This week, Washington gave the technology a new name, agents became the product, and machines quietly took over most of the web’s traffic.
The White House just renamed it “super intelligence.” On September 29, President Trump signed the White House Accord on Super Intelligence alongside the leaders of Google, Anthropic, Meta, OpenAI, xAI, and Nvidia, a voluntary pledge of internal controls, independent outside auditors, and board-level oversight. A separate executive order the same day directed federal agencies to use “super intelligence” instead of “artificial intelligence” in official communications.
Why it matters for how we live and build: the language we already use is now official vocabulary, but the accord carries no penalties and no deadlines, and the FTC opened a safety probe into OpenAI and Anthropic the day after. Promises on camera are not rules.
OpenAI’s DevDay made delegation the product. At DevDay on September 29, OpenAI unveiled Dots, always-on agents that work in the background across thousands of apps, plus ChatGPT Space for teams, Codex running fully in the cloud, and a Decisions API for handing narrow, repetitive decisions to AI. ChatGPT reached 1.2 billion weekly users. Notably, OpenAI also paused training of its most capable models on September 25 and scrapped its newest model, GPT-6.1, over safety concerns.
Why it matters for how we live and build: the interface is shifting from prompts to delegation. Learning to brief, supervise, and audit an agent is becoming a life skill, not a technical one.
Google answered with Gemini 4 Argon. Announced September 30, Argon is aimed at software development, legal and financial workflows, and cyber defense. Output capacity jumps to 1 million tokens, introductory pricing is $2 per million input tokens and $10 for outputs, and independent analysis puts its factual error rate at 15 percent, the lowest among frontier models. It is rolling out first to cyber defenders through a program Google calls Fairwind.
Why it matters for how we live and build: the frontier race is converging on long, agentic tasks done cheaply, and the winning edge is accuracy. A model that gets it right the first time beats a faster one you have to supervise.
AI agents now generate most of the web’s traffic. Cloudflare data shows bots and agents account for about 57 percent of web requests as of May. CEO Matthew Prince expected that crossover by the end of 2027; it happened a year and a half early. Infrastructure stocks are rallying on the traffic: Cloudflare is up 121 percent from its February low, Fastly is up 159 percent this year, and Anthropic committed $11.6 billion over seven years to Akamai on September 24.
Why it matters for how we live and build: the web is being rebuilt for machines first, and the pick-and-shovel trade on the agent era is compute, bandwidth, and uptime.
Robinhood let AI agents trade with no human sign-off. At its HOOD Summit on September 29, Robinhood unveiled Robinhood Agents inside the main app. Customers approve each order by default but can turn approvals off and let the agent act alone, and any bad trade is the customer’s loss either way. More than 150,000 customers are already in agentic accounts, and agents hit Robinhood’s tools nearly 30 million times a day.
Why it matters for how we live and build: money management is moving from advice to action. The question is no longer whether we trust the agent, but what guardrails we set before we hand it the keys.
The agent toolkit is growing up. Cloudflare shipped Clef and Clef-flash, open-source decision models that return structured choices instead of freeform text, plus a reinforcement learning fine-tuning path. Strands, AWS’s agent framework, released Strands Decider 2B, a tiny open model that scores choices in tens of milliseconds. And Restate raised $20 million on September 30 for durable execution, so an agent that crashes at step 37 of 40 resumes instead of starting over.
Why it matters for how we live and build: when routine decisions stop needing a giant model, agents get cheaper, faster, and safer. That is what turns demos into things we can rely on every day.
Humanoids came to Pittsburgh. Shenzhen-based Astribot brought its T1 humanoid robot to North America for the first time at IROS 2026 in Pittsburgh, which ran September 27 through October 1. T1 runs on the company’s Lumo models, built to connect reasoning with physical action, and has demonstrated autonomous tidying, sorting items into a backpack. Meanwhile, Singapore inaugurated a humanoid robotics center on September 30 with a goal of public-safety deployment by 2028.
Why it matters for how we live and build: physical AI is moving from lab demos toward household and civic work. The idea to watch is co-design, intelligence and body built together from the start, rather than software bolted onto a frame.
