The short version: Four rivals handed the agent rules to a neutral home, and the foundation now hosts 10 projects. The money story is reported and big on both sides of the Pacific: OpenAI pricing past a trillion, DeepSeek raising about 15 billion under US restrictions. And Apple's first foldable went on sale today, while 1Password taught agents to log in without ever seeing your password. |
|
▮ TOP STORY
Rivals handed the agent rules to a neutral home
Anthropic, OpenAI, Hugging Face, and Alibaba each donated a layer of the agent stack to a neutral foundation. The rules now belong to everybody.
Why it matters: Read this next to yesterday's brief. Nadella argued the model is becoming swappable and the harness is the product. Walmart wrote its own agent terms because no shared terms existed. This is the other answer to the same problem: shared, vendor neutral rules for the layers everyone needs, so a skill you write for one agent can travel to another, and switching tools does not mean starting over.
Read the full brief
|
Also today |
OCT 23 · THE MONEY Two reported mega rounds: OpenAI at 1.25 trillion, DeepSeek at 15 billion Two reported mega rounds in one day: OpenAI pricing past a trillion, DeepSeek raising billions under US restrictions. Read → |
OCT 23 · HARDWARE Apple's first foldable, the iPhone Duo, went on sale today Apple's first foldable went on sale today at $1,999, and its split screen is quietly shaped like an agent's workspace. Read → |
OCT 23 · AGENT SECURITY 1Password will log your agent in without showing it your password 1Password built the approval rope into the login itself: the agent asks, you approve, and it never sees the password. Read → |
OCT 23 · WASHINGTON The labs set lobbying records in Washington Nvidia, Anthropic, and OpenAI each spent more on lobbying last quarter than ever before. The rule-writing moved to Washington too. Read → |
|
Signals from leadersKrishna (IBM), All-In Podcast: the average enterprise uses 500 to 5,000 software vendors, and AI could let it land at 5 to 10 main platforms. Massive consolidation, on a six month to six year clock. Soderstrom (Spotify): Spotify's AI bet is about being the music service that agents choose. Being the default an agent reaches for is the new placement fight. MIT researchers: 13 agents left alone on a simulated island for 30 hours with no task climbed, built, drew, and invented games, and afterward generally beat fresh agents on assigned challenges. A pre-print, not peer reviewed. |
New models, tools, and hardwareOpenAI Decisions API, generally available: a dedicated endpoint for bounded answers (a probability, a choice, a score) on GPT-6 Luna, up to 10 times faster than the Responses API by OpenAI's account, at 0.10 dollars per million input tokens. It competes head on with TypeSafe's Jev. Claude Platform on Google Cloud, Anthropic managed: Claude through your Google Cloud project, billed through Marketplace against existing commitments. Google's shelf now holds Anthropic run Claude, Google run Claude, and GPT in private preview. Microsoft Copilot Code, Frontier rollout: describe a tracker, dashboard, or app inside the Copilot app and publish it under your organization's governance, with a path into engineering review as a GitHub draft pull request. NVIDIA DGX Spark 64GB, on sale today: same GB10 chip, 64GB of unified memory, from 4,999 dollars. Capable agents on hardware you own. |
For buildersKeep your skills in your own repo, version them, and write the description field like it matters, because that description decides when an agent loads the skill. Audit any skill you did not write: a skill folder can include scripts. Read them the way you read a dependency. Steal the Decisions pattern. Much of agent work is small bounded questions: route this, score that, is this a duplicate. A dedicated decision endpoint turns those calls into a cheap utility instead of a full chat model run. Copy 1Password's login flow anywhere your agent holds a credential: the agent identifies itself, a human approves one specific access, and the secret never touches the model. |
Career signalConsolidation is becoming the word to plan around. The middle of the market, small tools with no moat and no distribution, is getting squeezed from both ends. The roles that hold up sit at the seams: integrating platforms, governing agents across them, evaluating outputs, and owning the workflow knowledge that gets packaged into skills. A million published skills means a well written SKILL.md is now a legible, portable credential. Build two or three excellent ones in your field and publish them. |
One project ideaPackage one workflow you repeat every week as a portable Agent Skill: a SKILL.md that states the task, the steps, and what good output looks like, plus one reference file with your real examples. Run the same skill in two different agents on the same three inputs, log where the outputs match and where they drift, and publish the folder and the writeup. |
ClosingFive days, five versions of the same lesson, and today it got a foundation. The rivals who build the agents agreed the rules underneath should belong to nobody in particular, so they can belong to everybody who builds. Compete hard on the product, share the plumbing, and keep the human holding the keys at the login, the purchase, and the send button. (researched and drafted by my personal AI assistant and approved by me — full articles always at jeremymckellar.com) |
|