Welcome, humans.

Now, for today’s featured podcast release:

He built GitHub Copilot at Microsoft. Then he left for Google, built an AI coding tool in his terminal, and his team now ships 100 to 150 features and bug fixes every single week—using AI to build itself.

In his first-ever in-depth interview about Gemini CLI, Google Principal Engineer Taylor Mullen came on the pod and revealed how he manages swarms of parallel AI agents, why the humble command-line terminal is having a full-on renaissance, and the viral “Ralph Wiggum” technique that recently made waves among devs on X and Reddit.

Here's our favorite parts:

• (2:14) The origin story: Taylor built Gemini CLI at a hackathon two years ago—but scrapped it because AI was too slow and expensive. Fast forward to today and it's the most popular open-source CLI on GitHub.

• (8:21) The "holy crap" moment: Taylor told Gemini CLI to clear his packed schedule, DM everyone affected, and reschedule—all while he was at the gym. Done in five minutes.

• (17:24) Live bug-fixing demo: Taylor pastes a real GitHub issue URL into his terminal without even reading the bug first—and watches the AI solve it in real-time.

• (18:47) Spawning CLIs that spawn CLIs: His team runs 7-10 parallel AI agents simultaneously—each one working on a different task, notifying the human only when it needs approval.

• (23:05) Conductor — "Planning dialed to 11": Google's new extension doesn't just plan your project. It asks clarifying questions, writes implementation specs, self-improves over time, and helps your whole team get smarter.

• (45:01) "10x is the new normal": Taylor's team has moved past the 10x engineer debate entirely. Everyone is 10x now. The question is how you get to 100x—and the answer is parallelism.

• (45:56) The Ralph Wiggum Technique: The viral prompting method sweeping developer Twitter. Feed the AI's output back into the same prompt, over and over, until the answer is polished. Taylor runs it 5 times. Every time.

• (31:54) Why it's open source: "With a tool this powerful, how else do you trust it?" Over a million users are helping build and audit Gemini CLI in the open—every single day.

Why watch this? Because Taylor's live demos at (17:24) and (23:05) will show you exactly how to use a “CLI” / command line interface terminal agent actually works, which will help demystify it for you whether you want to try Gemini CLI, Claude Code, Codex, or any other tools.

Watch and/or Listen now: YouTube | Spotify | Apple Podcasts

P.S. Taylor's team prefers Gemini 3 Flash over Pro for almost everything—and he's only fallen back to Pro 10 times in the last month. Find out why at (43:44).

Keep scrolling for what's new with Gemini CLI, details on our upcoming live stream with the CEO of a humanoid robotics company tomorrow, and some other recent episodes you’ll love.

Real quick: Want to see your AI-adjacent product or service show up right here, below these podcast promos? Click the button below to advertise to our 650K readers!

NEW: Gemini CLI Just Got a Self-Correcting Upgrade

Timed with our interview, Google just launched Conductor's Automated Reviews—a major new feature for Gemini CLI that automatically audits AI-generated code for security risks, style compliance, and accuracy against your original plans.

Why it matters: The biggest concern with AI-written code isn't speed—it's trust. This update makes Gemini CLI the first major terminal tool to close the loop: the agent writes code, tests it, and peer-reviews its own work—all before a human ever sees it.

What Automated Reviews actually check:

• Code quality: Flags race conditions, null pointer risks, and logic errors—not just syntax.

• Plan compliance: Verifies every phase of your roadmap was addressed. No skipped requirements.

• Security scanning: Catches hardcoded API keys, potential data leaks, and injection vulnerabilities.

• Test-suite validation: Runs your full test suite and folds coverage data into the review report.

• Style enforcement: Ensures all new code follows your project's guidelines automatically.

Every finding is categorized by severity (High / Medium / Low) with exact file paths—so you know exactly where to look and can start a fix track inside Conductor immediately.

Also new: Agent Skills—a modular skills library that extends Gemini CLI with task-specific expertise (security auditing, cloud deployments, codebase migrations, etc.) without cluttering the model's primary context window. 

Based on the Agent Skills open standard, think of it like plug-and-play superpowers. Gemini autonomously decides when to use a skill based on your request and pulls it in on demand.

Oh, and the whole thing (Gemini CLI) is free and open source. Just thought we should mention that somewhere.

P.S: If you’re just getting into coding with AI, Taylor mentioned Test-Driven Development as a technique you can use to code with agents; there’s actually a popular plugin called “Superpowers” you can install to help you with this.

We wrote more about it on the website here.

🔑 The bottom line: The race isn't about who can write code the fastest. It's about who can write code you don't have to fix. Gemini CLI is betting hard on "trust, but verify."

Try it free: geminicli.com | GitHub (open source) | Extensions blog | Launch announcement

🤖 LATER THIS WEEK: Physical AI for Humanoid Robots — Neuron LIVE

Wednesday, February 18 at 8:00 AM PT / 10:00 AM CT / 11:00 AM ET

Humanoid robotics challenges go beyond movement and servo motors; there are massive AI challenges behind getting intelligence into the physical world, where gravity is real, friction matters, and mistakes break hardware.

This week on Neuron Live, we're joined by Nikita Rudin, Cofounder & CEO of Flexion Robotics, to unpack what it actually takes to build intelligence for humanoid systems. From training control policies and perception models to bridging simulation and the real world, we'll explore the AI stack powering the next generation of embodied systems.

If large language models are the brain in the cloud, what does intelligence look like when it has to walk, grasp, and not fall over?

Oh, and best of all? We’ll be getting a LIVE DEMO on a REAL ROBOT… you won’t want to miss this one!

Click to watch live based on your preferred channel below:

• YouTube

• LinkedIn

• X (Twitter)

IN CASE YOU MISSED IT…

1. Interested in New AI Architectures? Watch: Is This Energy Based Reasoning Model The Next AI Breakthrough? (w/ Eve Bodnia)

TL;DW: We sit down with Eve Bodnia, Founder and CEO of Logical Intelligence, to discuss Energy-Based Models (EBMs). Unlike LLMs that predict the next token, EBMs reason over an "energy landscape," allowing them to evaluate many possible solutions at once. Eve explains how this physics-inspired approach could solve the hallucination problem and enable true planning.

Why you should watch: This might be the clearest explanation of why current LLMs struggle with reasoning—and what a credible alternative looks like. If you want to understand the future of AI architecture beyond the Transformer, this conversation is essential.

• YouTube: Watch Here

• Apple Podcasts: Listen Here

• Spotify: Listen Here

2. Not an engineer (and don’t want to be)? Watch this: Inside Google Labs’ 3 AI New Tools That Will Change How You Create

TL;DW: We go hands-on with three cutting-edge tools from Google Labs: Mixboard (an AI-powered concepting board), Flow (an AI filmmaking tool for editing and animating), and Opal (a no-code AI app builder). Product leaders from Google join us to demo exactly how these tools fit into your creative workflow.

Why you should watch: Stop guessing what "multimodal" means and see it in action. These aren't just chat interfaces; they are practical, visual tools that let you build apps and videos using natural language. Perfect for creatives and builders who want to stay ahead of the curve.

• YouTube: Watch Here (Check Recent Videos)

• Apple Podcasts: Listen Here (Link for reference)

• Spotify: Listen Here

3. Want Agents that Adapt to Whatever Coding Platform You Like? This AI Agent Builds Better Code Than Most Developers (Factory AI)

TL;DW: Eno Reyes, co-founder and CTO of Factory AI, joins us to explain "Droids"—fully autonomous agents that can take Jira tickets, modify real codebases, and run tests. We dig into their context compression research and why code quality is the single biggest predictor of whether an AI agent will succeed or fail.

Why you should watch: Autonomous coding is moving from "cool demo" to "production reality." This episode gives you the blueprint for making your own codebase "agent-ready" and explains why the future of software engineering might look less like typing and more like reviewing.

• YouTube: Watch Here

• Apple Podcasts: Listen Here (Link for reference)

• Spotify: Listen Here

4. Want a Deep Dive on Coding with Agent Best Practices? Watch: Our GPT Codex Deep Dive (Live Demo)

TL;DR: Speaking of learning to code with AI, check out our 2-hour, hands-on deep dive into OpenAI’s Codex coding agent with Alexander Embiricos, product lead for OpenAI Codex. Codex Max is OpenAI’s latest frontier agentic coding model, built on an upgraded reasoning backbone to handle real-world software engineering tasks end-to-end (PRs, refactors, frontend builds). We demonstrate how it can work independently for hours, compact its own history to maintain context, and we build real agents live to push it to its limits.

Why you should watch: You’ll walk out feeling like an agentic-coding wizard, even if you’re starting from zero. It’s part Q&A part tutorial on how to set up and run multi-hour agent loops that actually ship code. And if you prefer, here’s a blog version!

• YouTube: Watch Here

• Apple Podcasts: Listen Here

• Spotify: Listen Here

One more before you go:

Last week, we spent 3 hours building AI agents live on camera. Microsoft's Bryan Goode demoed a tenant running 128,000 agents. Corey's OpenClaw agent literally haunted his house (his wife texted mid-stream in a panic because a voice was lecturing about tokenization from the living room). We built a landlord-tenant law scout in Claude Co-work in five minutes. GPT 5.3 Codex Spark dropped mid-stream. And yes, we made Cat Doom (our personal AI coding benchmark) with the Codex app…

The full breakdown—every tool, every timestamp, every tutorial—is in our deep-dive article. Or just watch the whole stream.

Last thing: if you haven’t subscribed yet, please do! All you have to do is click the image below, confirm, and you’ll be subscribed to the channel 😄 

We have a goal to hit 50K subscribers by the end of the year (if not 100K), and we’re only 35K away! If you like learning about AI, and already watch some of our videos, do us a favor and click here to subscribe today.

Stay curious,

The Neuron Team

P.P.S: Love the newsletter, but don’t want to receive these podcast announcement emails? Don’t unsubscribe — adjust your preferences to opt out of them here instead.