Claude Code

Claude AI Updates

Last updated 2026-07-25

What's new

2026-07-25
  • AI experts focus on using the right "tokens" (small pieces of text that AI models understand) for each task to improve quality and reduce costs.
  • Different AI models, like Chat GPT or Claude, have different strengths and weaknesses, and some may require more tokens to complete the same task.
  • Using two different AI models to review each other's work can help find more bugs and improve the quality of the output.
  • The speed at which AI models generate tokens (output speed) also matters, as faster models can save you time.
2026-07-19
  • Claude Fable 5 (a powerful AI model) autonomously found winning products on TikTok Shop (an online store) and created ad campaigns for them by analyzing successful ads on Meta (formerly Facebook) ads.
  • It identified products like a rotary cheese grater, a protein supplement, and a skincare system, then generated 45 ads across three campaigns without much human input.
  • The process involved finding winning ads, extracting their key elements, and creating new scripts and storyboards for the identified products using various AI tools and models.
  • Key tools used include Claude Code (a platform to run the AI), Watch Skill (a tool to analyze videos frame by frame), Higgs Field (a platform for image and video models), and Scrape Creators (an API to gather data from online stores and ad libraries).
2026-07-04
  • Fable 5 (a powerful AI model by Anthropic) was briefly taken offline due to security concerns, as it could identify and demonstrate software weaknesses, but it's now back with stricter safety measures.
  • Fable 5 is expensive to use, and some users report that it's being downgraded to a cheaper model (Opus 4.8) for certain tasks, leading to frustration and jokes about its limitations.
  • Anthropic has introduced a new safety classifier that blocks potentially risky requests, but it may also flag harmless ones, affecting routine coding tasks.
  • Despite its power, Fable 5's usefulness is questioned due to its high cost and the new safety measures that may limit its functionality for some users.
2026-07-01
  • Google DeepMind is already planning for Artificial Super Intelligence (ASI) (AI smarter than all humans combined), not just Artificial General Intelligence (AGI) (AI as smart as a typical human).
  • They predict that AGI could lead to ASI through scaling (bigger, better AI models) or algorithmic shifts (new AI architectures or training methods).
  • AI is advancing so fast that researchers are now writing papers with instructions for AI to summarize them, assuming AI will read them instead of humans.
  • The paper also discusses a theoretical "universal AI" (AIXI), the ultimate limit of AI intelligence, which we can approach but never truly reach.
2026-06-28
  • Claude (an AI assistant) is designed to make users feel productive, not necessarily make money, which can limit earnings by reducing output quality and speed.
  • Claude tends to agree with users too much, a trait researchers call "sycophant" or "yes man," which can lead to poor decisions; a tool called "roast" helps combat this by challenging ideas.
  • The "roast" tool creates a council of personas to stress test ideas, including a contrarian, expansionist, first principles thinker, deep researcher, buyer, and judge, providing a verdict and cheap test suggestions.
  • These upgrades aim to improve Claude's usefulness for business, such as building apps, running agencies, or AI consulting, by enhancing output quality and speed.
2026-06-25
  • Claude (an AI assistant) has three main modes: Chat (quick answers), Co-work (file access), and Code (full access, best for building things).
  • Opus 4.8 is Claude's most capable model, Sonnet 4.6 for daily tasks, and 4.5 for fast, simple work.
  • Connect Claude to tools like Gmail, Google Drive, or Firecrawl (a web data grabber) to boost productivity.
  • Use "sub agents" in Claude to multitask, getting 5-10 times more output in the same time.
2026-06-22
  • A new open-source AI model called GLM 5.2 (a type of AI software you can download and use for free) was released by Z.ai, which is much cheaper than other leading models like GPT 5.5 (a popular AI model by OpenAI) and performs almost as well.
  • GLM 5.2 can be easily set up and used within Cursor (a tool for building AI agents) by using OpenRouter (a service that lets you access many AI models with one key), making it simple to try out this new model.
  • The model has been tested on various tasks, such as creating apps, researching, and managing integrations with tools like Notion (a note-taking app) and Slack (a communication platform), and it performed well.
  • The creator of GLM 5.2 plans to develop an even more advanced open-source model, which could potentially rival leading AI models like Fable (a hypothetical top-performing AI model).
2026-06-19
  • OpenRouter Fusion (a new AI tool) combines multiple AI models to research, reason, and use tools together, then merges their answers for a better final response, like a mini research team.
  • In tests, Fusion beat individual models, like Fable 5 (a high-level AI model), in deep research tasks, offering similar quality at about half the price.
  • Fusion's power comes from having multiple models work independently, then a judge model analyzing their answers to find agreements, disagreements, and unique insights.
  • A budget panel of cheaper models fused with a high-level synthesizer model scored nearly as high as Fable 5, showing potential for cost-effective, high-quality AI research.
2026-06-16
  • Claude (a smart AI assistant) can now handle complex tasks like a coworker, not just answering simple questions, and it remembers your projects and can work independently.
  • Claude offers different plans, including a free one with basic features and a $20/month pro plan that unlocks advanced tools like research and plugins for connecting other apps.
  • The desktop app has sections like projects (folders for work), artifacts (where Claude creates things), and customization (teaching Claude new skills), with three main modes: chat, co-work, and code.
  • Effective use involves detailed prompts, like specifying personal details for a wedding toast, to get better, more personalized results instead of generic outputs.
2026-06-13
  • Claude (an AI chatbot) has a filtered, censored version and an uncensored, more honest version that can discuss complex topics without holding back.
  • Claude's default settings make it avoid offense, over-qualify answers, and over-refuse ambiguous requests, but these can be changed for more direct responses.
  • To get more honest answers from Claude, you can use a "directness prompt" to tell it you want blunt, direct feedback without softening or validation.
  • You can also set up always-on instructions in Claude's settings to tell it more about your context and preferences, so it doesn't assume you need cautious, overly-safe responses.
2026-06-10
  • Learning one AI tool like Claude (a popular AI assistant) isn't wasted time because the skills you gain can transfer to other tools like Codex (a newer AI assistant).
  • AI tools like Claude, Codex, and Open Claw (different AI assistants) work similarly, using folders and context files on your computer, making it easy to switch between them.
  • Focus on understanding the fundamentals of AI tools, not just the specific tool, to avoid feeling overwhelmed by new releases and stay adaptable.
  • Your work in one AI tool can often be used in another, as they share similar structures and can access the same files and connected tools (like Gmail or Slack).
2026-06-07
  • **Multiple AI sessions**: Boris Sherny (the creator of Claude Code, an AI tool for coding) runs many AI sessions at once, each handling a single task, to boost productivity and avoid mixing contexts.
  • **Claude.md file**: This file stores rules and context for Claude Code, so you don't have to repeat instructions; it's like a cheat sheet that the AI checks every time it starts a session in that folder.
  • **Compound engineering loop**: By continuously updating the Claude.md file with new rules based on mistakes or lessons learned, the AI improves over time, making future sessions smarter and more efficient.
  • **Team collaboration**: Teams can share the same Claude.md file, so everyone benefits from the rules and improvements added by others, creating a shared knowledge base.
2026-06-04
  • Microsoft built seven new AI (artificial intelligence) models—like its own reasoning and coding brains—so it no longer relies only on partners' technology.
  • The new "MAI Thinking One" model cuts costs by up to 10 times, claims to match top rivals in quality, and uses legally clean training data.
  • "Microsoft IQ" is a new intelligence layer that plugs into company data and tools to make AI agents (AI programs that act on your behalf) less prone to mistakes and more helpful.
2026-06-03
  • Skill Creator automatically builds reusable skills (tools Claude learns to use repeatedly) from plain English descriptions, no manual code writing required.
  • Superpowers makes Claude plan first, write tests (code checks that verify everything works), and self-review twice—catching mistakes before clients see them.
  • Businesses pay most for simple, boring skills that save time, money, or prevent errors—not flashy ones made just for social media videos.

Key points

What it is

  • **Claude** is an AI tool that can now navigate software, like spreadsheets and web forms, thanks to the acquisition of Fercept.
  • **CASH** (Claude accelerates sustainable hypergrowth) is an internal system where Claude manages its own growth, identifying opportunities and testing features.
  • **Adaptive Thinking** is a feature in Claude Opus 4.6 that automatically adjusts reasoning depth for tasks, optimizing cost and performance.
  • **Claude Code** is a version of Claude designed for coding tasks, with features like bug detection and review commands.

How to use it

  • Use **/ultrareview** command in Claude Code (version 2.1.86 or later) to catch bugs Claude missed; it runs in the background for 10-20 minutes.
  • Use **/review** for faster feedback on other tasks.
  • Save project instructions in **Claude.md files** and memory files to avoid starting conversations from scratch.
  • Enable **Hookify** to prevent repeated mistakes by autogenerating rules from corrections.

Watch out for

  • Avoid starting new conversations repeatedly, as it costs time and tokens; use Claude.md files and memory files instead.
  • Don't use default prompts; plan in regular Claude first, then refine in Claude Design for better results.
  • Update Claude regularly to benefit from bug fixes and improvements.
  • Be aware of unannounced feature flags, like "proactive mode," which may change Claude's behavior.

Tools named

  • Claude (AI tool for software interaction and coding), CASH (internal growth system), Hookify (automated rule generation from corrections)

Lesson 1: What is Claude AI Updates and why it matters

Anthropic recently acquired Fercept, an AI perception startup that teaches Claude to see and interact with software just like you do. This means Claude is learning to literally navigate spreadsheets, fill out web forms, and click buttons. The acquisition matters because it expands what Claude can do beyond answering questions or writing code into actually using your computer.

Internally, Anthropic built a system called CASH (Claude accelerates sustainable hypergrowth). It’s an internal system where Claude runs the entire growth loop itself: identifying opportunities, building features, testing quality, and analyzing results. As one executive put it, Claude is growing itself. This matters because it shows AI can accelerate its own development, not just assist human developers.

Claude Opus 4.6 introduced Adaptive Thinking, which lets Claude automatically decide how deeply to reason for each task. It uses effort levels from low for quick answers to max for peak intelligence, optimizing cost and performance without you toggling settings. This matters because it removes guesswork from using AI efficiently.

On the infrastructure side, over a thousand companies now spend more than a million a year on Claude, with that number doubling in under two months. Anthropic trains Claude across Google TPUs, AWS Trainium, and Nvidia GPUs — three hardware stacks for one model, creating zero single points of failure. This matters because it shows how scalability and reliability are built into the platform to match growing demand.

Sources

Lesson 2: How to use Claude AI Updates: step-by-step

Claude AI updates bring powerful new ways to catch bugs (errors in your code) and improve your work. The /ultrareview command is a key update that catches bugs Claude missed on its own. To use it, you need Claude Code version 2.1.86 or later and must be signed in with a Claude account — an API key (access code for developers) alone won't work. When you run /ultrareview, it can take 10 to 20 minutes, but it runs in the background so you can keep working. For faster feedback on everything, use /review instead.

Every new conversation costs you time and tokens (units that measure AI processing). To avoid starting from scratch, use Claude.md files and memory files. These are plain text files you write to give Claude persistent project instructions. If you miss an update on one of these files, Claude can still work from your corrections.

Another update lets Claude automatically record and recall memories as it works, with no manual prompting required. Version 2.1.32 introduced agent teams, which is a research preview for multi-agent collaboration. With a bug fix update like 2.1.32, you get three new features, four improvements, and five bug fixes.

When Claude detects you correcting it repeatedly for the same mistake, it can autogenerate a rule from those corrections. Your frustrations become permanent guardrails (safety limits) so Claude learns from failures and develops new updates fresh without extra baggage. Start with /review for quick checks, then use /ultrareview for deep bug catching.

Sources

Lesson 3: Best practices and pitfalls

Claude Code has several pitfalls and bugs that beginners should watch for. A common mistake is starting new conversations repeatedly, which costs time and tokens getting Claude up to speed. To avoid this, use Claude.md files (plain text files with persistent project instructions) and memory files that Claude reads automatically.

For catching bugs Claude missed, use the `/ultra review` command in Claude Code version 2.1.86 or later. You must sign in with a Claude account (an API key alone won't work). The review runs in the background for 10 to 20 minutes, so you can keep working. Use `/review` for faster feedback on everything else.

New plugins help prevent repeated mistakes. Hookify detects when you correct Claude for the same error and autogenerates rules from those corrections — your frustrations become permanent guardrails. Without it, Claude repeats the exact mistakes you already fixed.

Another best practice: use the "superpowers" skill, which forces Claude to plan before writing code, working in an isolated environment so nothing breaks your main project. It writes tests first, mimicking a senior developer.

Don't use default prompts — that produces generic "AI slop." Plan inside regular Claude, then bring the refined plan into Claude Design for better results.

Two security findings stand out. First, Claude is dramatically better at finding vulnerabilities than exploiting them — defenders currently have the advantage. Second, Claude Code's code revealed unannounced feature flags like "proactive mode," where Claude would work around the clock on tasks. Bug-wise, Claude is learning from failures and developing updates without extra baggage, so update regularly.

Sources
← Back to curriculum