Hands-on AI Weekly Summary - August 10, 2025
From Headlines to Hands-On: The AI Intel You Can Use Now
August 10, 2025 · Est. reading time: 10 min
Graymatter: Hands-on AI, actionable strategy, and self-leadership insights to innovate your career and business.
Welcome to the First Weekly AI Summary
Each week, I cut through the AI noise to deliver the most important product launches, breakthroughs, and market shifts — and, more importantly, what they mean for your strategy, team, and career. You’ll get the key developments, strategic context, and a hands-on way to try each capability yourself, saving hours of research and moving faster from insight to impact.
TL;DR
OpenAI drops GPT-5 — Unified reasoning boosts coding performance +75%, cuts hallucinations by 80%
OpenAI releases GPT-OSS — First open models since GPT-2; 20B runs locally, 120B rivals o4-mini
Anthropic launches Claude Opus 4.1 — New coding leader at 74.5% SWE-bench Verified
Claude Code Subagents — Specialized AI teams for anyone, no code required
Claude adds MCP connectors — Expands real-world workflow execution capabilities
Google’s Gemini Deep Think — Parallel multi-agent reasoning for complex problems
ChatGPT Study Mode — Built-in learning companion for deep topic mastery
Google NotebookLM — Now creates video content from your research sources
Mo Gawdat forecasts AI disruption — 15 years of upheaval, white-collar job collapse, middle class at risk; winners adapt fastest
OpenAI: GPT-5 Unifies Speed + Reasoning → 700M Weekly Users
TL;DR: Combines fast responses with extended reasoning; now on free tier.
Why it matters: No more model-switch confusion; 45% fewer factual errors than GPT-4o. Enterprise-grade reliability for 5M paid business users.
What to watch: API pricing at $15/$75 per million tokens; enterprise adoption curve.
Hands-on AI: In ChatGPT, use “vibe coding” — describe an app idea and watch GPT-5 build it end-to-end.
Learn more:
My initial evaluation is that GPT-5 is faster, smarter, but I was surprised it did not correctly generate the OpenAI Responses API code. More to come!
OpenAI: GPT-OSS Open Models → First Release Since GPT-2 in 2019
TL;DR: 120B and 20B parameter models under Apache 2.0 license; 20B runs on 16GB RAM laptops.
Why it matters: Strategic counter to Chinese open models; broadens frontier reasoning access while protecting GPT-5 IP.
What to watch: Developer uptake on Hugging Face, Azure, Databricks. Pressure on Meta’s Llama and Mistral.
Hands-on AI: Deploy
gpt-oss-20blocally via Hugging Face Transformers and test reasoning tasks offline.Learn more:
I will be following up with a detailed article on how to run GPT-OSS models on laptops.
Anthropic: Claude Opus 4.1 Surpasses GPT-5 in Coding Benchmarks
TL;DR: Scores 74.5% on SWE-bench Verified; excels at multi-file refactoring without new bugs.
Why it matters: Ideal for long-running agent workflows; raises competition in enterprise coding assistants.
What to watch: Developer migration between Anthropic and OpenAI ecosystems; enterprise pricing parity.
Hands-on AI: In Claude.ai, upload multiple files and request architectural refactoring suggestions.
Learn more: Claude Opus 4.1 announcement
Anthropic: Claude Code Subagents → Specialized AI Teams for Everyone
TL;DR: Spawn task-specific AI assistants from a simple description. No coding required.
Why it matters: Breaks down barriers for non-technical teams; turns ideas into operational AI agents.
What to watch: Expansion into legal, marketing, and ops workflows.
Hands-on AI: Use
/agentsin Claude Code to build a “business analyst” subagent that proactively analyzes data and creates visual dashboards.Learn more:
Build & Activate Agents in Natural Language with Claude Code - on Friday, I hosted a live, 30-minute Maven Lightning Lesson on how to get started creating and running agents. Check out the video, deck, and whiteboard to get up and running quickly.
Anthropic: MCP Connectors Expand Claude’s Workflow Execution
TL;DR: Anthropic continues to add Model Context Protocol (MCP) connectors, enabling Claude to execute workflows across more tools and data sources.
Why it matters: MCP connectors make Claude more useful in real business settings by integrating with critical apps, databases, and APIs — reducing the gap between AI reasoning and actual task execution.
What to watch: Growth in connector library; enterprise adoption in ops, analytics, and customer-facing workflows.
Hands-on AI: Explore the official Anthropic MCP documentation and try setting up a connector that links Claude to your applications.
Learn more: Getting started with connectors
Google: Gemini Deep Think → Multi-Agent Reasoning for Complex Problems
TL;DR: Runs parallel reasoning processes; wins gold at International Math Olympiad.
Why it matters: Competes on creativity and strategic problem solving, not just speed.
What to watch: Adoption at $250/month; how it stacks up to OpenAI o3 and Anthropic’s long-context modes.
Hands-on AI: Use Gemini Ultra’s Deep Think to map multi-step strategic initiatives for your business.
Learn more: Google Gemini Deep Think announcement
OpenAI: ChatGPT Study Mode → AI-Powered Learning Companion
TL;DR: New Study Mode in ChatGPT lets users dive deep into any subject with structured lessons, quizzes, and progress tracking.
Why it matters: Transforms ChatGPT from a Q&A assistant into a persistent tutor — ideal for leaders building AI fluency or professionals learning technical skills.
What to watch: Adoption in corporate training and professional development; integration with enterprise knowledge bases.
Hands-on AI: Activate Study Mode in ChatGPT (Settings → Features) and create a custom learning path on AI strategy, tools, or coding.
Learn more: ChatGPT Study Mode documentation
Google: NotebookLM Now Creates Video Content
TL;DR: Google’s NotebookLM can now turn your research sources into AI-generated video explainers, complete with narration and visuals.
Why it matters: Moves NotebookLM beyond static summaries, enabling rich multimedia learning materials for education, marketing, and internal training.
What to watch: Adoption by educators, content marketers, and corporate trainers; integration with Google Workspace and YouTube.
Hands-on AI: Import a set of documents into NotebookLM, choose the “Create Video” option, and generate a narrated explainer video on your topic.
Learn more: Google NotebookLM updates
Graymatter Insight: The Contrarian Take
Benchmarks Are a Shell Game. The Real Moat Is Workflow and Tools Integration.
This week was dominated by benchmark wars—GPT-5 vs. Claude 4.1 on coding, Gemini on math. While impressive, these scores are becoming a distraction. Winning a benchmark proves a model can solve a sterile, academic problem. It doesn’t prove it can solve a messy, real-world business problem.
The real competitive edge isn't benchmark performance; it's deep, reliable workflow integration with software tools we use daily. Anthropic's quiet expansion of MCP Connectors is arguably more significant than its SWE-bench score. An AI that is 5% less "intelligent" but can reliably execute tasks across Salesforce, Zendesk, and internal databases is infinitely more valuable than a genius AI that lives in a chat window.
I am loving how easy it is to connect apps in Claude and turn individual tools on or off. I've found that I am using it more often to run repeatable workflows.
What to watch: Pay less attention to benchmark leaderboards and more to the size and quality of each platform's connector/tool library. The winner will be the platform that gets work done, not the one that scores highest on a test.
Hands-on AI Learning Resources 🙌
Claude Code: A Highly Agentic Coding Assistant — Free DeepLearning.AI course to master Claude Code, perfectly timed for the recent Subagents launch.
Hands-on AI Content Catalog - Based on feedback from Hands-on AI for Leaders course alumni, I published a searchable catalog to improve access to course resources, including lessons, assignments, and videos. The catalog also includes access to free resources. Browse the catalog here.
What I’m Watching
Former Google X exec Mo Gawdat delivers a blunt forecast: AI’s next wave will upend white-collar work, gut the middle class, and reshape society before any utopian promise arrives. The winners? Those who adapt fast.
15 Years of Turbulence Ahead (2027–2042): Mo Gawdat, former Google X exec, warns of an unavoidable “hell” period as AI transforms work and society before any utopian benefits emerge.
White-Collar Jobs at Risk: AGI will outperform humans in nearly all professional roles—including developers, executives, podcasters, and CEOs.
Middle Class Collapse: Without intervention, the middle class could vanish, leaving a tiny elite and a struggling majority.
Social & Mental Health Fallout: Mass unemployment could fuel unrest, loneliness, and mental health crises.
A Path to Utopia—If We Act: Ethical AI governance, universal basic income, and human-centered policies could create a future of equality, leisure, and fulfillment.
Graymatter Insight: Disruption this fast doesn’t reward “wait and see.” It rewards leaders who move first—those willing to rewire their skills, workflows, and business models before the old rules collapse. The future won’t be won by the strongest or the smartest, but by the fastest to adapt.
Recap
Since this is the first edition of the Hands-on AI Weekly Summary, I’d love your feedback. Did this format help you quickly understand the most important AI news and how to apply it? What would make it more valuable for your AI learning and leadership journey? Reply and let me know — your input will shape future editions.