thevibeworks · lroolle · Jul 8, 2026 · Jul 8, 2026
diff --git a/content/.metadata.json b/content/.metadata.json
diff --git a/content/CHANGELOG.md b/content/CHANGELOG.md
diff --git a/content/blog/engineering/effective-harnesses-for-long-running-agents.md b/content/blog/engineering/effective-harnesses-for-long-running-agents.md
@@ -13,7 +13,7 @@ We developed a two-fold solution to enable the [Claude Agent SDK](https://platfo
 
 The Claude Agent SDK is a powerful, general-purpose agent harness adept at coding, as well as other tasks that require the model to use tools to gather context, plan, and execute. It has context management capabilities such as compaction, which enables an agent to work on a task without exhausting the context window. Theoretically, given this setup, it should be possible for an agent to continue to do useful work for an arbitrarily long time.
 
-However, compaction isn’t sufficient. Out of the box, even a frontier coding model like Opus 4.5 running on the Claude Agent SDK in a loop across multiple context windows will fall short of building a production-quality web app if it’s only given a high-level prompt, such as “build a clone of [claude.ai](http://claude.ai/redirect/website.v1.bca174ce-80b1-492e-a9e5-406e4054beb1).”
+However, compaction isn’t sufficient. Out of the box, even a frontier coding model like Opus 4.5 running on the Claude Agent SDK in a loop across multiple context windows will fall short of building a production-quality web app if it’s only given a high-level prompt, such as “build a clone of [claude.ai](http://claude.ai/redirect/website.v1.94106f97-b333-46a9-bf6d-5b6129e2cec5).”
 
 Claude’s failures manifested in two patterns. First, the agent tended to try to do too much at once—essentially to attempt to one-shot the app. Often, this led to the model running out of context in the middle of its implementation, leaving the next session to start with a feature half-implemented and undocumented. The agent would then have to guess at what had happened, and spend substantial time trying to get the basic app working again. This happens even with compaction, which doesn’t always pass perfectly clear instructions to the next agent.
 
@@ -34,7 +34,7 @@ In the updated [Claude 4 prompting guide](https://docs.claude.com/en/docs/build-
 
 ### Feature list
 
-To address the problem of the agent one-shotting an app or prematurely considering the project complete, we prompted the initializer agent to write a comprehensive file of feature requirements expanding on the user’s initial prompt. In the [claude.ai](http://claude.ai/redirect/website.v1.bca174ce-80b1-492e-a9e5-406e4054beb1) clone example, this meant over 200 features, such as “a user can open a new chat, type in a query, press enter, and see an AI response.” These features were all initially marked as “failing” so that later coding agents would have a clear outline of what full functionality looked like.
+To address the problem of the agent one-shotting an app or prematurely considering the project complete, we prompted the initializer agent to write a comprehensive file of feature requirements expanding on the user’s initial prompt. In the [claude.ai](http://claude.ai/redirect/website.v1.94106f97-b333-46a9-bf6d-5b6129e2cec5) clone example, this meant over 200 features, such as “a user can open a new chat, type in a query, press enter, and see an AI response.” These features were all initially marked as “failing” so that later coding agents would have a clear outline of what full functionality looked like.
 
 ```
 {

diff --git a/content/blog/engineering/equipping-agents-for-the-real-world-with-agent-skills.md b/content/blog/engineering/equipping-agents-for-the-real-world-with-agent-skills.md
@@ -85,7 +85,7 @@ We recommend installing skills only from trusted sources. When installing a skil
 
 ## The future of Skills
 
-Agent Skills are [supported today](https://www.anthropic.com/news/skills) across [Claude.ai](http://claude.ai/redirect/website.v1.9e5dc3d2-19b7-401c-bb25-4724e0906ec5), Claude Code, the Claude Agent SDK, and the Claude Developer Platform.
+Agent Skills are [supported today](https://www.anthropic.com/news/skills) across [Claude.ai](http://claude.ai/redirect/website.v1.94106f97-b333-46a9-bf6d-5b6129e2cec5), Claude Code, the Claude Agent SDK, and the Claude Developer Platform.
 
 In the coming weeks, we’ll continue to add features that support the full lifecycle of creating, editing, discovering, sharing, and using Skills. We’re especially excited about the opportunity for Skills to help organizations and individuals share their context and workflows with Claude. We’ll also explore how Skills can complement [Model Context Protocol](https://modelcontextprotocol.io/) (MCP) servers by teaching agents more complex workflows that involve external tools and software.
 

diff --git a/content/blog/engineering/how-we-contain-claude.md b/content/blog/engineering/how-we-contain-claude.md
@@ -15,7 +15,7 @@ The first is to supervise the agent’s behavior via a human-in-the-loop. Claude
 
 The second approach to capping the blast radius—and the focus of much of this post—is containment. Rather than supervising what the agent does, we supervise what it’s _able_ to do by enforcing access boundaries through, for example, sandboxes, virtual machines, and egress controls. This is where Anthropic engineering has devoted the most effort, and also where many of the most surprising security failures have occurred.
 
-Over the past two years, we’ve shipped three primary agentic products: [claude.ai](http://claude.ai/redirect/website.v1.846f50a1-2f9a-481d-bacd-de031776c9e6), Claude Code, and Claude Cowork. Each serves a different audience, requiring a different containment architecture. This article shares what’s held up, what’s broken, and what we’ve learned about agent security along the way.
+Over the past two years, we’ve shipped three primary agentic products: [claude.ai](http://claude.ai/redirect/website.v1.94106f97-b333-46a9-bf6d-5b6129e2cec5), Claude Code, and Claude Cowork. Each serves a different audience, requiring a different containment architecture. This article shares what’s held up, what’s broken, and what we’ve learned about agent security along the way.
 
 ## **Three types of risk, three components of defense**
 
@@ -49,13 +49,13 @@ _Three components to defend: the model, the environment in which it runs, and th
 
 ## **Patterns for containing agents**
 
-Focusing on the environment layer, we describe three isolation patterns and how they’re tailored for each Claude platform—[claude.ai](http://claude.ai/redirect/website.v1.846f50a1-2f9a-481d-bacd-de031776c9e6), Claude Code, and Cowork. We arrived at each design gradually, after finding the balance between the capabilities we need from the agent and the degree of intervention required from the user.
+Focusing on the environment layer, we describe three isolation patterns and how they’re tailored for each Claude platform—[claude.ai](http://claude.ai/redirect/website.v1.94106f97-b333-46a9-bf6d-5b6129e2cec5), Claude Code, and Cowork. We arrived at each design gradually, after finding the balance between the capabilities we need from the agent and the degree of intervention required from the user.
 
 ### **Pattern 1: The ephemeral container (claude.ai code execution)**
 
 Though best known as a chat interface, claude.ai also writes and runs code, generates files, and calls connectors. When Claude runs code inside claude.ai, it does so in a [gVisor](https://en.wikipedia.org/wiki/GVisor) container on isolated infrastructure. The agent is entirely server-side; no code runs on the local machine, and the filesystem is ephemeral (per-session). The blast radius is minimal, but so is the ceiling on what Claude can do—there's no persistent workspace and no access to the user's filesystem.
 
-This also makes [claude.ai](http://claude.ai/redirect/website.v1.846f50a1-2f9a-481d-bacd-de031776c9e6) subject to a more traditional threat model. We're not protecting user machines from agents; we're protecting our own infrastructure and each tenant from one another. Our pre-launch work for [claude.ai](http://claude.ai/redirect/website.v1.846f50a1-2f9a-481d-bacd-de031776c9e6) was dominated by traditional security work like network configuration, internal service auth, and orchestration.
+This also makes [claude.ai](http://claude.ai/redirect/website.v1.94106f97-b333-46a9-bf6d-5b6129e2cec5) subject to a more traditional threat model. We're not protecting user machines from agents; we're protecting our own infrastructure and each tenant from one another. Our pre-launch work for [claude.ai](http://claude.ai/redirect/website.v1.94106f97-b333-46a9-bf6d-5b6129e2cec5) was dominated by traditional security work like network configuration, internal service auth, and orchestration.
 
 That work reinforced the oldest lesson in security: the weakest layer is the one you built yourself. gVisor and [seccomp](https://en.wikipedia.org/wiki/Seccomp) have been hardened against well-resourced adversaries for far longer than agentic AI has existed, so the review effort went into the newer pieces we'd built around them. We’ll come back to this later, since our custom proxy is also the piece that broke in our most consequential incident.
 

diff --git a/content/blog/news/enabling-claude-code-to-work-more-autonomously.md b/content/blog/news/enabling-claude-code-to-work-more-autonomously.md
@@ -58,14 +58,18 @@ These updates are available now for Claude Code users.
 
 ## Related content
 
-### Anthropic appoints KiYoung Choi as Representative Director of Korea ahead of Seoul office opening
+### Government of Alberta uses Claude to find and fix cybersecurity vulnerabilities across government systems
 
-[Read more](https://www.anthropic.com/news/kiyoung-choi-representative-director-anthropic-korea)
+Since 2025, the Government of Alberta has been using Claude Code with both Opus and Sonnet models to review its systems, find vulnerabilities, and fix them.
 
-### Anthropic co-founder Chris Olah's remarks on Pope Leo XIV's encyclical "Magnifica humanitas"
+[Read more](https://www.anthropic.com/news/alberta-government-claude-cybersecurity)
 
-[Read more](https://www.anthropic.com/news/chris-olah-pope-leo-encyclical)
+### More details on Fable 5’s cyber safeguards and our jailbreak framework
 
-### Widening the conversation on frontier AI
+[Read more](https://www.anthropic.com/news/fable-safeguards-jailbreak-framework)
 
-[Read more](https://www.anthropic.com/news/widening-conversation-ai)
+### Introducing Claude Sonnet 5
+
+Sonnet 5 delivers frontier performance across coding, agents, and professional work at scale.
+
+[Read more](https://www.anthropic.com/news/claude-sonnet-5)
diff --git a/content/blog/research/81k-economics.md b/content/blog/research/81k-economics.md
@@ -91,6 +91,18 @@ There are key caveats to our analysis, owing to the nature of the data. First, o
 
 Still, the interviews surface real insights about people’s feelings around the economics of AI, showing how qualitative data can surface quantitative hypotheses. And the large share of economic-related concerns is a strong signal in itself.
 
+### Citation
+
+```
+@online{massenkoff2026interviewer,
+author = {Maxim Massenkoff and Saffron Huang},
+title = {What 81,000 people told us about the economics of AI},
+date = {2026-04-22},
+year = {2026},
+url = {anthropic.com/research/81k-economics},
+}
+```
+
 ### Appendix
 
 See the final section of the [linked PDF](https://cdn.sanity.io/files/4zrzovbb/website/3a8d990bc90098038eabd77b0d12ff636ed58d50.pdf).
@@ -107,20 +119,20 @@ Additionally, we thank Miriam Chaum, Ankur Rathi, Santi Ruiz, and David Saunders
 
 ## Related content
 
-### Coding agents in the social sciences
+### A global workspace in language models
 
-Results from a survey of 1,260 social scientists about AI and coding agent use.
+New interpretability research reveals an emergent mental workspace in Claude that holds internal thoughts that don’t appear in the model’s output.
 
-[Read more](https://www.anthropic.com/research/coding-agents-social-sciences)
+[Read more](https://www.anthropic.com/research/global-workspace)
 
-### Project Glasswing: An initial update
+### Anthropic Economic Index report: Cadences
 
-An early update on what we've learned from Project Glasswing.
+In our latest Economic Index report, we sample hourly for the first time to ask: When do people come to Claude? What do they produce with it? And how do they perceive AI's impact on their work?
 
-[Read more](https://www.anthropic.com/research/glasswing-initial-update)
+[Read more](https://www.anthropic.com/research/economic-index-june-2026-report)
 
-### 2028: Two scenarios for global AI leadership
+### Project Fetch: Phase two
 
-Our views on the AI competition between the US and China.
+We report results from our latest test of whether Claude can help Anthropic employees perform sophisticated robotics tasks. We found that Claude Opus 4.7, operating without human assistance, was about 20 times faster than the fastest human team at all tasks completed by participants less than a year ago.
 
-[Read more](https://www.anthropic.com/research/2028-ai-leadership)
+[Read more](https://www.anthropic.com/research/project-fetch-phase-two)
diff --git a/content/blog/research/AI-assistance-coding-skills.md b/content/blog/research/AI-assistance-coding-skills.md
@@ -3,7 +3,7 @@ Title: How AI assistance impacts the formation of coding skills
 URL Source: https://www.anthropic.com/research/AI-assistance-coding-skills
 
 Markdown Content:
-Research shows AI helps people do [parts](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4945566) of their job faster. In an observational [study](https://www.anthropic.com/research/estimating-productivity-gains) of [Claude.ai](http://claude.ai/redirect/website.v1.609423b6-230c-43ef-a7ee-385200ce6008) data, we found AI can speed up some tasks by 80%. But does this increased productivity come with trade-offs? Other research shows that when people use AI assistance, they become [less engaged with their work](https://www.nature.com/articles/s41598-025-98385-2) and [reduce](https://www.microsoft.com/en-us/research/wp-content/uploads/2025/01/lee_2025_ai_critical_thinking_survey.pdf) the effort they put into doing it—in other words, they offload their thinking to AI.
+Research shows AI helps people do [parts](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4945566) of their job faster. In an observational [study](https://www.anthropic.com/research/estimating-productivity-gains) of [Claude.ai](http://claude.ai/redirect/website.v1.94106f97-b333-46a9-bf6d-5b6129e2cec5) data, we found AI can speed up some tasks by 80%. But does this increased productivity come with trade-offs? Other research shows that when people use AI assistance, they become [less engaged with their work](https://www.nature.com/articles/s41598-025-98385-2) and [reduce](https://www.microsoft.com/en-us/research/wp-content/uploads/2025/01/lee_2025_ai_critical_thinking_survey.pdf) the effort they put into doing it—in other words, they offload their thinking to AI.
 
 It’s unclear whether this cognitive offloading can prevent people from growing their skills on the job, or—in the case of coding—understanding the systems they’re building. Our latest study, a randomized controlled trial with software developers as participants, investigates this potential downside of using AI at work.