Skip to content
Weekly newsletter

The robots now share one instruction manual

Agent Skills goes neutral, a forest map gets sharp, Sweden bans the chatbots


ciaosapiens
Mon, Oct 26, 20264 min read8 stories + quick hits

Hey, sapiens.

Bad news: the machines just agreed on a common file format, which is exactly how every good heist movie starts. This week Anthropic hands a standard to a foundation, DeepMind maps the planet's trees, and Sweden shows its government staff the chatbot door.

Story 1 · Lead1 min

Robot of the day: one folder to teach every agent

The gist

Anthropic is donating Agent Skills to the Agentic AI Foundation, which means a standard born inside one company now belongs to a neutral home. The idea is almost insultingly simple: a folder, a SKILL.md file with plain instructions for one job, and whatever scripts or reference docs the agent needs alongside. That's the whole trick. No grand protocol, no doorstop of a spec. It grew out of Claude Code. The agent was fine at general knowledge work but missed the step-by-step know-how real business chores need. Within weeks of the October 2025 launch, developers had published thousands of skills on GitHub and written loaders for other platforms. Anthropic then released it as an open standard in December 2025. The adoption count is the headline: more than 70 agent platforms, including ChatGPT, Cursor, GitHub Copilot and VS Code, have integrated it. Anthropic promises the standard stays open and vendor-neutral, steered by its community. Why care, sapiens? When rival agents all read the same instruction folder, your expertise becomes portable. Write a skill once, and every robot colleague can learn it. Slightly unnerving. Mostly convenient. Your move: turn one weekly chore into a skill.

Story 260 sec

DeepMind maps every forest at 10-meter detail

The gist

Google DeepMind and Google Research, with World Resources Institute and IIASA, released a global map sorting land into natural forests, plantations and farmed areas at 10-meter resolution. It also includes an early time series tracking yearly change from 2017 to 2025.

Why it matters

From orbit, rainforest and oil palm estates look like the same green blob. Telling them apart shows where conservation works and where forests are being converted. The map is also a baseline for the EU's deforestation regulation.

Story 360 sec

OpenAI ships a Decisions API for quick, cheap calls

The gist

OpenAI announced the Decisions API on October 22, 2026, pitching it as fast, focused, low-cost decision-making for any application. The launch post is short and the tagline does most of the talking.

Why it matters

It's positioned as a way to build more responsive apps, which suggests small, cheap model calls are becoming their own product category. Details on pricing and availability are thin, so judge it once the docs land, not the slogan.

Story 460 sec

OlmoEarth v1.3 puts the whole planet in embeddings

The gist

Allen Institute released OlmoEarth v1.3, a satellite embedding model, along with precomputed annual embeddings for the whole planet at 10m, in 64- and 128-dimension variants. A browser-based Embeddings Explorer comes with it.

Why it matters

Until now, using OlmoEarth meant fine-tuning with GPU access. The explorer needs no sign-in and no advanced technical knowledge, so tracking change over time or finding similar places is now open to anyone with a browser.

Meanwhile in Europe

Story 5 · Europe60 sec

Swedish AI provider open-sources per-request CO2 tracking

The gist

Stockholm-based Berget AI released an open-source method for calculating emissions per AI request, built with Stockholm Environment Institute, WattTime and Klimatkollen. Its headline number: the same request emits 8.5 times more CO2 in Texas than in Sweden.

Story 6 · Europe60 sec

Swedish government offices ban ChatGPT and Claude

The gist

Sweden's government offices have banned ChatGPT and Claude, reflecting concerns about data security and intellectual property in public sector work. One reliable source so far, but the signal on data sovereignty is loud.

Story 7 · Europe60 sec

OpenAI teams up with Poland's NASK on AI safety

The gist

OpenAI and NASK, Poland's National Institute for Telecommunications, announced a partnership on AI safety, security and technical research, aiming to strengthen safeguards and build a more resilient digital ecosystem.

Survival kit

Story 8 · Kit60 sec

Survival kit: write your first SKILL.md

The gist

Think of a skill as a recipe card for your agent: a SKILL.md with the steps for one task, plus any helper scripts or docs. Pick a chore you repeat weekly, like a report format or a review checklist. Write the steps in plain language, drop in your templates, and load the folder into an agent that supports the standard. Over 70 platforms already do, including Cursor, GitHub Copilot and VS Code.

Quick hits1 min
Anthropic opens an 8-12 week guest researcher program for life sciences, with free model access for academics and nonprofits.Hugging Face launches a Space that checks whether a model runs on AWS Trainium via torch-neuronx.LitPivot helps researchers pivot when the literature says their big idea already exists.AWS publishes a TorchNeuron profiling guide for squeezing more out of Trainium.Anthropic and Theorem show LLMs can produce formal proofs at scale, making verified software less of a years-long slog.Apollo research sees no clear signs of AI payoff outside the tech sector. The spreadsheets are unimpressed.Twelve AI models were asked for a controversial take on humans; 29 of 35 answers said we put group loyalty over truth. Rude, sapiens.Caristo's AI found coronary inflammation in over 42% of asymptomatic adults in a US study by Biograph.
Our take

Write one skill this week, before the agents write it about you. Ciao, sapiens.

Ciao, sapiens.