The AI OS · Letter #37
May 31, 2025

Don't Learn Gemini. Use It.

The 90% miss Gmail switch, model selection science, and how to game-ify any PDF in 60 seconds.

Don't Learn Gemini. Use It.

Last Tuesday, I watched my startup founder friend Marco stare at his laptop for 20 minutes, paralyzed by a simple decision: which AI tool should handle his investor deck analysis.

"I've got ChatGPT Plus, Claude Pro, and now everyone's talking about Gemini," he muttered. "Do I really need another subscription?"

I pulled up Gemini, uploaded his 47-page deck, and asked it to cross-reference three competitor analyses from his Drive folder—all simultaneously. Thirty seconds later, we had a synthesis that would've taken him hours to compile manually.

"Holy s***," Marco said. "It already knows my entire Google Workspace."

That's the thing about Google's AI blitz. While everyone debates which chatbot is "smarter," Google built something fundamentally different: AI that lives where you already work.

Since Gemini 2.5 Pro dropped and claimed the #1 spot, surpassing GPT-4o/Claude-3.5 on key benchmarks, I've been stress-testing it in real workflows. Here's what actually works—and what definitely doesn't.

Google's Absurd 30-Day AI Sprint

No one's shipping faster than Google these days. While OpenAI debates strawberry flavors and Anthropic perfects reasoning, Google dropped 24 AI tools in 30 days:

The Workspace Revolution:

The Creative Suite:

The Developer Tools:

The Infrastructure:

The Specialized Models:

Bookmark this list. You'll sound brilliant in your next team meeting. But here's what matters: The Gemini app now has over 400 million monthly active users and the best of Google AI is now included in Workspace Business & Enterprise plans, giving you AI features like Gemini and NotebookLM Plus with no add-on needed.

Translation: This isn't experimental anymore. It's infrastructure.

The Setup 90% Miss (Start Here or Stay Disappointed)

Most people jump into Gemini and think it's "meh" because they skip the most critical configuration step.

You don't start in the Gemini app.

You start in Gmail.

Here's the magic sequence:

Step 1: Unlock the Workspace Integration

  1. Open Gmail → Settings → General Tab
  2. Search for "smart features"
  3. Enable both:
    • Smart Features
    • Google Workspace Smart Features
  4. Hit Save

This single setting transforms Gemini from a basic chatbot into a workspace-aware AI assistant that can access your emails, docs, drive files, and calendar context.

Step 2: Configure Your Gemini Apps

  1. Head to gemini.google.com
  2. Click Settings → Apps
  3. Turn OFF everything you don't use
  4. Keep ONLY:
    • Google Workspace (essential)
    • YouTube (if you have Premium)

Why the selective approach? You want intentional triggers, not accidental noise from apps you never touch.

The Result: Now when you ask Gemini to "summarize this week's project emails from Sarah," it actually knows what you're talking about.

The Great Model Selection Science

Here's what nobody tells you about choosing between Gemini Flash and Pro: It's not about which is "better"—it's about matching cognitive load to task complexity.

Recent benchmarks tell the story: GPT-4o demonstrated elite-level performance, achieving scores above 85 on multiple benchmarks like MMLU, HumanEval, and MGSM. However, no single model dominated across every task. Meanwhile, Gemini 2.5 Pro shines in technical accuracy but struggles with audience empathy. Claude 4 Sonnet offered a balance of creativity, practicality.

Forget about "research papers" benchmarks.

It always boils down to one question: chat model (LLM) vs. reasoning model.

Don't Learn Gemini. Use It. — Charafeddine Mouzouni | Cohorte

The Simple Rule:

Task Type Model Choice Why This Works
Routine & Repetitive Flash (Chat Model) Fast execution, consistent output, direct answers
Strategic & Complex Pro (Reasoning Model) Deep analysis, multi-step logic, nuanced decisions

Flash Use Cases (Speed Wins):

Pro Use Cases (Depth Wins):

Pro Tip: Those claiming "always use reasoning models for everything" clearly haven't optimized for speed. Using Pro for simple tasks is like hiring a consultant to make coffee—you'll wait 40 seconds for an over-engineered answer.

Quick Reference Across Platforms:

Platform Chat Model Reasoning Model
Gemini Flash Pro
ChatGPT GPT-4o, 4* o*
DeepSeek V-* R-*

Real-World Test: Can Gemini Run My Entire Project Loop?

(Spoiler: Halfway)

I wanted to test Gemini's workspace integration with a common workflow: extract → analyze → distribute.

The Challenge: Transform scattered email threads into actionable project management without manual copy-paste hell.

I want to make this guide useful for everyone - whether you're using Gemini Premium or the free version.

So I'll focus on the features that are available now for everyone — no premium subscription needed.

Note: I've already built this workflow in my Claude desktop app using MCP integrations (if you're curious about that setup, check my previous newsletters).

Step 1: Information ExtractionI prompted: "Extract all actions, tasks, and owners from the last email from Alessandro. Put everything in a table."

Using @Gmail integration, Gemini pulled the email context and delivered a clean table with:

Bonus: There's an "Export to Sheets" button right in the interface.

Don't Learn Gemini. Use It. — Charafeddine Mouzouni | Cohorte

‍

Step 2: The Agentic Test (Where It Breaks)Then I tried: "Put all those items in a Google Sheet, name it 'actions_from_alessandro' and save it in my Drive folder called 'O_Project'"

Here's where Google's workspace integration hits a wall. Gemini excels at pulling information from your workspace but struggles with pushing actions back. It can't create files, update spreadsheets, or execute complex workspace actions.

Don't Learn Gemini. Use It. — Charafeddine Mouzouni | Cohorte

The Reality Check: Gemini Agents are brilliant at synthesis and analysis but not yet ready for full workflow automation. Recent coding benchmarks show mixed results—Claude takes the lead in most programming tasks, while Gemini 2.5 Pro shows strength in specific areas but still has execution gaps.

What Actually Works:

Bottom Line: Gemini transforms information work but doesn't replace workflow tools yet. Think "supercharged research assistant" rather than "AI employee."

Lab Test: Turn Any PDF Into an Interactive Game (This Is Ridiculous)

Here's where Gemini's 1-million-token context window creates magic.

I grabbed a dense 30-page OpenAI prompting guide and threw it at Gemini Pro with this challenge:

Don't Learn Gemini. Use It. — Charafeddine Mouzouni | Cohorte

Prompt:

Turn this paper into an interactive fun game designed to test my knowledge.

Requirements:

  1. Questions start easy, get progressively harder
  2. 10 questions total
  3. Explain wrong answers with examples
  4. Make it genuinely engaging with:
    • Achievement progression
    • Storyline elements
    • Interactive graphics
    • Professional polish

What Happened Next:

The Result: A publication-ready interactive learning tool in under 60 seconds.

Try This With:

This feature works across ChatGPT and Claude too, but Gemini 2.5 Pro takes the lead in single-shot execution—especially with longer documents thanks to that massive context window.

The Workspace Swiss Army Knife

The real superpower isn't Gemini as a chatbot—it's Gemini as ambient intelligence throughout your existing workflow.

With Smart Features enabled, the Gemini Side Panel becomes productivity infrastructure:

Email Mastery

Scenario: You get a vague conference invite. You're interested but need to negotiate timing.

Old Way: Spend 10 minutes crafting the perfect diplomatic response.

New Way:

  1. Open Gemini Side Panel
  2. Paste your rough thoughts: "interested but timing might be tough"
  3. Get: "Thanks for thinking of me! I'd love to explore this opportunity—could we discuss a date that works around my current commitments?"

Professional. Warm. Perfect tone.

Spreadsheet Intelligence

Scenario: You're tracking workshop signups across multiple channels.

Instead of manual table building:

=AI("Create signup summary by channel with conversion rates")

Need sentiment analysis on feedback?

=AI("Categorize this feedback: Price, UX, Feature Request, Other", D2)

Translation needs?

=AI("Translate to business-appropriate Mandarin", D2)

Gemini fills the logic. You refine the results. Hours saved.

Document Collaboration

Writing a project retrospective?

Prompt: "We received this client feedback [paste]. Write a professional summary of our planned response."

Working with meeting notes?

Need cross-document insights?

The Integration Advantage

Gemini in the side panel of Gmail, Docs, Sheets, Slides, Drive, and Chat—Get AI assistance directly in the flow of your work represents a fundamentally different approach than standalone AI tools. Instead of context-switching between apps, intelligence flows through your existing workspace.

This is where Google's integration beats Microsoft Copilot (which feels clunky) and standalone tools like ChatGPT (which require constant copy-paste).

The Unfair Context Advantage

While everyone debates reasoning capabilities, Gemini 2.5 Pro quietly offers an unfair edge: 1 million tokens of context—roughly 700,000 English words.

The Competition:laude 4 Sonnet:

Translation: You can feed Gemini two Harry Potter books plus your annotations and get coherent analysis in a single conversation.

Real-World Applications:

Example Prompt:"Here are three competitor analysis reports [upload], our Q4 performance data [upload], and customer interview transcripts [upload]. What strategic opportunities do you see that others are missing?"

This isn't just "more context"—it's qualitatively different analysis that becomes possible when an AI can hold massive amounts of information simultaneously.

The One Significant Flaw

Gemini has an over-safety problem that creates real workflow friction.

Example: I asked about Jimmy Carter's presidency challenges for a historical analysis piece. Gemini refused, deeming presidential criticism "sensitive political content."

Meanwhile, ChatGPT and Claude provided balanced, factual analysis without drama.

The Pattern: Google's safety filters are tuned so conservatively they block legitimate research, historical analysis, and business strategy discussions.

Impact: You'll hit walls when exploring:

Workaround: Rephrase sensitive queries as "academic research" or "case study analysis." Sometimes it works.

But this represents a deeper issue: AI safety shouldn't feel like AI censorship. When safety measures create more friction than actual risks, adoption suffers.

The Strategic Verdict

After stress-testing Gemini across real workflows, here's the honest assessment:

What Google Got Right:

What's Still Missing:

The Bottom Line: Gemini isn't the smartest AI, but it's becoming the most integrated AI. In productivity workflows, integration often beats raw intelligence.

My Recommendation: Use Gemini for workspace tasks, document analysis, and routine automation. Keep Claude or ChatGPT for creative work, complex reasoning, and sensitive research.

The future isn't about finding the "best" AI—it's about orchestrating the right AI for each context.

What's Next

Google's AI integration represents a inflection point: AI transitioning from "tool you visit" to "intelligence that lives in your workflow."

The best of Google AI is now included in Workspace Business & Enterprise plans, making this transformation accessible to millions of knowledge workers without additional subscriptions.

Try these workflows this week:

  1. Enable Smart Features in Gmail (if you haven't already)
  2. Test the PDF-to-game conversion with your team's training materials
  3. Use @AI formulas in Sheets for your next data analysis
  4. Upload 3+ documents to Gemini and ask for strategic synthesis

What I'm Watching:

Your Feedback Makes This Better: Email me and tell me which workflows you've tested, what broke, and what blew your mind. Every response makes the next issue sharper.

The AI landscape shifts weekly. The builders who win are those who use rather than just learn.

Until next time,

— Charafeddine

P.S. If this breakdown saved you hours of testing, forward it to someone still trying to figure out which AI tool to use. They'll thank you.

‍

New letters now publish on charafeddine.co

Read the latest letters