How to Build an AI Agent with Claude Code (Claude AI Agent Tutorial) — Transcript
Full transcript
- 0:00Most people are still using Claude like
- 0:02a fancy search engine. Ask a question,
- 0:04copy the answer, ask the next one.
- 0:06There's a way smarter way to work with
- 0:09it. And once you see it, you can't go
- 0:11back. Let me show you what an AI agent
- 0:13actually is and how to build one inside
- 0:16Claude without writing a single line of
- 0:19code. Before we build anything, I want
- 0:21to get clear on one thing because the
- 0:23term AI agent gets thrown around
- 0:25constantly and most explanations are
- 0:27either too technical or too vague to be
- 0:30useful. Here's how I think about it.
- 0:32There are three distinct levels of
- 0:34working with AI and most people are
- 0:36stuck at level one without realizing
- 0:39levels two and three even exist. Level
- 0:41one is basic chat. You ask something,
- 0:44Claude answers. You need a definition, a
- 0:46quick summary, an explanation. You type
- 0:49it in, you get it back. This is
- 0:50genuinely useful, but it's also the most
- 0:53limited possible way to work with the
- 0:55technology. You're essentially using a
- 0:57very sophisticated search engine. Level
- 0:59two is what I call builder mode. You
- 1:02stop asking questions and start asking
- 1:04Claude to make things. Write me this
- 1:06script, draft this email, analyze this
- 1:09document. Now you're getting real
- 1:11output, not just information. But the
- 1:13catch is that you're still managing
- 1:15every single step yourself. You decide
- 1:18what comes next, you review each output,
- 1:20you carry the whole project forward
- 1:22manually. Level three is agentic work.
- 1:25This is where the real shift happens.
- 1:27Instead of managing every step, you hand
- 1:30Claude a goal, a real complete outcome
- 1:33and it figures out how to get there. It
- 1:35thinks through the task, breaks it into
- 1:37phases, asks questions if something is
- 1:39unclear, executes each phase in
- 1:42sequence, reviews its own work, and
- 1:44delivers a finished result. You're not
- 1:46in the loop for every micro decision.
- 1:48You set the destination and the agent
- 1:50drives. And what makes Claude
- 1:52particularly powerful for this right now
- 1:54is that you don't need any developer
- 1:56setup to get started. The Claude desktop
- 1:58app already has a built-in code
- 2:00workspace where you can create projects,
- 2:03manage files, and run agentic workflows
- 2:05without touching a single line of actual
- 2:08code. That used to require serious
- 2:10technical infrastructure. Now, it's a
- 2:12desktop app. If you're running a
- 2:14business and juggling 10 different tools
- 2:16right now, what if I told you that one
- 2:18platform could replace all of them? This
- 2:20is GoHighLevel. You get access to pretty
- 2:23much all the features. Social media
- 2:25planner, you can schedule posts, see
- 2:27your content calendar, and even repost
- 2:29content across multiple platforms with
- 2:31one click. Email marketing, GoHighLevel
- 2:34has a full email suite. You can create
- 2:36campaigns, build email templates with a
- 2:39drag-and-drop editor. GoHighLevel has a
- 2:41built-in calendar and booking system.
- 2:43This replaces tools like Calendly. You
- 2:46can create different calendar types,
- 2:48one-on-one meetings, group calls,
- 2:50round-robin scheduling if you have a
- 2:52team. And here's the best part.
- 2:54GoHighLevel has a ton of pre-built
- 2:57workflow templates. I've linked 30-day
- 2:59free trial in the description, not the
- 3:01standard 14 days, an exclusive extended
- 3:04trial. Not every Claude conversation is
- 3:07actually agentic. There's a real
- 3:08difference between a sophisticated
- 3:10prompt and a genuine agent workflow, and
- 3:12it comes down to three things. The first
- 3:14thing is that a real agent follows a
- 3:16process, not just responds to a message.
- 3:19A chatbot gives you one reply per
- 3:21message. An agent moves through stages.
- 3:24It gathers context, forms a plan,
- 3:26executes the plan step by step, reviews
- 3:29its own output, and adjusts if something
- 3:31isn't working. It's operating through a
- 3:33system. The second thing is that a real
- 3:35agent makes decisions under uncertainty.
- 3:38A chatbot will guess if it doesn't
- 3:40understand your request. A real agent
- 3:42stops and asks. It adapts its approach
- 3:44based on what it finds. If the initial
- 3:46direction turns out to be wrong, it
- 3:48corrects course rather than pushing
- 3:50forward with a bad assumption. That
- 3:52ability to reason through ambiguity,
- 3:54rather than just pattern matching to the
- 3:56most probable answer, is what separates
- 3:59an agent from a fancy auto complete. The
- 4:01third thing is clarification before
- 4:03execution. This one is probably the most
- 4:06underrated. The vast majority of bad AI
- 4:08output happens because the model
- 4:10misunderstood what you actually wanted.
- 4:13It assumed an audience, a format, a
- 4:15scope, a tone. A well-designed agent
- 4:18asks before it acts. It treats your
- 4:20initial request as a starting point for
- 4:23a brief conversation, not as a complete
- 4:25specification to execute immediately.
- 4:27That clarification step is where most of
- 4:30the quality lives. All right, let's
- 4:32build something. Open the Claude desktop
- 4:34app and go to the code workspace. If you
- 4:37haven't downloaded it yet, it's
- 4:38available directly from Anthropic. Just
- 4:40search Claude desktop app and it'll be
- 4:42the first result. Once you're in the
- 4:44code workspace, create a new project
- 4:47folder. You can name it anything. I'll
- 4:48call mine agent workspace for this demo.
- 4:51This folder is your AI's operating
- 4:53environment. Think of it as the office
- 4:55your agent works in. Everything it
- 4:57creates, organizes, and references lives
- 4:59here. Right now, it's completely empty,
- 5:01and that's totally fine for what we're
- 5:03about to do. The first thing we need to
- 5:05add is the most important file in the
- 5:07entire setup. It's called Claude.md, and
- 5:10I genuinely believe most people who
- 5:12struggle with AI agents are struggling
- 5:15because they skip this file entirely.
- 5:17And creating it is genuinely the easiest
- 5:20part of this whole setup. You don't open
- 5:22a text editor. You don't mess with the
- 5:24file system. You just tell Claude to do
- 5:26it. Type create a file called Claude.md
- 5:30in the root of this project right into
- 5:32the chat, and it'll spin up the file for
- 5:34you in seconds. That's it. Now, we just
- 5:36need to fill it in. Here's the problem
- 5:38with most agent setups. People spend
- 5:40hours designing elaborate workflows,
- 5:43then launch them without giving Claude
- 5:45any context about who they are, what
- 5:47they're trying to accomplish, or how
- 5:48they want things done. And then they
- 5:50wonder why the output feels generic.
- 5:53Claude.md solves this. It's a plain text
- 5:56file written in markdown. Just a simple
- 5:58formatting system that Claude
- 6:00automatically reads whenever it starts
- 6:02working inside your project. Think of it
- 6:04as an onboarding document for your AI.
- 6:06Except instead of onboarding a new hire
- 6:09once, you write it once and Claude reads
- 6:11it every single session in your
- 6:12workspace. Let me walk you through what
- 6:15goes into a good Claude.md file because
- 6:18the structure matters. Start with a
- 6:20project context section. Two or three
- 6:23sentences about what this workspace is
- 6:25for. Something like, "This workspace is
- 6:27used for AI-assisted research, content
- 6:30creation, and workflow automation.
- 6:32[music] Outputs are primarily for
- 6:34non-technical audiences who want
- 6:36practical, actionable information about
- 6:38AI tools." Next, add an about me
- 6:41section. Tell Claude who you are and how
- 6:43you work. your background? Who's your
- 6:45audience? What tone do you prefer? For
- 6:47example, I create practical AI content
- 6:50for everyday users. I prefer concise,
- 6:53direct explanations over academic or
- 6:55jargon-heavy language. Every output
- 6:57should feel like advice from a
- 6:59knowledgeable friend, not a corporate
- 7:01report. Then write a rules section. This
- 7:03is where you encode your working
- 7:05preferences so Claude applies them
- 7:07automatically every time. Some rules I
- 7:09use in my own setup. Always ask at least
- 7:12three clarifying questions before
- 7:14starting any complex task. Always show
- 7:17your plan before you execute it. Keep
- 7:19every written output at or under the
- 7:21requested length. Never pad. When saving
- 7:24files, use lowercase names with hyphens,
- 7:27not spaces. These might sound like small
- 7:29things, but when they're encoded in the
- 7:31file, you stop having to re-explain them
- 7:33every session. Finally, define your
- 7:36folder structure. Something like a
- 7:38workflows folder for your instruction
- 7:40files, an output folder for completed
- 7:42work, and a resources folder for
- 7:44reference material you want Claude to
- 7:46draw from. This gives the agent a clear
- 7:49map of where things live and where to
- 7:51put things when it's done. Once you save
- 7:53this file, Claude's behavior inside your
- 7:56project changes immediately. It's no
- 7:58longer working from scratch every
- 8:00session. It has context, rules, and a
- 8:02structure to operate within. That one
- 8:05file, done well, is worth more than most
- 8:08of the prompt engineering tricks you'll
- 8:09find anywhere online. Before we create
- 8:12our first real workflow, I want to talk
- 8:13about one habit that makes or breaks AI
- 8:16agent work. It's called planning mode,
- 8:18and the idea is simple. You never let
- 8:20the agent execute immediately. Here's
- 8:23why this matters. When an AI agent
- 8:25starts operating autonomously, it
- 8:28commits to a direction. It creates
- 8:30files, structures content, makes
- 8:32assumptions, and if any of those initial
- 8:34assumptions are wrong, which happens
- 8:36more often than you'd expect, you've
- 8:38wasted time and you're now correcting a
- 8:40half-finished project instead of
- 8:41steering a clean one. Planning mode
- 8:44means you always ask Claude to show you
- 8:46the plan before it does anything. You
- 8:48want to see how it interpreted your
- 8:50goal, what steps it intends to take,
- 8:52what files it plans to create, and what
- 8:54questions it still has. That review
- 8:57takes maybe 2 minutes, but it routinely
- 8:59saves 10 minutes of cleanup on the back
- 9:01[music] end. The simplest way to build
- 9:03this into every workflow is to add one
- 9:06line in your
- 9:08section, "Always present a written plan
- 9:11and wait for approval before beginning
- 9:13any multi-step task." Once that's there,
- 9:16Claude defaults to showing you its
- 9:18thinking before it acts. You review, you
- 9:21adjust, you approve, then it executes.
- 9:24Quick pause before we start building. If
- 9:27what you just heard about agents has you
- 9:29thinking, "Okay, I actually want to get
- 9:31good at this," we built a place exactly
- 9:33for that. It's called AI Master, and
- 9:35there are two things inside that work
- 9:37together. First, the course. It's called
- 9:39Generative AI Essentials, and it's a
- 9:41hands-on encyclopedia of prompt
- 9:42engineering and automation. Modules on
- 9:44system prompts, multimodal work,
- 9:46marketing, data analysis, real
- 9:48workflows, not theory. Second, our
- 9:50studio. Once you've learned something in
- 9:51the course, you go practice it in the
- 9:53same platform on every major model.
- 9:55ChatGPT, Claude, Gemini, DeepSeek, Grok,
- 9:58plus Seedance and Kling for video and
- 10:00Nano Banana for images. One [music] tab,
- 10:01no juggling 10 subscriptions. Links in
- 10:04the description. And honestly, the
- 10:05annual plan is the smartest move, up to
- 10:0742% off the platform and every
- 10:10generation you run inside. Okay, back to
- 10:12building. Now, let's look at the
- 10:14underlying architecture of what we're
- 10:16building because understanding the
- 10:18structure makes you much better at
- 10:20designing your own workflows later. Most
- 10:23agentic systems have three components,
- 10:25and they work together in a specific
- 10:27way. The first component is the workflow
- 10:30file. This is a written document,
- 10:32usually a markdown file, that describes
- 10:35process the agent should follow for a
- 10:37specific type of task. It covers the
- 10:39goal, the steps, the rules, the expected
- 10:42output format, and what to do if
- 10:44something goes wrong. Think of it as a
- 10:46standard operating procedure written in
- 10:48plain English. You're not writing any
- 10:50code here, you're really just writing
- 10:52out a process in plain words. The second
- 10:54component is the agent itself, Claude.
- 10:57Claude reads your workflow file,
- 10:59understands the objective, and acts as
- 11:02the coordinator. It decides what to do
- 11:04at each step, what questions to ask,
- 11:06when to proceed, and when to pause for
- 11:08input. It functions like a project
- 11:11manager who's read the SOP and is now
- 11:14running the process. The third component
- 11:16is the tool set, the actual capabilities
- 11:18Claude can use inside the code
- 11:20workspace, reading and creating files,
- 11:23organizing folders, searching through
- 11:25documents, analyzing and editing text.
- 11:27For most beginner and intermediate
- 11:29workflows, these built-in capabilities
- 11:31are more than [music] enough. You don't
- 11:33need external integrations or APIs to
- 11:35build something genuinely useful. Here's
- 11:38the key insight, and this one runs
- 11:39counter to what most people expect. The
- 11:42most important component is the workflow
- 11:45file, not the tools. A thoughtfully
- 11:47designed workflow with basic tools will
- 11:50outperform a poorly designed workflow
- 11:52with sophisticated integrations every
- 11:54single time. The intelligence is in the
- 11:57instructions, not the technology stack.
- 11:59That's actually good news because
- 12:01instructions are something anyone can
- 12:03write. All right, let's build something
- 12:05real. We're going to create a research
- 12:08agent, a system where you give Claude a
- 12:10topic and it handles the entire research
- 12:12and report writing process from start to
- 12:15finish. Open a new conversation inside
- 12:17your Claude code workspace. Don't give
- 12:20it a topic yet. Instead, tell it what
- 12:22kind of system you want to build. Here's
- 12:24the kind of prompt that works well for
- 12:26this. I want to design a research
- 12:28workflow. When I give you a topic, you
- 12:31should first ask me clarifying questions
- 12:33about scope, audience, and desired
- 12:35depth. Then, form a written plan. Once I
- 12:38approve the plan, you research the topic
- 12:40thoroughly, organize your findings into
- 12:42clear sections, and save a structured
- 12:44report to the output folder. Before we
- 12:46build this, show me your plan for the
- 12:48workflow itself. Notice what we're doing
- 12:50here. We're not jumping straight to the
- 12:52task. We're designing the system that
- 12:54will handle the task. That extra step is
- 12:57what turns a one-time response into a
- 12:59repeatable workflow you can run over and
- 13:02over with different topics. Claude will
- 13:05come back with a proposed structure for
- 13:07the workflow. It might suggest things
- 13:09like a clarification phase with specific
- 13:12questions that will always ask, a
- 13:14research phase broken into subtopics, a
- 13:16synthesis step where it connects
- 13:18findings across sources, and a review
- 13:20step before saving the final file.
- 13:23That's a real agent workflow taking
- 13:25shape, built in plain English, requiring
- 13:28zero code. Once the structure looks
- 13:30right to you, tell Claude to go ahead
- 13:32and write the actual workflow file.
- 13:34It'll create something like research
- 13:36agent.md
- 13:38inside your workflows folder. When you
- 13:40open that file, you're reading the
- 13:42operating instructions for your agent,
- 13:44written in plain language, fully
- 13:46editable, and reusable every time you
- 13:49want a research report on a new topic.
- 13:51Now, let's put the workflow to work.
- 13:53Open a new session inside the same
- 13:55workspace and tell Claude you want to
- 13:57run the research workflow on a specific
- 13:59topic. I'll use the example I actually
- 14:01ran recently. I want a research report
- 14:04on the current state of AI agents in
- 14:062026. What's working in practice, what's
- 14:09over hyped, and where is the industry
- 14:11actually heading? If your workflow is
- 14:13set up correctly, Claude should not
- 14:15immediately start writing a report. The
- 14:17first thing it does is ask questions.
- 14:20Answer those questions specifically.
- 14:22Don't just say make it good. Something
- 14:24like audience is non-technical
- 14:26professionals who are curious about AI
- 14:28but don't have a dev background. Keep
- 14:30the tone conversational. Focus on tools
- 14:33and workflows people are actually using
- 14:35today, not theoretical frameworks.
- 14:37Target length around 1,500 words. Claude
- 14:41then forms a written plan. Here are the
- 14:43sections I'll cover. Here's the
- 14:44structure. Here's what I'll prioritize.
- 14:47You review it, maybe adjust one or two
- 14:49things, then you approve it, and then
- 14:50the agent runs. What happens next is
- 14:53genuinely different from a normal chat
- 14:55experience. Claude works through the
- 14:57phases of the workflow sequentially. It
- 14:59researches, synthesizes, writes,
- 15:02structures the output, and saves the
- 15:04report to your output folder as a
- 15:06completed file. You're not managing any
- 15:08of those steps. You approve the plan,
- 15:10and now it's executing. That moment when
- 15:13you watch the agent move through its
- 15:15workflow independently is when it stops
- 15:18feeling like a chatbot and starts
- 15:19feeling like something that actually
- 15:21works for you. Here's where a lot of
- 15:23people don't realize how powerful
- 15:25persistent project context actually is.
- 15:27Once your agent produces an output, you
- 15:30don't restart from zero to make changes.
- 15:32Claude already has full context of what
- 15:35it just created, so refinements are
- 15:37fast. Say the executive summary in your
- 15:39report runs a bit long. You just say,
- 15:42"Trim the executive summary. I want it
- 15:44down to three key points." Claude
- 15:46updates that section without touching
- 15:48the rest of the document. Or maybe you
- 15:49want to add something. "Add a short
- 15:51section at the end comparing the top
- 15:53three AI agent tools by use case."
- 15:56Claude goes back into the document,
- 15:58references what it already wrote, and
- 16:00adds the new section in a way that fits
- 16:02the existing structure and tone. This is
- 16:04the compounding effect of working inside
- 16:07a persistent workspace. Each session
- 16:09builds on the last. The context
- 16:12accumulates. Instead of starting over
- 16:14every time you need something new,
- 16:16you're refining and expanding a living
- 16:18body of work. That's a fundamentally
- 16:21different relationship with AI than the
- 16:23copy-paste loop most people are still
- 16:25stuck in. Okay, so the research agent
- 16:27was useful, but let's build something
- 16:29that genuinely saves you time every
- 16:32single week. We're making a repurposing
- 16:34agent. You feed it one long script, and
- 16:36it spits out three short scripts and a
- 16:39full social media pack, each one as a
- 16:41separate PDF sitting in a folder ready
- 16:43to publish. Here's what we're actually
- 16:45building. You drop your main script into
- 16:48an input folder. You give the agent one
- 16:50instruction. It reads the whole thing,
- 16:52picks the three strongest moments, and
- 16:54turns each one into its own 60-second
- 16:57short script with a fresh hook. Then it
- 16:59writes a social pack, an X thread, a
- 17:01LinkedIn post, an Instagram caption, and
- 17:04finally it converts every artifact into
- 17:06a clean PDF and drops them all into an
- 17:08output folder. Now, let's write the
- 17:10system prompt. Same approach as before.
- 17:13We describe the role, the inputs, the
- 17:15steps, and the output format. The role
- 17:17is a content repurposing specialist. The
- 17:20input is a single long-form script in
- 17:22the input folder. The steps are: read
- 17:25the script, [music] identify three
- 17:27standalone moments that work as shorts,
- 17:29write each one with its own hook and
- 17:31payoff, then draft platform-specific
- 17:33social posts that match the host's
- 17:35voice. The output is four markdown files
- 17:38converted to PDF, saved to the output
- 17:40folder. I'll run it live. One prompt and
- 17:43you'll see the files appear in the
- 17:45folder one after another. Three shorts
- 17:47PDFs, one social pack PDF, from a single
- 17:51script in under 2 minutes. I've spent a
- 17:53lot of time running these workflows and
- 17:55watching other people try to get started
- 17:57with them, and the failures almost
- 17:59always come from the same handful of
- 18:01mistakes. Let me walk you through them
- 18:03so you can skip the frustrating part.
- 18:05Mistake one is skipping Claude.md.
- 18:08Without that file, Claude has no
- 18:10persistent context about who you are,
- 18:12how you like to work, or what standards
- 18:14your outputs need to meet. Every session
- 18:17starts from zero. The quality is
- 18:19inconsistent and you spend more time
- 18:21correcting output than you save by using
- 18:23an agent in the first place. Write the
- 18:25file before you build anything else.
- 18:27Mistake two is using vague goals.
- 18:30Telling Claude to do some research on AI
- 18:32will produce something that feels like a
- 18:34Wikipedia article. Telling it to
- 18:35research how non-technical professionals
- 18:38in marketing are using AI agents for
- 18:40content production, focus on tools with
- 18:42free tiers, and target an audience
- 18:44that's never written a line of code,
- 18:46produces something you can actually use.
- 18:48The specificity of your goal determines
- 18:50the quality of the output. This is true
- 18:52at every level of AI work, but it
- 18:54matters most in agent workflows because
- 18:57the agent is running multiple steps
- 18:59based on that initial goal. Mistake
- 19:01three is skipping the plan review. It
- 19:03feels like an extra step and the
- 19:05temptation is to just let it run and fix
- 19:07things afterward, but agents that run
- 19:10without a reviewed plan can go
- 19:12significantly off track. Correcting a
- 19:14half-finished multi-stage workflow is
- 19:16much harder than adjusting a plan before
- 19:18execution starts. Two minutes of review
- 19:21up front saves 10 minutes of cleanup.
- 19:23Always approve the plan first. Mistake
- 19:26four is not requiring clarifying
- 19:28questions. If your workflow file doesn't
- 19:30explicitly tell Claude to ask questions
- 19:33before starting, it will assume, and
- 19:35those assumptions are the source of most
- 19:37mediocre AI output. Build the
- 19:39clarification step into every workflow
- 19:41you create. The agent should always
- 19:43understand your specific requirements
- 19:45before it starts working, not after.
- 19:47Mistake five is trying to build
- 19:49everything at once. People get excited
- 19:51about agents and want to build a
- 19:53research system, a content planner, a
- 19:55CRM workflow, an email responder, and an
- 19:58analytics dashboard all in the same
- 20:00week. The result is five half-working
- 20:02systems that none of them trust or use.
- 20:05Start with one workflow. Run it until
- 20:08it's reliable and you understand how it
- 20:10behaves. Then build the next one. Once
- 20:12you have one or two workflows running
- 20:15well, the natural question is, "What's
- 20:17next?" Here's a practical progression
- 20:19for how to expand your agent stack
- 20:22without overwhelming yourself. The
- 20:23progression that works best starts with
- 20:26simple self-contained research or
- 20:28writing tasks. Things with a clear input
- 20:31and a clear output format. A research
- 20:33report, a content outline, a document
- 20:36summary. These are low-stakes,
- 20:38high-feedback tasks where you can see
- 20:40immediately whether the agent performed
- 20:42well and where the cost of a bad output
- 20:45is minimal. From there, you can move to
- 20:47more iterative workflows. Ones where the
- 20:49agent produces a first draft and then
- 20:51refines it based on your feedback across
- 20:54multiple rounds. Script writing works
- 20:56well here. So does editing and
- 20:58improvement workflows, where you give
- 21:00Claude a rough draft and it applies
- 21:02specific criteria to improve it. The
- 21:04more advanced level is multi-workflow
- 21:07systems where one workflow triggers or
- 21:09informs another. A research agent
- 21:12produces a findings report and then a
- 21:13content planning agent reads that report
- 21:16and generates video ideas based on what
- 21:18the research identified. That's a real
- 21:21pipeline. It's not as complicated to
- 21:23build as it sounds. Once your individual
- 21:25workflows are solid, the thing I want
- 21:28you to take away from this progression
- 21:29is that agent capability isn't something
- 21:32you unlock all at once. You build it
- 21:34incrementally one reliable workflow at a
- 21:36time and the system gets more powerful
- 21:39as each component gets more refined.
- 21:41That's your first two working agents,
- 21:43one for research, one for repurposing.
- 21:45If you actually build even one of them
- 21:47this week, you're already ahead of 90%
- 21:50of people who just talk about agents.
- 21:52For a deeper dive on building agents
- 21:54from scratch, I've got another video
- 21:56linked up here. And if you want to
- 21:58actually learn this properly instead of
- 22:00just watching tutorials, that's what we
- 22:02built AI Master for. Links in the
- 22:04description.
About this transcript
This page contains the full transcript of How to Build an AI Agent with Claude Code (Claude AI Agent Tutorial) by AI Master, generated from the public captions YouTube serves with the video. The transcript has 3,776 words across 594 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.