Anthropic isn’t slowing down

Welcome back to The Shift. Let’s get straight to what matters in AI today…

Today we have:

🤖 Anthropic Just Released Claude Opus 4.7

🧑‍⚕️ How to Connect Your AI Agent to a Human Expert With Humwork

🖥️ OpenAI Just Turned Codex Into a Lot More Than a Coding Tool

🔨Tools and Shifts you cannot miss

🤖 Anthropic Just Released Claude Opus 4.7

Anthropic released Claude Opus 4.7 this week, its new top publicly available model that tops GPT-5.4 and Gemini 3.1 Pro on agentic coding, though Anthropic's own unreleased Mythos Preview still sits above it.

The Shift:

What improved? Opus 4.7 jumps from 53.4% to 64.3% on the SWE-bench Pro coding benchmark. It handles complex long-running tasks with less supervision, verifies its own outputs, and accepts images up to 3.75 megapixels, more than three times prior Claude models.

What else is new? A new xhigh effort level lands between high and max, and Claude Code defaults to xhigh for all plans. The new /ultrareview slash command flags bugs and design issues in a dedicated review session. Pro and Max users get three free ultrareviews to try it.

What is the cost? Pricing stays identical to Opus 4.6 at $5 per million input tokens and $25 per million output tokens. The catch is token usage; the updated tokenizer means the same input can now map to roughly 1.0 to 1.35 times more tokens, depending on content type.

Anything else to know? Opus 4.7 has built-in cybersecurity safeguards that automatically block high-risk requests. Security professionals can join the new Cyber Verification Program for legitimate use. Early reactions from users are divided despite the strong benchmark numbers.

Opus 4.7 is available now across all Claude products, the API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. If you are migrating from Opus 4.6, check the migration guide, the instructions following changes alone may affect existing prompts.

Together with hCaptcha

In a World of AI Agents: Intent > Identity

AI-powered bots aren’t just logging in anymore. They’re mimicking real users, slipping past identity checks, and scaling attacks faster than ever.

Thousands of companies worldwide trust hCaptcha to protect their online services from automated threats while preserving user privacy.

Now is the time to take control of your security.

Book a demo today.

🧑‍⚕️ How to Connect Your AI Agent to a Human Expert With Humwork

When your agent gets stuck, Humwork connects it to a verified human expert in real time, automatically, in under 30 seconds.

Step 1: Connect Humwork via MCP - Add Humwork to your agent stack through MCP. Setup takes 60 seconds. One integration gives your agent access to 1,000+ verified experts across every domain.

Step 2: Agent hits a wall and calls Humwork - When your agent encounters a bug it can't fix, a design decision it can't make, or a problem outside its capability, it calls Humwork automatically. No manual intervention needed.

Step 3: Expert gets matched in under 30 seconds - Humwork routes the request to the right verified expert, engineer, designer, strategist, based on what the problem actually needs.

Step 4: Expert talks directly to your agent - The expert sees the full context. No re-explaining. They diagnose the issue and push the solution directly back into your agent's context.

Step 5: Agent keeps going - The solution gets picked up and the agent resumes exactly where it left off, like nothing happened.

Average first reply under 2 minutes. 87% resolution rate. 24/7 across every timezone. You can get started here.

🖥️ OpenAI Just Turned Codex Into a Lot More Than a Coding Tool

OpenAI released a major Codex update this week, shifting it from a coding agent into a full development platform with background computer use, memory, an in-app browser, image generation, and parallel agents all in one app.

The Shift:

What Does It Do Now? Background computer use lets Codex operate any Mac app on its own with multiple agents running in parallel. It sees, clicks, and types with its own cursor across any app, even those without an API.

What Else Is New? An in-app browser lets developers mark up pages to give Codex precise instructions. Inline image generation via gpt-image-1.5 creates mockups without switching apps. Over 90 new plugins are live, including JIRA, GitLab, and Microsoft Suite.

How Does It Carry Work Forward? Memory retains preferences and context across sessions, so future tasks complete faster. Automations let Codex schedule and resume long-running work across days or weeks without any manual triggering.

Who Can Use It? The update is live now for desktop app users signed in with ChatGPT. Codex has 3 million weekly users with 70% month-over-month growth. Computer use is macOS only for now.

Codex head Thibault Sottiaux said OpenAI is building the super app out in the open. With background computer use, memory, and parallel agents now in one place, that description is starting to feel accurate.

Together with The Code

What 200K+ Engineers Read to Stay Ahead

Your GitHub stars won't save you if you're behind on tech trends.

That's why over 200K engineers read The Code to spot what's coming next.

• Get curated tech news, tools, and insights twice a week

• Learn about emerging trends you can leverage at work in just 5 mins a day

• Become the engineer who always knows what's next

Join 200k+ engineers

🔨AI Tools for the Shift

🧠 RedShip - Track real conversations where people need your solution and turn Reddit into a high-intent acquisition channel.

📸 StageFlow - Generate professional product photos instantly for your shop without needing a camera or studio.

🤖 Fello AI - An AI coworker that helps you handle real tasks, conversations, and daily work more efficiently.

🔗 APIClaw - Access real-time commerce data streams so your AI agents can analyze thousands of products at scale.

📩 Calyo - Automate social media prospecting, outreach, and negotiations with AI that finds and engages the right leads.

🚀Quick Shifts

🪖 Google is reportedly exploring allowing Gemini in classified Pentagon settings, signaling a shift toward deeper military AI integration beyond its current unclassified use limits.

🎬 Luma launched an AI-powered production studio with Wonder Project, using real-time “hybrid filmmaking” to create high-quality films faster and cheaper with AI-driven creative workflows.

🎨 Anthropic’s CPO left Figma’s board amid plans for new AI design tools, signaling growing tension as AI labs move to compete directly with traditional software platforms.

🏛️ The U.S. government may grant select agencies access to Anthropic’s Mythos AI, despite ongoing legal disputes, highlighting rising tensions between regulation and AI adoption.

🤖 Windsurf launched 2.0, adding an Agent Command Center to manage parallel AI agents and integrating Devin directly into the IDE for more autonomous development workflows.

🧩 Prompt of the Day

How to Analyze Price Sensitivity Without Killing Conversions

Most brands guess pricing. The smartest brands test how price changes reshape demand, perception, and buying urgency.

Turn pricing into a growth lever, not just a margin decision.

Paste the prompt: Drop this into ChatGPT, then fill in your product details.

Create a price sensitivity analysis for [Insert product or category]. Explore multiple pricing scenarios including price increases, discounts, and bundle offers. For each scenario, estimate how it impacts conversion rate, demand, and perceived value. Identify the tipping point where higher prices reduce volume and where lower prices hurt brand perception. Highlight key customer reactions and behavioral shifts. Focus on how these insights can be used to optimize pricing, promotions, and overall revenue. Keep it clear, strategic, and grounded in real buying behavior.

🤳AI Nugget of the Day

That’s all for today’s edition see you tomorrow as we track down and get you all that matters in the daily AI Shift!

If you loved this edition let us know how much:

Forward it to your pal to give them a daily dose of the shift so they can 👇