Hands-on AI Weekly - August 17, 2025
Strategic AI Intelligence for High-Performance Leaders
Graymatter: Hands-on AI, actionable strategy, and self-leadership insights to innovate your career and business.
Every week, I share the AI tools and techniques I'm actually using as an AI entrepreneur and instructor—the same ones driving measurable results in my Hands-on AI for Leaders cohort. You'll get three things: the week's most impactful upgrades, why they create competitive advantage, and a 15-minute action plan to implement each one immediately.
This Week's Intelligence
Perplexity Comet rolled out to US Pro users on August 14, and to Enterprise Pro on August 15. This isn't just another browser—it's positioning itself as your AI work console for multi-step tasks.
Gemini gains personalization from past conversations and temporary chats. Smart privacy controls are rolling out on 2.5 Pro in select regions first.
Claude introduces a learning style that coaches rather than answers, plus the ability to reference your past conversations on demand. Sonnet 4 now handles one million tokens in public beta.
Claude Code adds output styles—Explanatory and Learning modes that help developers understand decision-making processes and build skills through guided practice with TODO markers.
ChatGPT 5 fine-tuned its default tone after user feedback and restored 4o access. New routing modes (Auto, Fast, Thinking) give you tactical control over speed versus depth. Plus users now see higher weekly thinking limits near 3,000 messages.
OpenAI published comprehensive GPT-5 prompting standards and a Playground optimizer on August 7—the closest thing we have to "prompt engineering best practices" from the source. The guide introduces new API controls like reasoning_effort and verbosity settings.
The Strategic Move That Changes Everything
Anthropic's one million token context window for Sonnet 4 isn't just bigger—it's fundamentally different. Available now via API and Bedrock, coming soon to Vertex AI.
Why this matters strategically: Your team can now ingest entire codebases, policy libraries, or research archives in a single pass. This eliminates the brittle chunking strategies that break complex analysis. Think comprehensive audits, root cause analysis across systems, and discovery work that previously required weeks of document review.
The competitive advantage? While others are still piecing together fragmented insights, you're seeing the whole picture.
Tools That Deliver Immediate Impact
Perplexity Comet: Your AI Work Console
What it is: Perplexity's AI browser that handles multi-step workflows across your core tools.
The opportunity: Agentic browsing can triage your email backlog, prep your calendar briefings, and execute research workflows. This reduces the "swivel chair time" that kills productivity across operations, sales, and support.
Watch this: How deeply it integrates with Gmail and Calendar, what admin controls emerge, and the pace of global rollout.
Try this (15 minutes): Install Comet, connect your Gmail and Calendar, then prompt: "Brief me on tomorrow's priorities and draft replies to my three highest-impact email threads."
Learn more: Perplexity Enterprise Pro
Google Gemini: Personalization With Privacy Controls
What's new: Gemini uses your chat history to personalize responses while offering temporary chats that don't save anything.
Why it matters: You get the continuity of personalized AI without sacrificing control. Personal context becomes a strategic toggle, and temporary chats protect sensitive conversations.
The test: Whether personal context actually improves task completion rates—personalization without performance gains is just noise.
Try this (15 minutes): Since personal context is on by default, ask: "Summarize our last five conversations and propose specific next steps for my Q4 objectives." To check your settings, go to Settings → Personal context → Your past chats with Gemini. Follow up with a temporary chat for any sensitive planning.
Claude: Memory On Demand + Socratic Learning
The memory upgrade: Claude can search and reference your prior conversations when you explicitly ask—no automatic logging, stronger privacy defaults.
The learning mode: A new style option that shifts Claude into Socratic coaching, asking questions and guiding your thinking rather than delivering answers.
Strategic value: Continuity across multi-week projects without repeating context. The learning mode is particularly powerful for onboarding, training scenarios, and developing judgment rather than just getting answers.
Try this (15 minutes): Start a new chat and say: "Find our strategy discussion from last week and summarize decisions, owners, and open risks." Then switch to learning style and prompt: "Coach me through building Q4 OKRs for revenue growth. Ask probing questions and wait for my reasoning before suggesting next steps."
Learn more: Anthropic Memory Feature | Engadget Learning Mode
OpenAI GPT-5: Professional Prompting Standards
What dropped: Comprehensive prompting guide plus a Playground optimizer that shows measurable improvements in runtime, memory usage, and task adherence.
Why this is strategic: Common standards reduce the ambiguity that tanks AI performance. The optimizer shows concrete gains—faster execution, lower resource consumption, higher accuracy on coding and analysis tasks.
Key controls: reasoning_effort (minimal for simple tasks, medium/high for analysis) and verbosity (default low, override to high only for detailed explanations).
Try this (15 minutes): Paste a production prompt into the OpenAI optimizer, review the changes, then run a quick A/B test across ten tasks. Track runtime, memory, and accuracy. Keep what performs better.
Learn more: GPT-5 Prompting Guide | Prompt Optimization
What I'm Processing
Agentic browsers as work consoles: Comet's integration depth hints at a future where your browser executes complex workflows, not just displays pages. This is operating system-level competition emerging.
Memory with guardrails: Anthropic's on-demand approach provides a privacy-forward alternative to automatic memory systems. Adoption in regulated industries will be the real test.
Prompt management as IP: Your accumulated prompt libraries are becoming intellectual property and competitive moats. While others experiment randomly, systematic prompt management turns AI use into strategic advantage.
Level Up Your Prompt Game
Live Lightning Session: I'm hosting a 30-minute demonstration on Friday, August 22, at 12 noon US Eastern, showing how to manage your prompts and agent instructions like strategic assets.
Here's the reality: Your prompts are competitive moats. While competitors burn time on random queries, systematic prompt management turns ad hoc AI use into productivity multipliers that save hours weekly, ensure consistent quality, and scale across teams.
This turns your accumulated AI expertise into a durable business advantage.
Your Turn
What AI techniques accelerated your impact this week? Which tools moved from experiment to essential workflow?
Reply and share your wins—I read every response and often feature the best insights in future editions.
—James