LAB ENTRY #002 · Tool Review
The AI Stack Review
From "Frozen Screens" to Agentic Building — A Non-Technical Leader's Honest Verdict
The Context War
LLM Layer — Gemini vs. ChatGPT
I started building Agni QR in ChatGPT. As the project grew, I hit the Context Wall. The longer the chat went, the more the screen froze. Starting a new chat felt like a reset — there was no thread connecting the previous logic to the new one. Switching to Gemini changed everything. The moment I mentioned a topic from a previous thread, it switched gears, personalised the response, and carried the context forward. It felt like an ecosystem that actually knew my project. This is where I found my confidence in building with AI.
Gemini
GooglePros
- Superior context management — remembers projects across threads
- Switches gears instantly when you reference a previous topic
- Deep Google ecosystem integration reduces context-switching
- Personalised experience that compounds over time
Cons
- Can be overly cautious with safety filters on creative tasks
- Gemini Advanced is paywalled — free tier does not reflect the real product
- Image generation lags behind dedicated visual tools
The winner for long-form project building. Gemini is the only tool that feels like it is genuinely learning alongside your project — not just responding to your last message.
ChatGPT
OpenAIPros
- Best-in-class for one-shot copywriting and quick ideation
- Intuitive interface with the lowest barrier to entry
- Strong plugin and integration ecosystem
Cons
- The Context Wall — long chats degrade noticeably and screens freeze
- No shared brain between separate conversations on the same project
- Starting a new chat feels like explaining yourself to a new hire
Excellent for isolated, quick creative tasks and first-draft brainstorming. Loses its edge the moment your project gets complex — the context wall is real and frustrating.
Research & Learning
NotebookLM
While the rest of the internet was drowning in the complexity of n8n and Make.com documentation, I used NotebookLM differently. I dropped in dozens of YouTube resources and listened to them like a personalised podcast. It is the ultimate hack for learning technical workflows without reading a single manual.
NotebookLM
GooglePros
- Synthesises dozens of sources into a coherent, personalised knowledge base
- Audio overview feature turns dense documentation into a podcast format
- Answers questions from within your uploaded sources — no hallucination drift
- Dramatically reduces time-to-understanding for complex technical topics
Cons
- Quality is entirely dependent on the sources you feed it
- Cannot browse live web content or pull real-time data
- Gets you to the vocabulary layer — not a replacement for hands-on practice
The ultimate cheat code for non-technical leaders. While everyone else was drowning in n8n documentation, I turned a stack of YouTube links into a personalised automation podcast. This is how you learn technical stacks without the manual.
Visual & Creative Suite
Nano Banana & VEO 3
Product marketing runs on visuals. Nano Banana gave me a designer on demand — I could manipulate dashboards, swap company logos, and create pitch-ready mockups by describing exactly what I wanted. VEO 3 is a different story: the potential is obvious and enormous, but I am still the bottleneck. I see the 8-second cinematic magic others are making. I am still building the prompting vocabulary to get there.
Nano Banana
IndependentPros
- High-fidelity mockups from plain language descriptions
- Perfect for PMMs who need pitch-ready visuals without a design budget
- Handles brand asset integration — logos, dashboards, UI components
- Fast iteration cycle: describe a change, see it, refine
Cons
- Requires high-precision prompting — vague inputs produce vague outputs
- Less suited for animation or motion design
- Limited community resources compared to established tools
My secret weapon for product marketing. Manipulating dashboards, swapping logos, and creating high-fidelity pitch visuals in plain language — when the prompt is precise, the output is indistinguishable from a professional designer's work.
VEO 3 / Google Vids
Google DeepMindPros
- Cinema-quality 8-second clips from text descriptions when prompted correctly
- Massive scope for product storytelling and social content
- Google Vids integration creates a full script-to-video pipeline
Cons
- High learning curve — video prompting is a distinct skill from text or image
- Difficult to achieve a specific envisioned output without advanced expertise
- Community guidance and best-practice resources are still thin
The ceiling is clearly visible and it is extraordinary — cinematic 8-second product storytelling with no production budget. I am still the bottleneck here. The tool is ready; my video prompting vocabulary is not. Yet.
The Builder Layer
Antigravity — The Endgame
When I hit the Agent Manager in Antigravity, the game changed. I did not need a separate AI chat window running alongside a code editor. The agentic layer was inside the build environment itself. I described what I wanted in natural language. The agent reasoned through it and built it. I have zilch knowledge of coding, and yet I have a functional prototype. I do not know if this product will succeed in the market. But I know one thing: I am not being left behind in this AI hustle.
Antigravity
AntigravityPros
- Agent Manager removes the gap between "AI assistant" and "build environment"
- Natural language development with sustained project context
- Scaffolds real, working code — not pseudocode or suggestions
- Enables non-coders to ship beyond what chat-based AI makes possible
Cons
- Steeper initial setup than a chat-based tool
- Still encounters roadblocks on complex API integrations
- Requires clear logical thinking from the user — even without code, you must know the "why"
The endgame. The Agent Manager made me realise I did not need a separate AI assistant anymore. I could build the product, in natural language, inside the environment itself. A non-coder holding a functional prototype with zip downloads, branded QRs, scan analytics, and an advanced search bar — this is the tool that made that possible.
The Verdict
What a non-coder actually built
"I have zilch knowledge of coding, and yet I've built a functional prototype. I don't know if this product will succeed in the market. But I know one thing for sure: I am not being left behind in this AI hustle."