This post was originally published on this site
In this episode of the Agent Factory, Smitha Kolan and Vlad Kolesnikov are joined by Brandon Hancock, a full-stack engineer and the creator behind the YouTube channel AI with Brandon, where he teaches AI concepts to over 80,000 developers.
This was a very special recording, taking place just hours after Google released several major updates, including the new flagship model Gemini 3, the Antigravity coding environment, and updates to the Gemini CLI. We spent the episode exploring these new tools through live demonstrations.
Gemini 3 and Gemini CLI
Gemini 3 is Google’s newest flagship model, designed for advanced high-level reasoning and complex agentic operations, making it ideal for orchestration. Gemini CLI is a command-line interface that allows developers to interact with Gemini models directly from their terminal. It supports piping inputs and outputs, making it a powerful tool for chaining prompts and building lightweight “AI employees” without complex infrastructure.
Building a Personal Portfolio with Gemini 3 Pro
Timestamp: 01:18
Smitha kicked off the demos by using Google AI Studio and Gemini 3 Pro to build a personal website from scratch in minutes. The goal was to convert a LinkedIn profile into a deployed portfolio site.
-
Input Data: She uploaded a PDF export of her LinkedIn profile, a personal headshot, and an inspiration image generated by Nano Banana (an image generation tool).
-
Multimodal Prompting: She instructed Gemini 3 to use the PDF for content and the Nano Banana image for stylistic direction.
-
Iterative Refinement: When the initial output had a placeholder image, Smitha used the annotation feature in AI Studio to highlight the specific section and ask Gemini to fix it using her uploaded photo.
-
Deployment: The final result was deployed directly to Cloud Run with a single click.
Creating “AI Employees” with Gemini CLI
Timestamp: 07:54
Brandon showcased how he uses the Gemini CLI to build what he calls “AI Employees.” His demo focused on a market research agent designed to find potential customers for his startup in the Emergency Medical Services space.
-
Standard Operating Procedures (SOPs): Brandon treats prompts as SOPs stored in markdown files. He uses Gemini 3 Pro to write the high-level instructions and the faster, cheaper Gemini 2.5 Flash to execute the “worker bee” tasks.
-
Parallel Execution: He demonstrated a Python script that calls the Gemini CLI to run multiple agents in parallel. This allowed him to search for leads across multiple cities (e.g., Miami) simultaneously.
-
The Cheat Code: Brandon explained how passing a Python script to Gemini allows the model to understand the parameters and kick off the workflow itself, effectively automating the automation.
The “Anya the Capybara” Video Agent
Timestamp: 18:15
Vlad demonstrated an agent built with the Agent Development Kit (ADK) that turns technical documentation into engaging educational videos hosted by an AI character, “Anya the Capybara.”
-
Agent Architecture: The system uses a root orchestrator agent that manages two sub-agents: a Script Sequencer (which adapts technical text into a natural script and chunks it into 8-second segments) and a Video Generator.
-
Consistency: To keep the character consistent, Vlad used Nano Banana to generate four specific views of the Capybara. The agent randomly selects a view for each chunk to keep the video dynamic.
-
Antigravity in Action: Vlad revealed that he used the new Antigravity feature to refactor his agent’s code. He simply asked Antigravity to make changes to the prompts and sub-agents, and the tool handled the coding work automatically.
-
Video Stitching: Finally, he used the Gemini CLI to write an FFmpeg command to stitch the 92 generated video chunks into a single, cohesive 11-minute video.
Check out the code for this video generating agent here.
The AI-First Mindset
Timestamp: 10:21
In between the demos, we discussed the philosophy behind building these tools. Brandon shared his approach to productivity in the age of AI.
The Concept of AI Employees
Timestamp: 15:50
Brandon explained his goal to be “AI first” for everything. He views his agents not just as scripts, but as employees that free him up to focus on high-level strategy. By creating folders of “Standard Operating Procedures” (markdown files), he can assign tasks like market research or drafting emails to GeminiI. He noted that the barrier to entry is now so low that you did not need any coding experience to spin up a “workforce” of Gemini models to handle repetitive tasks at scale.
The “Ghostwriter” Workflow
Timestamp: 15:50
Beyond market research, Brandon touched on how he uses Gemini CLI as a “ghostwriter.” By hooking up an MCP (Model Context Protocol) server to Gmail, he can feed the AI his previous writing styles and have it draft responses to emails. The key, he emphasized, is providing the AI with the right context so it “knows who it is” before it begins the task.
Conclusion
This episode highlighted just how fast the development cycle has become. With tools like Gemini 3 Pro for reasoning, Gemini CLI for rapid execution, and Antigravity for code refactoring, we are seeing a shift where automating everything is becoming a reality. Whether you are building a personal website or a fleet of market research agents, the friction between idea and execution has never been lower.
Your turn to build
We challenge you to try building your own “AI Employee” this week using the Gemini CLI. Start with a simple task you do repeatedly, write an SOP in markdown, and see if you can get Gemini to take it off your plate.
Check out the full episode to see the code in action!
Resources and links
Google AI Studio and Gemini 3 → https://goo.gle/487Fnde
Antigravity → https://goo.gle/49z1sE9
Video Avatars Agent Repo → https://goo.gle/4oP8Vnd
