Testing My AI Stack

All right. Here we go.

I’ve been watching Matt Wolf again and finally decided to test some of the AI tools he mentioned in a recent video. My curiosity was overdue.

The list:

• Nano Banana (tested previously in lm arena)
• ElevenLabs
• Granola
• Cursor
• WhisperFlow
• Claude
• Perplexity Pro

Let’s start in reverse.

Nano Banana (Google)

Nano Banana lives inside Gemini. I clicked around, found the image generator, and asked it to create a Bukovian dog living in The Hague.

Naturally.

It generated something surprisingly close to Bert — similar coloring, decent vibe. Not quite The Hague, but close enough. Entertaining. Practical. I’ll probably use it again.

Image tools are becoming table stakes.

ElevenLabs

This one intrigued me.

I had some signup issues (thanks, Apple VPN), but eventually got in. The part I really wanted — voice cloning — is behind the paid tier. Apparently it only needs two minutes of your voice.

That’s… mildly unsettling.

I did test text-to-speech and even found a Trinidadian voice option, which was honestly impressive. The UI feels like a mini sound studio: voice changer, isolator, dubbing, audiobooks, soundtrack generator.

Useful? Yes.
Necessary? Not yet.
Security implications? Questionable.

Part of me wants to spend €5 just to hear myself say things I’d never actually say.

Granola

Granola turns your messy meeting notes plus transcript into something polished.

Which sounds great.

But also slightly dystopian.

As a security person, I can’t help wondering where all that data is going. Every productivity tool is also a data pipeline. That tension is hard to ignore.

Still, if you’re presenting frameworks or executive summaries often, I can see the appeal.

Cursor

Cursor is interesting

AI-native coding platform. Looks like a Visual Studio fork with agents built in. Integrates with Cloudflare, Databricks, Stripe, Vercel, Slack.

This is the one I need to spend real time with.

I’m thinking about using it to build out my M365 SRA framework. If it actually accelerates structured security design instead of just autocomplete, then it’s worth exploring.

But I don’t judge tools in 15 minutes. This one needs a proper test.

WhisperFlow

Voice-to-text that removes the “ums” and “ahs.”

Apple’s dictation isn’t great. If WhisperFlow actually cleans up speech into usable prose, that’s powerful — especially for someone rebuilding public speaking confidence.

Still skeptical about installing too many background apps.

Claude & Perplexity

Claude has a strong reputation. I’ve been quietly comparing it against ChatGPT for a few prompts.

It performs well.

It’s also slightly cheaper.

Perplexity Pro is search-first. Supposedly better than Google. My honest first impression? I don’t like the font. It sounds trivial, but UX matters. If it’s unpleasant to read, I won’t use it.

Small things matter.

So What’s My Stack?

Right now:

• ChatGPT
• ChatGPT Atlas
• Maybe Perplexity

Switching tools isn’t just about features. It’s about familiarity, voice alignment, and workflow friction.

ChatGPT understands my tone. My structure. My thinking patterns.

That’s not trivial.

The question isn’t which tool is “best.”

It’s which tool compounds.

For now, I’m staying put — but experimenting.

We’ll see.