In partnership with
🧠 THAT ONE AI - This Week’s Signal
Here’s what’s shaping AI right now - without the noise:
🧩 Anthropic folded Cowork and Design into Claude and added Docs and Slides
⚙️ TypeSafe shipped a model that cannot write a sentence and costs almost nothing
🗣️ Gemini 3.8 Live keeps talking while it reasons
🧠 That One AI Tip: write Claude skills that actually fire
🧰 Tools worth testing
🧩 Anthropic Folded Cowork and Design Into Claude, and Added Docs and Slides
Anthropic merged Cowork and Design into Claude chat. Two new products arrived with the change: Claude Docs and Claude Slides.
The real change is routing. You start a conversation. Claude brings in the right tool by itself. Your skills and your context stay in one place across all of them.
It is rolling out now. Pro and Max plans go first. Read the announcement.
The bigger signal: 👉 Context is now the product. The chat window, the document and the slide deck are just views of the same memory.
Want to get the most out of ChatGPT?
ChatGPT is a superpower if you know how to use it correctly.
Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.
Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.
⚙️ TypeSafe Shipped a Model That Costs $42 per Billion Tokens and Cannot Write
TypeSafe came out of stealth with Jev. The company calls it a System One model. It chooses between options you define in advance and returns a confidence score. It cannot produce free text at all.
The pricing is the part to read twice. Input runs $42 per billion tokens. Output is free. TypeSafe puts that at roughly 238 times cheaper than Claude Fable 5.1. Responses land in 70 to 500 milliseconds, which the company claims is 40 to 200 times faster than current large language models.
The training method is new as well. TypeSafe uses an algorithm it calls RLCD. Founder Diogo Almeida, who worked on the research behind ChatGPT, describes the result as closer to a database than a coworker. The intended jobs are routing requests, scoring records, screening another model's output for jailbreaks, and gating what reaches an expensive model. Read the launch post.
The bigger signal: 👉 A large share of your agent stack is doing classification and calling it reasoning. Free output turns that line item into an architecture decision.
🗣️ Gemini 3.8 Live Removes the Thinking Pause From Voice Agents
Google shipped Gemini 3.8 Live. The models keep speaking while they reason, so the audio never stops while the system works out an answer.
There is an Extended Thinking variant for harder queries. It currently leads the public speech to speech quality ranking.
For anyone building a voice product, that pause was the hardest thing to design around. Teams filled it with hold music and filler phrases. Those workarounds can come out now. Read Google's post.
The bigger signal: 👉 Voice agents competed on latency for two years. That race is closing. The next one is about what the model says while it is still deciding.
🧠 That One AI Tip: Write Claude Skills That Actually Fire
You wrote a skill. Claude ignores it. This is the most common complaint about skills, and the fix is almost always in one field.
Only the frontmatter is loaded upfront
Claude reads the name and description of every installed skill at the start of a session. It does not read the body. The body loads later, and only if the skill fires.
That means the description carries the entire decision. A perfect 400-line skill body does nothing if the description does not win the match.
Treat the description field as a trigger list
Most people describe what the skill does. Claude needs to know when to reach for it.
State the situation in plain words. Name the exact phrases a user would type. Then say it directly:
yaml
---
name: pr-review
description: >
Reviews a pull request against our team checklist. Use this whenever
the user says "review this PR", "check my diff", "look at this branch",
or asks whether a change is ready to merge. Also use it when the user
pastes a diff with no instruction. Trigger this skill before giving any
code feedback of your own.
---The phrase "use this whenever" does more work than any amount of description polish.
Add the word "skill" to common trigger words
Short, common triggers collide with normal conversation. A skill triggered by "plan" will fight every ordinary use of the word.
Append the word to the trigger instead. "Plan skill" or "review skill" gives the model a clean signal that a casual sentence will not produce by accident.
Reinforce it once in CLAUDE.md
Add a short block to CLAUDE.md that names each skill and the moment it should run:
Skills and when to use them:
- pr-review: run before you comment on any diff or branch.
- release-notes: run when I mention shipping, tagging, or a version number.
- incident: run the moment I paste a stack trace or an error log.Two places agreeing on the same rule is far more reliable than one.
The bigger signal: 👉 A skill lives or dies at retrieval time. You are competing for a match against every other skill in the folder, and the description is your only entry.
Turn AI into Your Income Engine
Ready to transform artificial intelligence from a buzzword into your personal revenue generator
HubSpot’s groundbreaking guide "200+ AI-Powered Income Ideas" is your gateway to financial innovation in the digital age.
Inside you'll discover:
A curated collection of 200+ profitable opportunities spanning content creation, e-commerce, gaming, and emerging digital markets—each vetted for real-world potential
Step-by-step implementation guides designed for beginners, making AI accessible regardless of your technical background
Cutting-edge strategies aligned with current market trends, ensuring your ventures stay ahead of the curve
Download your guide today and unlock a future where artificial intelligence powers your success. Your next income stream is waiting.
That One AI 🧰 TOOLBOX
A few tools quietly worth exploring:
⚙️ Fusion → Devin's harness that pairs a frontier planner with a cheaper coding model.
🛒 ZeroClick → Makes your product listings readable and buyable by AI shopping agents.
😴 Dreambeans → Google Labs experiment that pulls a short daily list of suggestions out of your own apps.
🗂️ Resurf → A local context library for Mac and iPhone that feeds your notes and files to AI over MCP or CLI.
🔚 EXIT NODE
Three of this week's four stories are about consolidation. Anthropic collapsed its surfaces into one. TypeSafe priced the boring half of your stack at almost nothing.
The pattern underneath is that the interesting decisions moved. They are no longer about which model is smartest.
See you next issue.