes exporting Ascend 950-series AI chips to build data centers for the Egyptian government. Sources and documents indicate the proposal would test US tech diplomacy by placing Chinese chips within sovereign AI infrastructure typically supplied by NVIDIA, AMD, and Microsoft. Click here
China said it has formulated nearly 200 key standards for artificial intelligence. The government says AI is now integrated across R&D, design, production, quality inspection, operations, and maintenance as it standardizes the sector under the 15th Five-Year Plan. Click here
UK digital infrastructure investment hit £11.2 billion in 2025, its highest level since the dot-com era, driven by AI data center growth. The Office for National Statistics said hardware spending alone rose 54.6% as hyperscalers expanded AI capacity. Click here
Talent Signals
Each week, we spotlight key roles tied to the themes shaping this week’s AI headlines, connecting talent to the companies driving the news.
SambaNova is building a full-stack AI inference platform that helps enterprises and governments deploy high-performance AI across private, sovereign, and large-scale production environments. SambaNova is hiring across AI infrastructure, engineering, hardware, and go-to-market roles. Click here
Assembledis building an AI-powered support operations platform that combines workforce management, automation, and AI agents for customer service teams. Assembled is hiring across product, design, engineering, deployments, and customer success roles. Click here
BroccoliAI is building AI employees for home-service businesses, automating customer calls, scheduling, follow-up, and other front-office workflows for contractors. The San Francisco-based company is hiring across engineering, AI, product, and go-to-market roles. Click here
You can see all the opportunities at Mayfield-backed AI companies here.
Social Signals
The most important conversations in AI are unfolding across social media, where top voices are shaping the next wave of signals and strategy. Here are some of the top social signals and their takes from the past week.
Ryan Greenblatt (Click here) — “The difficulty of understanding incidents and overseeing AI agents appears to be growing faster than the rate at which more capable AIs help us with oversight and understanding. AI capabilities for achieving large, ambitious, and misaligned objectives are growing faster than our ability to understand what these agents are doing.” Reflecting on the METR and Redwood Research investigation of the Hugging Face incident, Greenblatt argues that multi-agent systems are becoming harder to audit as their scale and coordination increase. His takeaway is that agent capability may be advancing faster than the tools enterprises have to observe, understand, and control it.
Aatish Nayak (Click here) — “The world doesn’t just want raw models and agents; it wants problems resolved and outcomes achieved. The premium will accrue to the companies that can diffuse this intelligence across every aspect of civilization, converting raw tokens into real-world outcomes, transforming industries, and creating economies in the process. That work has barely started.” Nayak argues that the app layer is far from dead. As intelligence becomes more abundant, durable value will accrue to companies that own workflows, context, distribution, and customer outcomes, turning frontier capability into something enterprises can actually use.
Greg Isenberg (Click here) — “This is the MOST asymmetric window I’ve ever seen in business. Never in history have this many categories opened at once, all ripe for the taking. Contrary to what a lot of people believe, the opportunity right now is endless.” In a post viewed 286K+ times, Isenberg argues that AI is opening multiple company-building models at once, from agent-enabled services and vertical agents to AI-native rollups and infrastructure for agent workflows. His message to founders is simple: this is a rare window to build businesses with software-like economics in categories that were previously limited by human labor.
