5 practical experiments

Five Things in AI You Should Learn Today

Each one comes with a 30-minute experiment you can run today

These aren't ideas—they're real tools people are using right now. This lab shows you exactly how to try each one yourself this week.

Explore
the lab
Explore
Capability 01
1 / 5
01
Capability

NotebookLM Slide Decks + Infographics

Turn any document into presentation-ready slides (or a one-page infographic) in minutes

Try NotebookLM→
Swipe to compare. Scroll inside each card for details.
Quick start
What you need
  • A NotebookLM account (free for consumer accounts, limited for Workspace/Education)
  • At least one source document uploaded to a notebook (PDF, Google Doc, Word file, or plain text)
  • Edit access to the notebook (not just chat-only access)
  • 2-5 minutes for generation time
⚡5 minutes to try this
Step 2
Upload your doc, select 'Infographic' or 'Slide Deck', then add your custom prompt:
Step 2
Prompt (copy / paste ready)
Create a deck for executives using a professional and bold style with a focus on strategic implications and key decisions
Step 3
You'll get

A strong first draft you can download/share right away

Notes
  • Upload your source document: Click '+ Add source' in the Sources panel on the right
  • In the Studio panel on the left, click 'Slide Deck' or 'Infographic'
  • For Slide Decks: Choose 'Detailed Deck' (more text) or 'Presenter Slides' (clean visuals with minimal text)
  • Select output language and length (short, default, or long)
  • Wait 2–5 minutes while it generates, then download as PDF or share a link
  • Infographic options: Choose detail level (Concise/Standard/Detailed) and orientation (Web/Square/Portrait/Landscape)
Context
Why this matters now
Expand
NotebookLM now lets you generate complete slide decks and infographics directly from your uploaded sources. Instead of hand-building a deck from scratch, you can start with a strong first draft, then polish it for stakeholders.
Before / After
Before 2025
Chat + audio summaries only
Now
Slide decks + infographics from your sources
⚠️
Field note

Common mistake: Slide Decks and Infographics are AI-generated and may contain visual or factual inaccuracies - always review before sharing externally

Review the output carefully. Use the 'Good content' or 'Bad content' feedback buttons to improve future generations. For critical presentations, fact-check key claims against your source documents.

Success
Success looks like…
  • Combine multiple source documents in one notebook for comprehensive decks that synthesize across materials
  • Use the pencil icon to customize before generating (tone, audience, and what to emphasize)
  • Generate multiple versions with different formats (Detailed vs Presenter) or lengths to find what works best
  • For infographics, experiment with different orientations based on where you'll share it (Web for social, Portrait for reports)
  • Use the feedback buttons ('Good content' / 'Bad content') to push outputs toward your preferred style over time
02
Capability

Claude Agent Skills

Teach Claude how your team works — then let it apply that playbook automatically

Try Claude→
Swipe to compare. Scroll inside each card for details.
Quick start
What you need
  • A Claude account (Skills are available on Pro/Max/Team/Enterprise)
  • Skills enabled in Claude settings (Team/Enterprise: an admin may need to enable org-wide)
  • One repeatable workflow you want to standardize (reports, reviews, brand voice, spreadsheet tasks)
  • 5–10 minutes to create + test your first Skill
⚡7 minutes to try this
Step 2
Copy/paste this prompt to have Claude create your first Skill:
Step 2
Prompt (copy / paste ready)
Help me create a Claude Agent Skill called "Weekly Product Update". Goal: When I ask for a weekly product update, Claude should follow our standard format every time. Format rules: - Keep it under 300 words - Use clear section headers - Highlight blockers in bold Sections (in this exact order): 1) What shipped this week 2) What’s in testing 3) What’s blocked and why 4) Next week’s priorities Process: 1) Ask me 2–4 quick questions you need (audience, tone, and any must-include details). 2) Then generate the Skill structure (including the SKILL.md instructions) so I can save it. 3) Include 1 short example output so I can sanity-check it.
Step 3
You'll get

A reusable Skill that Claude auto-activates whenever you ask for a weekly product update — so you stop re-explaining the structure every time

Notes
  • If you don’t see Skills yet: go to Settings → Features and enable Skills (Team/Enterprise: your admin may need to enable them org-wide).
  • Notes: You can also create Skills for things like a code review checklist, brand voice rules, meeting notes formats, customer support replies, sales emails, or any repeatable workflow you want Claude to follow the same way every time.
  • Test it in a fresh chat: ‘Create this week’s product update’ and confirm Claude follows the sections + style automatically.
Context
Why this matters now
Expand
Claude now supports Agent Skills: folders that bundle instructions, scripts, and reference files Claude can load only when a task needs them. Skills are composable (they stack), portable (work across Claude apps, Claude Code, and the API), and efficient (Claude only loads what’s relevant). You can even create your own with the built-in skill-creator workflow, then share them org-wide.
Before / After
Before 2025
Copy-paste prompt templates
Now
Auto-invoked, shareable workflow folders
⚠️
Field note

Common mistake: Making a Skill either too vague (generic advice) or too brittle (so specific it only works for one case)

Write Skills like a playbook: define the repeatable process (sections, checks, tone), add 1–2 examples, and leave room for the actual content to change each time. Keep files minimal so Claude stays fast.

Success
Success looks like…
  • Stack Skills for bigger workflows (e.g., ‘Brand Voice’ + ‘Weekly Product Update’)
  • Add a real example doc/template inside the Skill so outputs match your team’s style
  • Use org-wide management to share Skills across the team (instead of everyone recreating them)
  • Browse partner-built Skills in the directory, then adapt them to your process
  • Version your Skills: when the workflow changes, create a new version and note what changed
03
Capability

ChatGPT Shared Projects

Share one workspace where your team’s chats, files, and instructions stay in sync

Try ChatGPT Projects→
Swipe to compare. Scroll inside each card for details.
Quick start
What you need
  • A ChatGPT Business, Enterprise, or Edu workspace (sharing Projects with teammates requires a team plan)
  • At least one teammate to invite
  • 1–3 reference files your team uses repeatedly (strategy doc, brief, spec, spreadsheet)
  • 5 minutes to set instructions + invite your first collaborator
⚡5 minutes to try this
Step 2
In your shared Project chat, paste this prompt:
Step 2
Prompt (copy / paste ready)
We’re working inside a shared ChatGPT Project. First, summarize (in 5 bullets) what this Project is for and the current decisions/assumptions from the chats + files here. Then create a one-page plan with: 1) Goals (max 3) 2) This week’s priorities (max 5) 3) Open questions (max 5) 4) Risks + mitigations (top 3) End with: ‘What should I ask you to clarify next?’
Step 3
You'll get

A single ‘source of truth’ plan any teammate can open, extend, and reuse — with ChatGPT pulling from the same shared files, instructions, and conversation history

Notes
  • Notes: Create a new Project, add project instructions (purpose, tone, output style), upload your key files, then use Project settings to invite teammates.
  • Notes: If you don’t have a team plan, you can still use personal Projects — but you won’t be able to share the Project with teammates.
Context
Why this matters now
Expand
ChatGPT Projects are persistent workspaces that hold related chats, uploaded reference files, and project-level instructions. On Business, Enterprise, and Edu plans, you can share a Project with teammates so everyone works from the same context (no re-sending docs, no ‘what did we decide?’ threads). Personal Projects exist for all users, but sharing a Project is a team-plan feature.
Before / After
Before 2025
Each person starts from scratch (and context splinters across chats)
Now
One shared Project where context compounds for the whole team
⚠️
Field note

Common mistake: Inviting teammates before setting project instructions — the Project turns into a messy pile of chats and inconsistent outputs

Before you share, add project instructions that define: (1) what this Project is for, (2) key context ChatGPT should assume, and (3) how responses should look (tone, length, format). Then upload the ‘one true’ docs so everyone gets consistent answers.

Success
Success looks like…
  • Create one shared Project per initiative (launch, quarterly planning, hiring) so context doesn’t leak across workstreams
  • Pin or maintain a ‘Decisions so far’ message and update it weekly to keep the Project clean
  • Keep project instructions short and specific (tone, deliverables, definitions). Put long docs in files, not instructions
  • Start a fresh chat for each sub-task (e.g., ‘Risk review’, ‘Meeting notes’, ‘Draft v1’) while keeping the same shared context
  • When the initiative ends, archive the Project — it becomes institutional memory your team can reference later
04
Capability

Nano Banana Pro

Create and edit images with clean, readable text — right inside the Gemini app

Try it in Gemini→
Swipe to compare. Scroll inside each card for details.
Quick start
What you need
  • A Google account
  • Gemini app (mobile) or Gemini on the web
  • A clear idea of what the image is for (slide, poster, social post, diagram)
  • 2–5 minutes to generate + refine
  • Optional: AI Pro / Ultra for higher usage quotas
⚡5 minutes to try this
Step 2
Paste this prompt:
Step 2
Prompt (copy / paste ready)
Create a clean process diagram with 4 steps: Discovery → Analysis → Solution → Implementation. Design requirements: - Minimal modern style for a business slide - Large, readable text labels - High contrast - No extra text beyond the four step labels Make it landscape (16:9).
Step 3
You'll get

A presentation-ready diagram with readable text you can download and drop into a deck

Notes
  • Refine it conversationally: ask 'make the text bigger', 'use a blue + cream palette', or 'add small icons above each step' to iterate on the same image.
  • Image editing workflow: upload a photo (or the generated image) and say what to change (background swap, add/remove objects, change colors) — Nano Banana keeps context across turns.
Context
Why this matters now
Expand
Nano Banana is Gemini’s image generation and editing model. The big unlock is text: it can render clearer, more readable words inside images, which makes it useful for real work (diagrams, posters, social graphics) instead of just ‘concept art’. The fastest way to use it is directly in the Gemini app: describe what you want, then keep the conversation going to refine the same image. If you have a paid plan (AI Pro / Ultra), you’ll typically get higher quotas and more headroom.
Before / After
Before 2025
AI images with messy or unreadable text
Now
Images you can actually use (including readable labels + headlines)
⚠️
Field note

Common mistake: Being vague about the exact text — you’ll get incorrect or inconsistent labels

Write the exact words you want inside the image (titles, labels, step names). If the text is close-but-wrong, ask for a correction like: ‘Keep everything the same, but change “Implement” to “Implementation” and increase label size by ~20%.’

Success
Success looks like…
  • Prompt in layers: start simple, then refine with 2–3 small changes per turn (color → layout → typography)
  • Lock in constraints: ‘keep the same layout, only change the color palette’ or ‘do not change any text’
  • For social: explicitly request ‘vertical 9:16’ or ‘square 1:1’ and specify safe margins for text
  • If you see watermarks or need quick edits, try doing the final pass in a Google surface that supports image editing (e.g., Google Vids)
  • Make a mini style guide you reuse: paste the same ‘brand palette + font vibe + mood’ block at the top of every prompt
05
Capability

Perplexity Spaces

A shared research workspace where threads, sources, and instructions stay together

Try Perplexity Spaces→
Swipe to compare. Scroll inside each card for details.
Quick start
What you need
  • A Perplexity account (Spaces work on free + paid plans)
  • One ongoing topic or project (competitive intel, market research, a purchase decision, a technical deep dive)
  • 2–5 minutes to create your Space and set instructions
  • Optional: files to upload (PDFs, Word files, images — up to plan limits)
  • Optional: teammates to invite as viewers or research partners
⚡4 minutes to try this
Step 2
In your Space, paste this prompt:
Step 2
Prompt (copy / paste ready)
Summarize the three most important developments in [your topic] from the past 30 days. For each, include: (1) what happened, (2) why it matters, (3) what we should do next. Keep it under 400 words and include links/citations.
Step 3
You'll get

A focused briefing that stays inside the Space, so future questions build on the same threads and sources instead of starting from scratch

Notes
  • Name your Space (e.g., ‘Q1 Competitor Intel’) and add a short instruction like: ‘Act like a research analyst. Always cite sources. End with 3 next steps.’
  • Bonus: Click ‘Share’ to invite teammates with View or Contributor access (note: new Threads aren’t auto-shared unless they’re added to the Space).
Context
Why this matters now
Expand
Perplexity Spaces are dedicated workspaces that let you group Searches and Threads by topic or project, then share that knowledge hub with others. Each Space can have custom AI instructions (and, depending on your plan, preferred models and sources) so responses stay consistent without re-prompting. Spaces are private by default, but you can share them with view or contributor access, and even share a link. On Enterprise Pro, Spaces can search both the web and your organization’s files (including via connectors like Google Drive, SharePoint, OneDrive, Box, and Dropbox).
Before / After
Before 2025
Research scattered across tabs, links, and one-off threads
Now
A Space that compounds knowledge (and can be shared with your team)
⚠️
Field note

Common mistake: Sharing a Space but forgetting to add Threads to it — teammates won’t see your best work

After you run a great search, make sure the Thread is actually inside the Space (or add it). Then your team can follow up in the same Thread and keep the research chain intact.

Success
Success looks like…
  • Use View vs Contributor access so the right people can add Threads (and everyone can learn from them)
  • Pick Sources per Space (web, academic, SEC filings, etc.) to match the job you’re doing
  • Upload key reference docs so the AI can synthesize across your files and the web (Enterprise Pro adds org-wide file search + connectors)
  • Start from the Space Templates gallery when you want a fast setup (then customize instructions)
  • Run a weekly ritual: open the Space and ask for ‘new developments since last week’ + ‘what to watch next’
  • Export or share the Space link with stakeholders who just need view access
Tip: swipe left/right to switch. Scroll inside a card for details.
Faisal Shariff

Need Help Making This Real?

You've seen what's possible. The hard part isn't understanding AI capabilities—it's implementing them across your organization and building workflows that actually scale.

I'm Faisal Shariff. I've spent 20+ years helping organizations—from 10 to 300,000+ people—implement technology that actually works. These experiments show you what's possible. That's where I come in.

Let's Work Together →
Not ready yet? Get weekly AI experiments in your inbox →