← Back to all issues
Issue 1 · Regular · Free

Claude Cowork is live. Here is what to do with it.

Four launches this week. One new skill. What to use and what to skip.

May 12, 2026·7 min read

What shipped this week

Four launches matter this week. Here is the operator take on each. **Claude Cowork went GA.** The shared-workspace product Anthropic teased in January is now live for all paid plans. If you spend more than 2 hours a day in chat threads with teammates, move them into a Cowork space this week. The shared context model means you stop pasting the same project background into every new chat. **Claude Design hit research preview.** Anthropic shipped the design-focused interface on April 17. It is not production ready. If you ship product UI, get on the waitlist and test it on 1 real project. Do not move your design work over yet. **Claude Code desktop got a full redesign.** New panel layout, faster skill loader, native tab support. If you already use Claude Code, update today. The skill loader alone saves a few seconds per invocation. **Codex now runs on GPT-5.5.** OpenAI quietly swapped the model underneath Codex last week. If you write code and you have been on the fence about Codex versus Claude Code, the gap is now small enough that you should test Codex on 1 real task this week. **Bonus: gpt-image-2 replaces DALL-E 3 today.** Same API, better outputs, lower cost per image. If you generate marketing images, your existing API calls keep working. Check the new sample outputs before your next campaign.

The new skill: weekly-product-update-brief

This is the skill that produces the section you just read. It reads 3 changelog URLs and outputs a 1-page operator brief: what to use, what to ignore, what to watch. Why now: most operators read changelogs reactively when something breaks. The 10 minutes a week to read proactively is what separates the operators who compound their stack from the ones who get surprised every quarter. Drop the SKILL.md into your ~/.claude/skills/ folder. Run it Monday morning. Feed it 3 URLs from products you pay for. The brief that comes back is the meeting agenda for your Tuesday team standup. The full skill body is at the bottom of this issue. Copy it, save it, run it once before you close this email.

Pattern I am seeing

The race between Claude Code skills and Codex skills is a leading indicator of where agentic work compounds first. Both products are converging on the same shape from different starting points. Claude Code began as a coding agent and is now generalizing into operator skills. Codex began as a code completion product and is now adding the same skill surface. The two products are aiming at the same point from opposite directions, and the operator who already has 5 working skills will get 10x more value from whichever product wins than the operator who treats them as chat windows. If you write 1 skill a week for the rest of 2026, you finish the year with 30 reusable workflows. That is the moat. The model swap underneath does not change it.

What I would skip

Skip the AI meeting note tools that launched 4 new features this month. Otter, Fireflies, Granola, all of them. The category is in a feature war that does not change the outcome: you still have to read the notes, decide what matters, and write the decision down. The 5 minutes you save on transcription get eaten by the 10 minutes of scrolling AI-generated summaries that miss the 1 thing the meeting actually decided. If you already use one, fine. Do not switch this week. Do not add a second one. Spend the evaluation time on the skill above instead.

New skill this issue

weekly-product-update-brief

Read 3 product changelog URLs and output a 1-page operator brief on what to use, ignore, or watch.

Why this week

Most operators read changelogs reactively. This skill flips it to a proactive 10-minute weekly review that produces the agenda for your Tuesday standup.

---
name: weekly-product-update-brief
description: Read 3 product changelog URLs and output a 1-page operator brief on what to use, ignore, or watch. Run weekly on Monday morning before your team standup.
---

You are a compiler that turns 3 raw product changelogs into a single 1-page operator brief. The brief tells the user what to use this week, what to ignore, and what to watch. You are not summarizing marketing copy. You are producing operator decisions.

## When to run

Monday morning. Block 10 minutes on your calendar. Run before the team standup so you can name 1 product change in the meeting and ship a decision the same day.

## Inputs

The user provides 3 things. If any are missing, ask once and stop.

- [CHANGELOG_URLS]: 3 URLs to product changelogs, release notes, or "what shipped" pages
- [PRIORITY_AREAS]: 1-2 sentences naming the work the user is doing this week (example: "Closing 3 enterprise deals and rewriting onboarding copy")
- [STACK]: 1 sentence on what tools the user already pays for (example: "Claude Code, Claude Cowork, Linear, Notion")

## Steps

1. Fetch each URL. If a URL fails to load, write the URL on its own line under a "Failed to fetch" section at the bottom and continue.
2. For each changelog, extract every named feature, product, or release. Use the exact product name as written. Never abbreviate, expand, or invent names.
3. Classify each item as Use, Ignore, or Watch using the rules below.
4. Write the brief in the exact output format.
5. Run the self-check at the end before responding.

## Classification rules

Use = the user should turn it on this week. The feature is shipped (not a preview), it sits in the user's existing stack or replaces something they already pay for, and it touches their [PRIORITY_AREAS]. Maximum 3 items classified as Use per brief.

Ignore = the feature exists but is not for this user. It is for a different audience (enterprise when user is solo, technical when user is non-technical), or it duplicates something the user already has working. Be specific about why.

Watch = the feature is in preview, beta, or research release. The user should know it exists but should not change workflows yet. Include the expected GA date if the changelog gives one. If no date is given, write "No GA date stated."

## Output format

Use this exact structure. No extra sections. No preamble.

```
Weekly product update brief
[Today's date in YYYY-MM-DD]

## Priority context
[1 sentence echoing the user's PRIORITY_AREAS so the reader can confirm the brief was scoped right]

## Use this week (max 3)
1. [Product name] -- [Feature name] -- [1 sentence on the specific action to take this week, starting with a verb]
2. ...
3. ...

## Ignore (max 5)
1. [Product name] -- [Feature name] -- [1 sentence on why it does not apply to this user]
2. ...

## Watch (max 5)
1. [Product name] -- [Feature name] -- [Expected GA date or "No GA date stated"] -- [1 sentence on what would change for the user when it ships]
2. ...

## The 1 decision for this week
[1 sentence naming the single action the user will take today, before they close this brief]
```

## Rules

- Use the exact feature name as written in the changelog. If a changelog calls it "Cowork", write "Cowork", not "the new collaboration feature".
- Every Use item must start with a verb (Turn on, Replace, Switch, Move).
- No item can appear in more than 1 category.
- Maximum total length: 400 words.
- If a changelog has zero items that apply to the user, write "Nothing applicable this week" under the relevant section. Do not pad.
- Source URLs go at the bottom under a "Sources" section, 1 URL per line.

## Banned phrases

Do not use any of these:
- exciting new features
- best-in-class
- industry-leading
- next-generation
- cutting-edge
- game-changer
- supercharge
- unlock
- transform
- streamline

If the changelog uses any of these words about a feature, strip them and write only the concrete change.

## Self-check before responding

Confirm all 5:

1. Did I use the exact product and feature names from the changelogs?
2. Is every Use item a verb-led action the user can take today?
3. Did I avoid all banned phrases?
4. Is the brief under 400 words?
5. Does the "1 decision for this week" name a specific action, not a category?

If any answer is no, fix the brief before responding.

What shipped

  • Claude Cowork -- General availability for all paid plansuse this now

    Move your team chat threads into a Cowork space this week if you spend 2+ hours a day in shared context with teammates. Stops the paste-the-background-again loop.

    Source
  • Claude Design -- Research preview released April 17, 2026watch closely

    Not production ready. Get on the waitlist and test on 1 real project. Do not move your design work over yet. No GA date stated.

    Source
  • Claude Code -- Desktop app redesign with new panel layout and faster skill loaderuse this now

    Update today if you already use Claude Code. The new skill loader saves a few seconds per invocation and the tab support makes multi-project work cleaner.

    Source
  • Codex -- Now running on GPT-5.5use this now

    OpenAI swapped the model last week. The gap with Claude Code closed enough that you should test Codex on 1 real coding task this week and compare.

    Source
  • ChatGPT -- gpt-image-2 replaces DALL-E 3 on May 12, 2026use this now

    Same API, better outputs, lower cost per image. Existing API calls keep working. Check the new sample outputs before your next marketing campaign.

    Source
Pattern observation

The race between Claude Code skills and Codex skills is a leading indicator of where agentic work compounds first. Both products are converging on the same shape from different starting points. Claude Code began as a coding agent and is generalizing into operator skills. Codex began as code completion and is adding the same skill surface. They are aiming at the same point from opposite directions. The operator who already has 5 working skills will get 10x more value from whichever product wins than the operator who treats them as chat windows. Write 1 skill a week for the rest of 2026 and you finish the year with 30 reusable workflows. That is the moat, and the model swap underneath does not change it.

Want every issue?

$37/mo or $370/yr. Cancel any time.

Subscribe