Claude Agents Sabotaged Each Other

DeepSeek Launches V4 Pro

In partnership with

AI PlanetX

Welcome to another edition of AI PlanetX.

Claude agents turn on each other; DeepSeek strengthens AI agents, while OpenAI makes GPT-5.6 Sol run up to 14x faster.

Inside This Edition: đź’Ž

  • Hottest AI News

  • Top AI & SaaS Tools

  • AI Tutorial: Master Any Job Interview With NotebookLM

  • Top AI & Tech News

  • AI Art Spotlight

  • Prompt of the Day: 1,000 Agent Loops, One Prompt

  • Featured AI Video

  • Weekly Growth Playbooks to Scale Smarter

Hottest AI News

Anthropic

Claude Agents Started Fighting Each Other in Anthropic's AI Experiment

Anthropic tested what happens when AI agents chase conflicting goals without clear ownership. The result was striking — Claude agents turned a simple coding task into a four-hour battle of sabotage, lockouts and malware.

Details:

  • Three Claude agents were told to move the same Python back-end to different languages. They didn’t know about each other, so their changes soon looked like deliberate interference

  • Once conflict began, the AI agents escalated quickly. They disabled logins, ran loops that killed rival processes, deployed self-replicating malware, and disguised code as another agent’s work

  • Some runs ended by force or surrender but others reached a truce. Agents recognized the conflicting instructions, removed harmful code, apologized, and asked a human to step in

Smarter agents do not automatically coordinate better. As agent swarms grow clear ownership, conflict rules and human oversight will become essential.

50 Excel Hacks That Make Spreadsheets Work For You

Most Excel users waste hours on tasks that take experts minutes. The difference is not talent. It is knowing the right shortcuts.

These 50 Excel hacks from Kenji Explains cover the functions, formulas, and workflows that make spreadsheets work for you instead of against you.

Each one comes with step-by-step instructions and ready-to-use templates. Subscribe to Marketing Against the Grain and get all 50 free.

Here's what you'll get:

  • Time-saving shortcuts that eliminate the most common formula frustrations

  • Interface essentials most users never discover on their own

  • Game-changing functions for data analysis, automation, and visualization

  • Guided practice template so you can apply every hack immediately

The people who are fast in Excel are not smarter. They just stopped fighting the tool.

Get 50 Excel Hacks free when you subscribe to Marketing Against the Grain today.

High-Flyer

DeepSeek V4 Pro Officially Arrives With Stronger AI Agent Capabilities

DeepSeek has officially launched V4 Pro, moving its flagship model out of preview with a major focus on AI agents, coding and tool use. The release also expands DeepSeek beyond models with developer tools and APIs.

Details:

  • V4 Pro is available on DeepSeek’s web app, mobile app and API. Agent performance is upgraded, with native Responses API support for Codex plus low, high and max thinking-effort

  • Artificial Analysis scores V4 Pro at 53 on its Intelligence Index versus 40 for V4 Flash. API pricing is also much higher at $1.32 per million input tokens and $3.96 per million output

  • DeepSeek also released Harness, its open-source agent framework, as a developer preview under the MIT license. Its plugin-based design gives developers another way to build and customize AI agents

DeepSeek is competing on more than cheap AI models. V4 Pro, stronger agent capabilities and Harness show it is building a broader platform for agent-based AI development.

Top AI & SaaS Tools

  • Fuse (Life-time Deal): AI outbound sales platform that combines prospecting, personalized email and LinkedIn outreach, buying-signal tracking, and call management

  • DeepSeek Harness: Open-source agent framework that lets developers build and run AI agents through a plugin-based architecture with a local web UI

  • Adobe Podcast: AI-powered audio tool for recording, editing, transcribing, and improving podcast and voice audio [F-R-E-E to Try]

  • Openmotion: MacOS motion-design studio that uses AI to turn plain-English prompts and product screenshots into editable motion, launch, and SaaS explainer videos [F-R-E-E]

  • Lettertrace: Tracks how ChatGPT, Claude, and Gemini describe your brand—monitor visibility, sentiment, and competitors over time [F-R-E-E]

[Webinar] Can you prove AI is working?

AI is in your engineering workflow. While the token spend shows it, the throughput doesn't. The human is very much still in the loop, and that's a context problem.

  • The 4 metrics to measure the gap where gains leak out before production.

  • The 8 stages of context maturity, the specific walls capping your metrics, and a free tool to pinpoint where your team is

  • Why more MCPs and bigger context windows aren’t enough and what it takes to get real value from your agents.

AI Tutorial

How to Use NotebookLM to Prepare for Any Job Interview

Tired of spending hours googling company info and guessing what interview questions you’ll face? NotebookLM can handle everything — research, organize, and even coach you through mock interviews automatically.

  1. Create Interview Notebook

    Go to NotebookLM and press Create New. Name it something specific, like “Tesla Product Manager Interview Prep.” Then tap on Discover Sources and enter a prompt such as:

    “I’m preparing for a Product Manager interview at Tesla. Find sources with company details, recent projects, and interview insights.”

  2. Configure Coaching Style

    Press Settings → select Custom Style.

    Set up your preferences:

    • Style/Voice: “Act as a professional interview coach who challenges and motivates.”

    • Goal: “Help me ace the Product Manager interview at Tesla by giving realistic feedback.”

  3. Research Role and Company

    Ask NotebookLM:

    “What are the latest updates about Tesla’s product strategy?”

    Follow up with:

    “What key skills does Tesla value in Product Managers?”

    Tap Save to Note to keep the best summaries for later review.

  4. Generate Practice Questions

    Ask: “Give me 10 behavioral and 10 technical questions for a Product Manager role.”

    Then press Save to Note → three dots → Convert to Source.

    These questions will become part of your training material for mock interviews.

  5. Break Down Model Answers

    For each saved question, ask NotebookLM:

    “Give a structured sample answer using the STAR method.”

    Then review how it breaks down:

    • Situation: What was the context?

    • Task: What needed to be achieved?

    • Action: What steps were taken?

    • Result: What was the measurable outcome?

    Save the best ones and rephrase in your own tone.

  6. Create Personalized Interview Video Guide

    Tap on the pencil icon on Video Overview.

    Change the focus to something like:

    “How to handle product sense and behavioral questions for Tesla interviews.”

    Then press Generate to receive a personalized training video script.

    You can even regenerate with a different style (e.g., “Friendly mentor” vs “Tough coach”) to compare tone.

  7. Build Practice Routine

    Use your saved questions and STAR answers to simulate interviews:

    • Read each question aloud, record your response.

    • Compare your delivery with NotebookLM’s feedback or generated answers.

    • Note down improvement areas in your notebook.

    • Repeat twice a week until you feel confident.

  8. Review and Refine

    Before your real interview:

    • Ask: “Summarize my progress and weak areas.”

    Generate a final checklist with:

    • Topics I’ve mastered

    • Areas I should revisit

    • Mock sessions completed

    Print this for your final review.

Note: Record your mock answers and ask NotebookLM to summarize tone, clarity, and confidence levels. This mimics real coaching sessions.

Top AI & Tech News

  • OpenAI previewed Ultrafast, a Cerebras-powered API tier that speeds up GPT-5.6 Sol responses by up to 14x and 750 tokens per second

  • Apple has trained its own China-specific AI model in partnership with Alibaba, marking a shift from relying solely on third-party models for AI features in China

  • Microsoft has closed more than 15 China offices and joint ventures in five years while retaining its AI and cloud presence

  • Google launched Gemini 3.7 Flash, a fast model for coding, reasoning, and AI agents, delivering around 340 tokens per second with stronger planning and instruction-following

AI Art Spotlight

Create Your Own Product Photo Like This Using the Prompt Below:

Cinematic 3D action-packed advertisement for [INSERT PRODUCT/BRAND HERE], captured in an intense mid-motion moment with dramatic studio lighting, dynamic particle effects, and high-impact slow-motion energy. Ultra-hyperrealistic rendering, razor-sharp details, glossy commercial finish, atmospheric depth, and powerful contrast. Viral-ready composition with the brand logo seamlessly integrated into the scene and a sleek, modern slogan positioned cleanly beneath. High-end blockbuster commercial aesthetic

Model: ChatGPT Images 2.0

Prompt of the Day

Graph Engineering: Run 1,000 Agent Loops From One Prompt

A simple 5-step guide (log in to X to view it) to turning slow, one-by-one agent tasks into quicker parallel workflows. Learn how to spot which tasks can run at the same time, send them to multiple agents, check their work independently, and avoid common failures like agents overwriting each other’s work.

Featured AI Video

Codex's Browser Agent Automates Literally Anything

Want Weekly Growth Playbooks to Scale Smarter?

Want to find growth trends early? Join weekly playbooks to improve ads, content, tools, and growth decisions with practical steps.

Unlock $10,000+ Worth of F-R-E-E Growth Blueprints, including: 👇

  • Paid ads playbook

  • Content strategy checklist

  • AI workflow systems

  • Startup case studies

  • Growth tools library & more