The argument in one line.
Gemini CLI excels at adding features to existing codebases with context but fails at generating visually compelling projects from scratch without sufficient architectural guidance.
Read if. Skip if.
READ IF YOU ARE…
- A developer working in an existing codebase who wants to quickly test whether Gemini CLI can handle real maintenance tasks like bug fixes and feature builds.
- A developer evaluating free AI CLI tools and needs a straightforward comparison of Gemini CLI's actual performance against its advertised capabilities and rate limits.
- A terminal-first developer already familiar with Claude Code or similar tools who's curious whether Google's free tier makes switching worth considering for incremental coding work.
SKIP IF…
- You're building greenfield projects or starting from scratch — the video concludes Gemini CLI fails on new projects with no existing context.
- You need to evaluate enterprise-grade or production-ready AI tooling — this tests the free tier at hobbyist scale, not paid plans or reliability guarantees.
- You're comparing IDE-integrated tools like Cursor or Gemini Code Assist — this breakdown focuses exclusively on terminal-based CLI usage.
The full version, fast.
Google's Gemini CLI is a free terminal-based AI coding agent that ships with Gemini 2.5 Pro, a 1M-token context window, 60 requests per minute, and 1,000 requests per day at zero cost, plus built-in Google Search, MCP server support, and a customizable GEMINI.md memory file. Three live tests reveal where it actually delivers: a one-line CSS scrolling fix passed instantly, and a medium-sized build adding a full memory feature with database migration, schema changes, new UI page, and API integration completed in minutes for three cents in tokens. A from-scratch Next.js landing page failed even after iteration, producing flat visuals. The rule for you: lean on Gemini CLI inside existing codebases where context is rich, not on greenfield projects.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:36
01 · Cold open — direct competitor, completely free
Talking-head intro promising the breakdown: what makes Gemini CLI special, how to set it up, and three real-world coding tests.
00:36 – 05:00
02 · Gemini CLI announcement — open source AI agent for your terminal
Walks through Google's launch blog post. Open-source CLI for the terminal, two modes (interactive REPL + single-shot prompt), mentions Gemini Code Assist as Google's Cursor competitor. Frames Google as 'AI everywhere developers are.'
05:00 – 07:30
03 · Free-tier breakdown — the price-objection killer
Full-screen card: 60 req/min, 1,000 req/day, Gemini 2.5 Pro (1M context), Open Source, Available free-of-charge. Brandon does the math: a single 1M-context request would cost $3 on the API; he just gave you $180/day for free.
07:30 – 11:23
04 · Major features tour + setup
Built-in Google Search from the terminal, MCP server support (image-gen + Veo video gen demo from launch post), GEMINI.md custom prompt file, scripting/automation. Then setup: npm install -g @google/gemini-cli, run gemini, log in with personal Google account for free tier, or paste GEMINI_API_KEY into .env to bypass limits. Tours /version, /theme, /editor, /tools slash commands.
11:23 – 14:54
05 · Case 1 — Simple code fix (PASS)
Real bug: profile page won't scroll, content cut off at the bottom. Brandon @-references the file, Gemini investigates, proposes a CSS overflow fix, opens the diff in his external editor (Cursor) for review, applies. Then he teaches Gemini a project rule on the fly: 'only run npm lint, never run npm run' — update memory — Gemini writes it to GEMINI.md so the rule persists. Verdict: clean pass.
14:54 – 26:18
06 · Case 2 — Medium task: full Memories feature (BIG PASS)
Adds an entire new Memories tab to his ShipKit chat template: new sidebar entry, CRUD page, Postgres schema migration, API changes so every chat injects memories into the system prompt. Uses an AI-driven workflow — screenshot + task_template.md + GEMINI.md as context, asks Gemini to plan first, reviews the multi-phase plan, requests Phase 0 (schema), then implements phase by phase. Gemini lints between steps, fixes its own errors, ships working feature in minutes. End-to-end test in-app: types a memory, sends a chat, response respects it. Verdict: massive pass.
26:18 – 34:16
07 · Case 3 — Hard test: new landing page from scratch (FAIL)
Empty folder, .env with API key. Prompt: build a Next.js landing page for the AI With Brandon channel, look me up on YouTube first, make it beautiful and modern. Gemini researches the channel, drafts copy. Hits a wall trying to run create-next-app interactively (CLI wizards confuse it) — Brandon escapes, runs npx manually, hands the scaffolded project back. Gemini styles it, but the result is generic and ugly. Second pass with a screenshot + 'do not stop until it's absolutely beautiful' — slightly better, still underwhelming. Verdict: fail. 'Gemini CLI thrives with context, struggles without it.'
Lines worth screenshotting.
- Gemini CLI's free tier — 60 requests per minute, 1,000 requests per day, Gemini 2.5 Pro, 1M context window, $0 — is not a trial; it is a permanent tier that makes professional-grade AI coding available without a subscription.
- The 1M token context window is Gemini CLI's structural advantage: it can load an entire large codebase into a single session where Claude Code would need to navigate selectively.
- Gemini CLI thrives on existing codebases and fails on greenfield projects — the inverse of what most developers assume, and the rule that determines where it belongs in a multi-tool workflow.
- A full memory feature with database migration and UI built correctly in a single Gemini CLI session is not a simple task — it is the category of work that justifies using a tool with deep codebase context over a tool that generates from scratch.
- Generating a from-scratch Next.js landing page without existing code context is where Gemini CLI produces generic, usable-but-not-good output — which is the accurate boundary test that separates demo conditions from production conditions.
- Gemini CLI is open source, which means the rate limits and free tier are not a marketing hook that disappears in six months — they are baked into a product with a public repository that anyone can inspect.
- The correct use case framing for Gemini CLI vs Claude Code is not 'which is better' but 'which task has more existing context to work from' — Gemini wins when context is rich, struggles when context is empty.
- A simple one-line CSS fix handled correctly by Gemini CLI confirms the floor: it can do the easy stuff reliably, which is the prerequisite before trusting it with the medium-complexity work.
- Google releasing a Claude Code competitor for free is a competitive pressure event that will force Anthropic to either match the free tier or differentiate on quality — neither outcome is bad for users.
- 60 requests per minute is a high enough rate limit to run Gemini CLI in agentic loops that make multiple tool calls per task without hitting the ceiling during a normal work session.
- The right workflow for teams is Gemini CLI for existing codebase work and a generative tool for greenfield — splitting by task type rather than choosing one tool eliminates the failure modes of each.
- A terminal-based coding agent from Google authenticating through an existing Google account removes the API key setup friction that stops non-technical users from starting.
- The medium-complexity test — a feature that required database migration logic, backend integration, and frontend UI — being handled correctly in one session is the result that matters most for real-world adoption decisions.
- Greenfield project failure is not a model quality issue; it is a context issue — the model has no existing patterns, no naming conventions, no architectural decisions to build on, so it produces generic output.
- The honest three-test verdict format — one pass, one pass, one fail — is more useful than a demo that only shows the tool succeeding, because it tells you exactly where the boundary is before you find it yourself in production.
Gemini CLI Is Free and Extraordinary — With Enough Context
Gemini CLI review
Brandon Hancock's three-test review shows that Gemini CLI thrives on structured context and multi-phase plans but fails at open-ended from-scratch builds where it has to make aesthetic and architectural decisions without guidance.
01Cold open — direct competitor, completely free
- Direct Claude Code competitor, completely free — three real-world tests to find where it passes and where it fails
03Free-tier breakdown — the price-objection killer
- 60 requests per minute, 1,000 per day, Gemini 2.5 Pro, 1M context window — completely free with a personal Google account
- A single 1M-context API call costs $3 — the free tier delivers $180 per day of compute at zero cost
04Major features tour + setup
- GEMINI.md is the persistent project rules file — the equivalent of CLAUDE.md for Gemini CLI sessions
- MCP server support enables image and video generation directly from the terminal
05Case 1 — Simple code fix (PASS)
- CSS overflow fix with the file referenced is a clean pass — at-reference the file, let Gemini investigate, review the diff
- Teaching a project rule on the fly — only run npm lint — and watching Gemini write it to GEMINI.md is the memory workflow in action
06Case 2 — Medium task: full Memories feature (BIG PASS)
- Screenshot plus task template plus GEMINI.md as context, plan first, implement phase by phase — this workflow produced a working full-feature build
- Gemini linted between phases and fixed its own errors — the structured workflow is what enabled the clean execution
07Case 3 — Hard test: new landing page from scratch (FAIL)
- Empty folder plus open-ended prompt produced generic output — no context means no quality
- Interactive CLI scaffolding broke the agent — run scaffolding commands manually and hand the project back rather than having Gemini attempt interactive terminal wizards
Terms worth knowing.
- CLI (Command Line Interface)
- A text-based way to control software by typing commands into a terminal instead of clicking through a graphical interface.
- Gemini CLI
- Google's open-source AI coding agent that runs inside a terminal, letting developers chat with Gemini models, edit files, run searches, and execute tasks from the command line.
- Claude Code
- Anthropic's terminal-based AI coding agent that reads, writes, and edits files in a project through natural-language conversation. Often cited as the direct competitor to Gemini CLI.
- Interactive mode
- Running a CLI tool as an ongoing chat session where you can send multiple messages, see streaming output, and approve actions, rather than firing one isolated command.
- Single-shot request
- Calling an AI tool with one self-contained prompt that returns an answer and exits, rather than starting an ongoing conversation.
- Cursor
- An AI-first code editor built as a fork of VS Code, with chat, inline edits, and codebase-aware suggestions powered by large language models.
- Gemini Code Assist
- Google's in-IDE AI coding assistant that adds chat, inline edits, and file-aware suggestions to editors like VS Code and JetBrains. Positioned as Google's competitor to Cursor.
- Gemini 2.5 Pro
- Google's flagship large language model at the time of the video, featuring a one-million-token context window and strong reasoning, coding, and multimodal capabilities.
- Context window
- The maximum amount of text, measured in tokens, that a language model can consider in a single request, including both the prompt and its response.
- Token
- The basic unit of text a language model processes, roughly equivalent to a short word or word fragment. Pricing and context limits are measured in tokens.
- MCP (Model Context Protocol)
- An open protocol that lets AI agents connect to external tools and data sources, such as image generators, Notion, or Gmail, through standardized server interfaces.
- MCP server
- A small program that exposes a specific tool or data source to an AI agent over the Model Context Protocol, allowing the agent to call its functions during a session.
- Veo
- Google's generative video model, used here through an MCP server to turn a still image into a short AI-generated video clip.
- GEMINI.md
- A markdown file placed in a project that Gemini CLI loads as persistent context, used to store project rules, conventions, and instructions the agent should follow on every run.
- npm
- The standard package manager for Node.js, used to install JavaScript libraries and command-line tools globally or per project.
- Node.js
- A JavaScript runtime that lets developers run JavaScript outside the browser. Required to install and run many modern CLI tools.
- Global install
- Installing a package system-wide rather than into one project, so its command can be invoked from any directory in the terminal.
- Workspace account
- A Google account managed by an organization through Google Workspace, distinct from a personal Gmail account and often subject to different access policies.
- API key
- A secret string that authenticates requests to a paid service, letting the provider track usage and bill the account that owns the key.
- .env file
- A plain-text configuration file holding environment variables such as API keys, kept out of source control and loaded by an app at runtime.
Things they pointed at.
09:00toolContext7 MCP server
09:00toolGoogle Veo (used in MCP video-gen demo from launch post)
Lines you could clip.
05:32
“If you ran 60 of those massive requests, you're easily getting up to a $180 worth of tokens from Gemini 2.5 Pro completely for free.”
31:48
“I never wrote a single line of code. I added in a whole new feature, database change, UI change, and updated API calls, and it worked in under two minutes.”
33:56
“Gemini CLI thrives with context, and without context, it's struggling.”
34:10
“Use Gemini CLI on your existing projects to add in new features and small changes. I would not recommend it right now for creating brand new projects.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphoranalogystory
00:00Hey, guys. Google just released Gemini CLI which is their direct competitor to Claude code. And the craziest part is not only is the new Gemini CLI tool insanely powerful, Google is basically giving it away to all of us completely for free.
00:15And in today's video, I'm gonna break down everything you need to know about Gemini CLI. So we're gonna cover what makes it so special, how to set it up, and then finally, we're gonna put it to the test on a few different code examples so that you can see firsthand should you use this tool, should you not.
00:32So without further ado, let's go ahead and dive in and cover this new amazing tool. Alright. So let's break down the major news that Google just released when they were announcing Gemini CLI, is their open source AI agent for your terminal.
00:45So first things first, Gemini CLI is a command line interface tool, which means you're going to be using it strictly inside of your terminal. And if you have seen it announced so far, 99% of the time when people are showcasing this tool, they're showcasing the interactive version of using Gemini where you just run Gemini, and then you get this amazing little terminal output right here so you can start chatting with your AI agents.
01:09But there is another option, which is to actually use Gemini to make single request where you say, hey, here is a prompt, and I just want you to answer this question really fast. So you can actually use Gemini CLI in multiple ways, not just interactive mode.
01:24You can also do single shot request. So pretty cool. Now, one thing I thought was super interesting when I first saw this announcement was, Well, I already have cursor.
01:32Now there's this thing called Gemini CLI to actually allow me to, like, chat with an agent in my terminal. But does Google have their own, like, cursor? I And just found out today, the answer is yes.
01:41Gemini has the code assist tool. So if you actually hop over here, you can see right here that Google has their direct competitor to cursor to where you can, like, chat just like you would in cursor. It'll make changes in your IDE.
01:56You can add in files you want to look at. So Google has . If you want to see a video where I cover this, you know, Gemini code assist as a competitor to cursor, happy to do it.
02:06But back to the main article, there is AI in everywhere, in your terminal, in your editor. So they are just at any point, you as a developer are interacting with code.
02:15Gemini is trying to be there to help you out. So let's dive into, I think, the most important piece of news when it comes using this new tool, which is basically, it is completely free to you as an individual developer. You're gonna get access to Google's most advanced model, Gemini 2.5 Pro, completely for free.
02:34You're gonna get up to 60 model request a minute, which is insane. That's like, you know, per minute. So that means a request per second is what you're getting.
02:42And you're gonna get to a thousand request per day completely for free. So if you kinda break down the math on this, Gemini 2.5 Pro has a 1,000,000 token context window, which if you were to use this just normally out in the wild, that would charge you $3 for that one request right there.
02:58And you can make dozens of those requests. So if you want to just ingest insane amounts of information and then get summaries, you could do that. So you're easily getting up to like, if you ran 60 of those massive request, you know, you're getting easily a $180 worth of tokens from Gemini 2.5 pro completely for free.
03:17So they're, uh, you know, holding back no expense when allowing developers to try out this new tool. And I've been loving it so far. You're gonna see this more in a minute.
03:25But just want to let you guys know it is wild how much free tokens they're giving away for you to use their most advanced model. So so pumped. Alright.
03:33So let's dive into some of the most important features of using Gemini CLI and you're gonna see these in action throughout the rest of the video. First things first. Gemini CLI allows you to actually run Google search which is a trippy concept at first that from your browser you can actually search using terminal, which is insane.
03:50The other thing that was really powerful is that you can use, you know, MCP servers directly in your terminal. So they have a nice little demo down here.
03:59Please make me a video of a cat that takes thirty seconds. So they're using MCP tools to first create an image, and then they're using Vio to convert that image into an actual video and saving it all locally.
04:11So that's only available by using MCP servers. So this is really cool that you're able to do this once again directly from your terminal. Then there's a few other important pieces of information to know, like you can update some customized prompts so you can make sure the Gemini CLI tool works exactly how you want it.
04:29That's gonna be done using a something called a Gemini dot markdown file. Don't worry. We're gonna cover that later.
04:33And then finally, you can actually automate task and integrate with existing workflows. You basically calling it through your scripts. So don't worry.
04:40You're gonna see most of this later throughout the rest of the video. But I just wanna quick before we dive into text like coding, I wanna show you some of these coolest examples that they're advertising completely for free. And let's go ahead and hop over to terminal fast so you can see these in action.
04:52And don't worry. We're gonna help you set up everything on your own in just a second. So let's hop over to the Okay.
04:57So I just turned on Gemini CLI, and I just wanna show you three of the most common ways you're gonna be using this tool so you can see in action. So first things first. Like I said, it comes built in with Google search.
05:07So you can say things such as, hey, please get me the weather today in San Francisco. You can actually run it. And what it's going to do is it's gonna go off, actually search the Internet.
05:18So you can see right now it's actually calling the Google search tool. It's actually putting in the input for the tool call, and then it's gonna spit back a really nice result for us. So we can actually like right from our terminal start interacting with the world.
05:30Thanks to the power of AI agent. So you can see, hey, the current weather is this. It has a west wind and it gets, you know, pretty chilly tonight.
05:38What's also pretty cool is Gemini CLI also allows you to hook up MCP servers. So I just put one in real quick. You can hit control plus t to actually see all of your MCP servers.
05:49So you can see right now I have one set up called Context seven, which allows me to look up any library doc for most software packages. So I can say please use context seven to see how to create a NextJS project.
06:05Now, obviously, this is a pretty straightforward example where we're just looking up documentation. But what's pretty cool is as you saw in the initial demo, you can use it to hook up to Gemini's different image creation tools, their video creation tools.
06:19Basically, if you can think of a MCP server, like connecting to your Notion, your Gmail, you can control it right here from your terminal, which is insanely powerful that you're allowing your agents to take action on your behalf right here from your terminal. And you can see it came back with the right answer. So in our case, if we want to create a new Next.
06:36Js project, we just run this command right here. Great. And the final thing that you guys are 99% of the time most likely going to use Gemini for is to make changes to your code.
06:45So you can easily come in here and you can actually type in just by using at. You can see at to file path. You can see at and I can do read me and perfect.
06:55You can see what does it say. And boom. Right then and there, you're going to be using Gemini CLI to analyze your doc.
07:02You can use it to make changes. Don't worry. We're gonna put this bad boy to the test here in just a second.
07:07But, yeah, you can see from one terminal using this tool, we can now interact call agent to do MCP lookups, Google searches, and to start editing our code. So super powerful. Now, what I wanna do is walk you through real quick how to set it up, and then we're going to dive into some code tests so you can see, hey, how powerful is this thing?
07:25And to answer the question, should you be using it? Alright. So now it's time to set up Gemini CLI on your computer and this is insanely simple to follow.
07:33You just first need to head over to GitHub. I'll have a link to this down description below. But all you need to do is run these two commands here.
07:39So first things first, you need to have node installed on your computer. And then second, you need to actually have this package installed.
07:47So what I like to do is run this command right here, which is going to install the Google Gemini CLI globally on your computer. So we're just gonna copy that. We're gonna head over to our terminal.
07:58We're going to paste it in, and this is going to just set everything up, and things are going to be good. Once it's done installing, you should be able to run the Gemini command just like this, and it will open up a new interactive shell just like this to where you can see everything that's going on. Now, mine's in a weird state because I've already done this like seven times.
08:19But most likely when you log in, you're gonna see a screen that looks just like this, which is gonna say, hey, how do you want to log in? Now to get access to all of those free credits that Google's giving away, you want to do login with Google. Whenever you do that, you're gonna see a window that looks just like this and you want to pick your personal account, not like a workspace account.
08:40You wanna do personal. So you're gonna click your personal account, say, yes, I would like to sign in, and then you are good to go. You're authorized to start working.
08:49And fantastic. You're off to the races. You could start coding.
08:51Now, for those of you who are out there just churning away code like there's no tomorrow, you will eventually hit the limits. So if you do hit the limit of using Gemini CLI, all you have to do is swap over to using a different key. So I'm not gonna show the key, but basically, what you do is you have a dot ENV file, and all it has in there is just gemini underscore a p I underscore key, and you paste in your gemini API key, and then you can use this tool as much as you want without limits.
09:24It is just gonna charge your account. So you can see now if I do off again, I can now say, hey, I'd like to use my Gemini API key, and it's like, Everything's cruising now. I can make request, and everything is looking good.
09:36Now that's everything you have to do. At this point, you could go ahead and start using it. I do just wanna show you real fast that this tool gives you a ton of different options just by using forward slash.
09:46So forward slash, you can check out which version you're using. You can also check out, you know, hey, I would actually like to change the different theme that I'm using. So I can change to use, you know, GitHub Lite.
09:59I can pick to use yeah. So I can change themes however I want. You can hit forward slash again.
10:04You can also change the editor. So anytime there's like a bug fix that you want to edit that the agent produces, you can say, oh, here's the editor I'd like to use.
10:13So on my case, I'm using cursor. So I wanna say, great. I wanna use cursor and I want to apply this globally.
10:20Fantastic. So now anytime I wanna make a change, it's gonna be done inside a cursor. So you can really I recommend playing around.
10:27Just hit forward slash and start playing with all the different commands and options that they give you. And then finally, you can say tools, and this will list out all the different tools that Gemini gives you access to. And there's a ton.
10:40You can basically search the web, read folders, search files, and do Google search. So, yeah, there's a ton that comes right out of the box. And don't worry, can always add more MCP servers.
10:49Show you how to do that later on at the end. Alright. So now that we have all the basics covered, that we have it set up running our computer, let's hop over to the code portion.
10:57So pumped to show you this in action. We're gonna first do a simple code fix, then we're going to do a more of a medium, like, big task. And then finally, we're gonna try and spin up a new project completely for scratch just to see how Gemini can handle it on.
11:10And so you can see for yourself, is this something you want to start adding to your developer workflow or not? So that's exactly what we're gonna test now. So let's hop over to making our first code change with Gemini CLI.
11:20Okay. So it's time to dive into our first code example using Gemini CLI to see if it can handle real world coding task. And this is going to be the simple example just so we can see how powerful this tool is.
11:32So the problem we're trying to solve is I'm working on a application here and you can see when I try and scroll for whatever reason, it's cutting off the bottom and not allowing me to scroll. So what we can do is start to use Gemini CLI to see if you can handle it. So what we can say is say, hey, my profile page.
11:52So we can scroll through all of our components right here. I can say, it looks like it looks like something is wrong with scrolling and it won't let me see the bottom of the page.
12:06So this is, as you can see, probably one of the most straightforward simple request. This is probably some sort of styling issue, but we're going to see a Gemini 2.5 Pro can handle these types of issues inside of Gemini CLI. So you can see right out the gate, it's going to investigate the issue to see if something's going wrong.
12:24Then you can see it's starting to ask, I think I have found the issue. Now, how would you like me to incorporate this change? So there's allow once, allow always, and modify with an external editor.
12:36I wanna show you the modify with an external editor real fast just so you can see in action. So I'm gonna show it right here. And now, you can see right out the gate, we're gonna make this a little bit bigger.
12:46So you can see oh, so here is my current code for this. And you can see here is my updated code for this. And you can see it's just a simple change right here to allow for overflow.
12:58So basically, hey, or can it go outside of the bounds vertically. So we can go, that looks pretty interesting. I think that's right.
13:05If I open this back up, you can see it's going to now allow me to say, yep, I want this change. I could have edited it, but I left it exactly how it was just so we can run it. So we're gonna go, yep, go ahead and make that change.
13:19Now we're gonna give it a second. It's going to apply it and instantly, it's going to try and run the code. Now in this case, I'm going to say no.
13:28So you can just hit escape and then what we can do is say, just for future reference, only run NPM lint, never run NPM run.
13:39And I think we say update memory. And then what this will do is it should start to update our memory. Yeah.
13:46Prioritize yeah. So you can now see I have updated our Gemini dot we'll scroll up so you can see it. Basically, we've updated our Gemini dot markdown file and we've refreshed memory.
13:56So we can say, yep. You can always run Lint to see if there's any issues. There shouldn't be.
14:02And you can see great. Things look great. And I just wanna show you too because this is pretty cool.
14:06So we can go ahead and hit escape. Well, I can now type in memory and I can do memory show.
14:12And what this should do is it'll show us all of the rules for this code. And you can also see, hey, there's a new piece of memory now to only run NPM run lint. So it's kinda cool.
14:23You could actually train the CLI tool to behave exactly how you would want it to. So just say update memory and it'll know what to do. Really freaking cool if you ask me.
14:31So now when I come back over here, refresh the page, you can see I can now scroll. So we're gonna give it a check. It was able to solve simple problems.
14:39And now what we're gonna do is go over to a bigger task so we can see if Gemini CLI can handle bigger task in my preferred way for working with these AI driven workflows. So let's hop over to code example number two.
14:51Okay. So now it's time to work on task number two, which is gonna be honestly a medium sized task inside of our project. And just to get a little bit more context, let's quickly go through what the app does and then cover how we can actually use Gemini CLI to do it for us.
15:05So long story short, this is a template that I'm working on for a project for you guys called ship kit to where you can easily crank out AI projects. So there's just a bunch of templates of like real world AI applications. So this is one of the templates that I was working on for you guys.
15:19So you can see it's basically a chat application to where you can swap between all the common models, try them out, and actually have like full blown conversation. So I was, you know, trying to let's see. Like, I was gonna go on a diving trip to Saint John's so you can see like, hey, it's chatting with ChatGPT except we can use every model here and swap between all of them.
15:38That's what the project is. Now, here's what I'd like to do. I want to, in our case, create a brand new tab, and this brand new tab should be called memory.
15:47And what memory should do is allow us to just like we can over here in Gemini CLI to where we can add memory, That's exactly what it should do over here.
15:57We should be able to create new memories that will automatically be included in all of our chats just to give it more context. So here's how we can do this. So first things first.
16:07I'm going to take a screenshot. From there, I'm going to paste this in our images folder. And now what's pretty cool is I can actually start using the Gemini CLI tool.
16:18I can start to interact with it. So let's actually run clear really fast. So we'll run clear.
16:24Cool. Things are ready for us to start working. And here's what I'd like to do.
16:28In the age of AI, I like to do AI driven task development. So as you can see, as I've been working on this project, I for every single major feature, I pretty much just come in here and create a brand new task. So you can see for every major task, I don't do the coding.
16:45I let AI do the coding for me. And in order to get, like, really high resolution task just like this, there are templates. So you can see I have a task template to where basically it gives really good instructions on how to create projects, what's the goal, tech stack, so forth and so forth.
17:02But we can take this a step further because when you're working with Gemini CLI, you'll create a Gemini markdown file, which will basically just be like one master source of truth, which can talks about everything inside of your project to help give Gemini more context about what it's working on that so when it's making changes, it knows what to do.
17:22So this is where you'll continually add in more and more information about your project. Every time it makes a mistake, you'll come down here and just add, like, you know, important and then basically add new rules of, like, you know, hey, don't do this. Do this.
17:36Keep this in mind. Yeah. So we basically this is where you're going to keep up with everything inside of your project.
17:41So let's see how we can use all of these together. So what we're gonna do is say, please let's do this. We're gonna say, let's click down here.
17:49Please use. And now what we can do is start to add in documents. So you can see we added in a new screenshot.
17:56So that's our screenshot that we just took of this section over here. And now what I can say, and we're also going to use task template. Great.
18:04And then now this is where we can actually start coding and telling it exactly what we want. So what we're gonna do is this. Can you please help me add a new piece of functionality to my project?
18:14Specifically, what I want to do is add in a new memory tab. And what this memory tab should do is it should be a CRUD table where I can add in new text memories.
18:29These text memories need to get included as context for all messages with the AI.
18:37So inside of the AI sorry. API forward slash chat, we need to update so that every request pulls all context, meaning all memories, adds it to context to each message so that when the AI is generating a response, it knows more about me, what I want, and will generate better results.
18:57Please let me know if you have any questions before implementing this feature because ideally, we're going to update the sidebar to include memory. We're gonna have a new memory page, and that memory page is gonna have a table to where I can click new and add a text memory. I should also be able to edit and delete previous memories.
19:13Please let me review the plan before you start implementing the code. Alright. Cool.
19:18So we're gonna let this one run for a minute because as you can see, we just gave it a ton of information. But this is how in the age of AI, this is exactly how you start to, you know, work at scale and implement features at lightning speed. So you can see what it has done so far is it's come up with a multistage phased approach to actually implement this.
19:39And you can see it's actually like, yeah, we'll make a memories table. From there, we'll make a memories page and then we'll update the chat API. The only thing I think it missed is the fact that it needs to update our actual schema.
19:49Let's close this out real fast so you can see I just wanna show you guys this in action. So, you know, because I think this is insanely powerful. But you can see we do have a schema over here.
19:59So for each model type in our database, so everything for messages, we have a new table for it. So we can actually just come back in here and say, that looks great. The only thing that you're missing is a phase zero, which is where you will update our schema to include a new memory table.
20:17But that's the only thing that's missing. I love everything else that you're doing. So please update the phases and then go off and generate that task document.
20:24Alright. So it should have done it. I'm gonna go ahead enter and then fantastic.
20:27It took in my advice. Now it's going to take in basically that piece of feedback. It's going to say, yep, I need to update your schema.
20:35I'm now gonna update the phases. Once it has updated the phases, it should create a new document for us to proceed. So you can see now it's gonna look at the task template to fully understand like, okay, cool.
20:47I know what to do. I'm gonna create the memories. And now it's just gonna take a second to generate everything.
20:52So we're gonna let it run. And then once it's ready, I will let you guys know and we'll go from there. So it's actually running real fast.
20:58I just wanna show you how you can interrupt So you can press escape at any moment, and it will actually cancel the request. And we can come back in here and adjust our terminal tool using and basically give it feedback along the way. So we can say, hey, you didn't follow proper instructions.
21:14You were supposed to create a document first in the AI docs folder using the, uh, task template. So please create that plan first so I can fully review it, and then we're going to implement it phase by phase. Okay.
21:29So now you can actually see in real time, it's going to hopefully generate a new document for us right here because it accidentally went off and started creating new files. Like, it did not let me review it, so it kinda jumped the gun. But, hey, it's trying to crank out code, so that's a, you you know, go it for trying to be super proactive on us.
21:47So we're gonna let it run and ideally, it should create a super long task document just like this, which is contains a ton of context, but it's gonna be our source of truth as we have the model generate everything for us. So we're gonna give it a few more seconds so that it can generate the plan, then we're gonna let it run the plan and I'm gonna show you guys the results.
22:06So stay tuned and I will show you in just a second. Okay. So the AI just finished updating our documents.
22:11So now what you can see is I have this nice purple color document right here. You can command click it and it will show you the final results. So you can see, hey, here's the entire plan.
22:20We're gonna implement an AI memory feature. Here's the exact tech stack that it pulled from our general task template.
22:28So right here, it pulled some of the general information it was told to look up, and it filled it in because we listed it all out here in our Gemini file. So everything's included right here. And then you can see it went through and listed out.
22:40I'll go down to the functional part. But you can see, like, this is the power of using AI to help you work faster. But you can see it says, alright.
22:46So here are the API changes we need to make, everything related to memory. From there, you can see the new front end components and pages it needs to make. And then it gives you a step by step phase approach of how it's gonna implement it.
22:58So this is where you can start to see the power of AI because now you can see I can go in here and say, great. Please implement phase one and let me review the results once you implement it and then we'll move on phase by phase. So hopefully, this is super helpful for you guys to see that a, this tool is insanely powerful.
23:15I'm just talking to my computer and Gemini CLI is doing all the heavy lifting for us. So you can see it's reviewing the code. You can see it's doing all sorts of amazing work for us.
23:26So what I'm gonna do is I'm just going to rinse and repeat this because at this point, I'm not coding anymore. AI is doing all the coding for me, and I'm gonna let it run all the way through, just not giving any more feedback and letting it generate the entire update for this feature and the segments working. I'll show you the final results over in our application.
23:43So we'll be right back. Alright, guys. It just wrapped up and this thing is a beast.
23:48So I just wanna show you exactly what it did real fast. So you can see what it did is main things is it made all the changes. As it was making changes, it was running Lint to make sure that there weren't errors.
24:00When there were errors, it fixed them. But what you can see it did, it just consistently cranked out code. Basically, like, you would expect inside a cursor.
24:08It read files. It updated files to make sure when we're chatting with AI, it gets memories first. It injects the memories as context into the system prompt.
24:17And that way, every time we you know, we're talking to the AI going forward, it's going to actually add a memory. So let me show you that it worked. And I also wanna show you real fast that it added everything we needed.
24:26So it added a memories table. It applied the database migration to my database. It created functions to get my memories so that I every time I chat, I can easily get memories.
24:37And then it also updated a ton of information. So, like, in my chat function, you can now see all this green right here. This is new code.
24:45So you can see we're gonna do some memories. So you can see right here, memories sorry. I I gotta spell.
24:52But you can see in memories, what it did is it added some context to where it shows all the content from each memory. So enough talking about the code.
24:59Let me just show you the final result. So we're over here. It made the brand new memories in the sidebar.
25:04It has a new add memory button, so I can actually come in here. I can paste in a memory so you can see, hey, make sure every response starts with Gemini CLI reporting for duty. Now I can hit save.
25:15It's going to save the new memory to the table. This is crazy. I never wrote a single line of code.
25:20And now whenever I chat where I can say, hey, exclamation mark, please tell me in one sentence who is Elon Musk.
25:31Then what's gonna happen is it's going to instant sorry. I hit a button. My bad.
25:34What it's going to do is instantly respond back and it's going to include our memory so that in every chat going forward, it actually includes it. So you can see I didn't write a single line of code. I added in a whole new feature, database change, UI change, and updated API calls, and it worked in under, like, two minutes for all the changes.
25:52That's insane. So I'm gonna give Gemini CLI a huge check when it comes to working on I mean, this is like a medium large fix. This would normally take developers a few hours to implement and it took minutes, which is insane.
26:04So now what we're gonna do next is we're gonna move over to the final code test where we're going to actually have it generate a brand new project from scratch to see just how well it can do. So let's hop over to test number three. Okay.
26:15So now, it's time to run our third test where we're gonna see if Gemini CLI can do a great job of creating a brand new project from scratch and just see how well it does so that you can know if you can trust this tool when you're creating your own project. So let's go ahead and dive in. So first things first, I've created a brand new empty folder.
26:34The only thing I have in here is an environment variable folder with my Gemini API keys. The reason why is I blown through all of my free credits because I've been testing this like a madman for you guys. And at this point, I am paying for it.
26:46However, using everything that you've seen for the previous task and a bunch of things you'd have not seen when I was testing, I'm currently paying 3 pennies. So this is insanely affordable to use. So I just wanna give you a heads up.
26:58If you do run out tokens, it's not gonna break the bank. So let's go ahead and run it. And what I wanna do, I'm just gonna go ahead and paste in a huge prompt for you guys.
27:05So let me just give you a, uh, paste. So What you can see is I was like, hey, please create a brand new landing page for me. I want to use Next.
27:13J s, and I want this website to be beautiful and modern. And most importantly, what I need you to do is look up the YouTube Creator AI with Brandon. Obviously, that's who you're watching right now.
27:24Like and subscribe if you haven't. But research him, find everything you can about what he does, and basically create a landing page. So what we're gonna do is we are going to run this.
27:33And what this should do is it should kick everything off. So it's so far, it's, you know, clarifying the approach. It's going to hopefully, here in just a second.
27:41Yep. There it goes. It's going to research AI with Brandon.
27:44It's gonna look up my YouTube channel, analyze everything about what I do, the content I create, who I'm trying to help, which is AI developers just like you to help you build real world AI applications. So it's gonna analyze everything about that. Then what it's hopefully gonna do next is it's going to start planning out some content.
28:01So, like, I now understand who Brandon is, who he serves. So let's start creating some copy for the landing page. So fingers crossed, it should start doing this in just a second.
28:11Oh my gosh. It is crushing it. It's doing a great job.
28:15So, yeah, this is all around looking fantastic. So what I wanna do is say proceed, and this should start to create the Next.
28:23Js project with everything that you see right here, which is a a nice hero section, about section, what you'll learn, some featured videos. So alright. It does a great job.
28:31So we're just gonna say proceed. Now it should start creating the project. It's going to start planning it all out.
28:38I'm gonna create a Next. Js project. So here's one thing that is important.
28:42I have noticed that when you run commands like this, create NextJS application, where it opens up a wizard towards, like, you run the command and then it asks you six questions of, like, do you wanna do a or b, c or d? It doesn't do well. So what I'm going to do is I'm just gonna copy this command.
28:59I'm going to press escape. I'm going to you can actually run it. So you can do you can hit the exclamation mark, and then you can paste in your own code.
29:08So that's exactly what I'm gonna do. Then it should actually run it. What we're gonna do real fast instead because it is struggling, I'm just gonna run the command manually.
29:15So we're gonna paste it. We're gonna run it. Now this should create a NextJS project for us.
29:20And I'm just gonna hit no. I'm just gonna hit all the defaults and it should create the project.
29:26Fantastic. So now we can get out of this the second it's done. So we're having to help the AI just a little bit, but I feel like that's fair because it's gonna build everything else for us.
29:34So we're gonna let it run. And then the second this is done installing all the dependencies, we're going to tell this AI over here.
29:42Yep. Alright. Great.
29:43We're done. So now we can kill this one. And now we can say, I actually ran the command for you so you have the proper Next.
29:49Js project created. So please look at the project and then proceed with the plan. Fantastic.
29:54So then now Gemini CLI should go off and actually update this entire project for us and start to create the landing page. It should hopefully clear out all the blank boilerplate code that we normally have. So you can see it understands the project structure and now it's just gonna go off to the races to make all the changes.
30:13So I'm just going to allow it to make every change that it wants. And in real time, you can see as it's making changes to files, it's going for it. So it's updating the color scheme right now and it's gonna keep on cruising.
30:23So I'm gonna hop back once it's done running because it usually takes about two to three minutes for big changes like this and then I'll show you guys the final result. So it just wrapped up and I'm not gonna lie, the results are pretty underwhelming for creating a new project right out the gate. So I'm curious if it just needs another round.
30:39What So I like to do, we'll give it one extra round of help. So we'll take a screenshot, then what we can do is pass it in over here and we can say and basically just give it feedback.
30:50So we can say at screenshot, we'll paste it in, and then now we'll give it more instructions to say, this website was supposed to be beautiful and modern and be really visually appealing.
31:02So you need to be a professional UI and UX developer and completely redo the landing page so it's absolutely beautiful.
31:12And look at the screenshot I just gave you so you can see the application in its current state. Do not stop until the website is absolutely beautiful. So now we're gonna give it another shot and hopefully this time it will just go off and do exactly what it was supposed to do the first time because this is not what it was supposed to do.
31:32So we're hopefully gonna let it do it. And what's so nice is like it is multimodal. So, like, it can take in images just like that.
31:40You can see it's like, yeah, this kind of this was a very polite way to say, like, yeah, that was bad. So it's gonna update it to have more colors, better typography, better component redesign, and some animations and transitions.
31:52So we're gonna give it a second because now it's actually updating the global CSS to have, like, more colors, be much more visually appealing. So we're just gonna give it one more attempt so we can hopefully see that it can actually create beautiful websites. And hey, this is just the hard part about working with AI is it's having to assume things and build at the same time.
32:11So it probably was just trying to do too much at the same time. So we're gonna give it a break and we'll come right back. Okay.
32:15So it just finished building everything. Not gonna lie, pretty disappointed. So if we hop over to the website, even though we told it multiple times like, hey, make this visually appealing.
32:25It did do a bunch of nice cool animations. But however, it just really like, this is not like I was expecting like the lovable version of like a beautiful landing page. And for whatever reason, it just kind of, you know, doing a bunch of animations, but not following like, you know, making it look super, super nice in a UI.
32:45So, hey, maybe just the issue is I'm not giving it enough prompting of, hey. Here's a example UI that you should go off of.
32:52However, when working with, like, a lovable or, you know, a Firebase Studio, this feature always dominates. So I was I had super high hopes for creating brand new Next. Js projects that were super visually appealing, but for whatever reason, it seems like it just can't do it.
33:07And maybe I just should have given a better, you know, better version of, a Gemini dot m d file, but I know a lot of you are just gonna wanna come in here and say, make a new project and just expect it to just crush it like I did, and it wasn't doing it. So all around, a little disappointed with hard test number three where we create a brand new project from scratch.
33:26So I'm gonna give this one a fail. However, as you did see in the previous project where we had a ton of existing codes where it could look through and understand what it needed to do.
33:37Basically, what the main takeaway I'm trying to say is , and I feel like without context, it's struggling. So I think if on my recommendation to you guys when you're working on this on your own, it would be to use Gemini CLI on your existing projects to add in new features and to add small changes.
33:57This is an amazing tool for that. I would not recommend it right now for creating brand new projects. And hey, if you have any ideas or suggestions on what I did wrong in this example, I'd love to hear and try it again.
34:08But for right now, it's just not doing what I thought it would. But So yeah. But that's a wrap for this video, guys.
34:13I hope you enjoyed seeing Gemini CLI in action. You were able to set up in this tutorial, learn how to set up your project, and you were able to see just how powerful this tool is, and you were able to see that, hey, you could do pretty much all this completely for free. And if you have to go over the limit, it cost a few pennies to implement some amazing features.
34:31So all around, I had a ton of fun using this. I can't wait to hear what you guys think of using this project as well. And I can't wait to see you guys in the next video.
34:40I have a ton of other AI related content right here on this channel. Everything from agent development kit, LangChain, CrewAI, Next. Js.
34:47I have it all right here, and I definitely recommend checking out those videos and whichever video is popping up right now on the screen. But until the next time, can't wait to see you guys. Hope you have a great night.
34:55Talk to you soon. Bye.
The Hook
The bait, then the rug-pull.
Brandon Hancock opens face-cam in front of an American flag and a bookshelf and lands the same three nails Google itself led the launch with: direct competitor to Claude Code, insanely powerful, completely free. Then he promises the only thing a developer actually wants — three real coding tests, no demos, watch it succeed and fail in real time.
Named ideas worth stealing.
16:54model
Three-Test Verdict Format
- Simple test (1-line fix) — proves baseline reliability
- Medium test (multi-file feature with DB) — proves real workflow value
- Hard test (greenfield project) — finds the breaking point
Brandon's review structure: escalating-difficulty live coding tests, each with an explicit pass/fail call. Lets the viewer make their own verdict without trusting his opinion.
Steal forAny AI tool review video — beats 'first impressions' fluff by 10x because every viewer wants to know where the tool breaks
16:00model
AI-Driven Task Development Loop
- Maintain a task_template.md (project shape, tech stack, conventions)
- Maintain a GEMINI.md (project rules + memory, updated continuously)
- For each feature: take screenshot + reference task_template.md, ask Gemini for a multi-phase plan first
- Review the plan, request edits (e.g. 'add a Phase 0 for schema changes')
- Have Gemini generate a long-form task doc in /ai_docs/
- Execute phase by phase, review between phases
- When Gemini makes a mistake, add a rule to GEMINI.md so it never repeats
Brandon's whole AI coding workflow, taught in passing inside the Case 2 demo. The pattern (template + context file + plan-then-execute) is portable across any agentic CLI.
Steal forJoeFlow / MCN / any project where Joe wants Claude Code to ship a multi-file feature without supervision — this is the cleanest version of the loop in the wild
13:20concept
Update Memory pattern
When Gemini makes a mistake, say the correction out loud and add the words 'update memory'. Gemini writes the rule into GEMINI.md automatically. Use /memory show to view all current rules.
Steal forTeach end-users how to train their agent in real time — same pattern works for Claude Code's CLAUDE.md
05:32concept
Free-Tier Value Math
Don't just list specs — convert them to dollars-per-day. 60 req/min × 1M-context Gemini 2.5 Pro at $3/req = $180/day in free tokens. Concrete dollar value kills the 'is this real?' skepticism in one sentence.
Steal forAny pricing or free-tier marketing page — translate units into dollars the reader would otherwise spend
How they asked for the click.
VERBAL ASK
34:16subscribe
“I have a ton of other AI related content right here on this channel. Everything from agent development kit, LangChain, CrewAI, Next. Js. I have it all right here, and I definitely recommend checking out those videos and whichever video is popping up right now on the screen.”
Soft channel CTA at the end — no hard ask for likes, no link drop, just 'next video is on screen.' Honest video earns the trust to skip the hard sell. The real sales pitch is the ShipKit.ai mention woven into Case 2 as the codebase being demoed.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
