Eight YouTube talks and playlists on Large Language Models, agentic AI, and building agents, selected for engineers who’d rather watch a 30-minute lecture than skim a 30-thread thread. The list runs from a high-level Stanford overview to a multi-hour “LLMs from Scratch” deep-dive, with stops for evaluation, MCP, and live-coded agent builds. If you learn better from a lecturer than from a notebook, this is your entry into the rest of the Learning Hub.
Why these eight videos for learning LLMs
The YouTube surface for LLM content is enormous and largely low-signal. This list filters to eight videos that hit one of three bars: a canonical introduction (the talk you’d send a new hire), a hands-on build (someone coding an agent live, not just diagramming it), or a playlist with serious depth (multiple hours, coherent arc). TED-style overview talks were excluded; vendor product demos were excluded.
Pair this roundup with the Papers roundup if you want the conceptual layer underneath these talks, or with the Courses roundup if you want a graded, structured version of the same content.
1. LLM Introduction
A ground-floor introduction to what a large language model is, how it’s trained, and what it actually does when you send it a prompt. The right starting point if you’ve used ChatGPT but never opened a model card. Free on YouTube.
The framing is non-technical enough for product managers and designers, but precise enough that an engineer won’t roll their eyes. Watch this first if you’re new to the field; skip it if you’ve already read the original Transformer paper.
Why it matters: the cleanest free ground-truth intro. Saves you from learning the field through Twitter.
↔ Watch LLM Introduction on YouTube
2. LLMs from Scratch
A long-form code-heavy walkthrough of building a small language model from the ground up: tokenization, embedding, attention, and the training loop. Not a math lecture with code snippets; a code lecture with math footnotes. The pace is fast but unhurried.
If you’ve ever finished an intro talk and thought “ok but how do I actually build one,” this is the next click. The same author also published a full book version (covered in our Books roundup); the video is the abridged, free version.
Why it matters: the fastest path from “I use LLMs” to “I could train one” without signing up for a course.
↔ Watch LLMs from Scratch on YouTube
3. Agentic AI Overview (Stanford)
A Stanford lecture giving the academic framing of what “agentic AI” means: the loop, the tools, the planning, the memory. The vocabulary is precise, which matters because most industry usage of “agent” is loose to the point of meaningless.
Stanford’s CS lecture series has a track record of producing the canonical talk on emerging topics; this one fits the pattern. Watch at 1.25× if you’re already familiar with the basics; the slides are dense.
Why it matters: the conceptual scaffold that turns “agent” from a buzzword into a defined system with components you can reason about.
↔ Watch the Stanford Agentic AI Overview on YouTube
4. Building and Evaluating Agents
A talk that treats evaluation as a first-class concern rather than an afterthought. Most agent tutorials end at “it works in the demo”; this one keeps going into how you’d know if it actually works in production, trace eval, tool-call eval, end-to-end eval.
If you’re shipping an agent into a real product, this is the talk to watch before you write your eval suite. It pairs well with the DeepLearning.AI “Evaluating AI Agents” course in our Courses roundup.
Why it matters: the talk that moves you from “agent demos” to “agent in production.” Eval is the gap; this is the bridge.
↔ Watch Building and Evaluating Agents on YouTube
5. Building Effective Agents
The video companion to Anthropic’s widely-cited “Building Effective Agents” essay (linked in our Guides roundup). Walks through the same patterns: augmentation vs. workflow vs. agent, when to use each, where the complexity lives, with diagrams and code sketches.
If you read the essay and wanted the authorial voice walking you through it, this is the version to watch. If you didn’t read the essay, read the essay first; the video assumes you have.
Why it matters: the agent design taxonomy that the rest of the industry has quietly converged on.
↔ Watch Building Effective Agents on YouTube
6. Building Agents with MCP
A talk centered on the Model Context Protocol, the open spec that lets an agent talk to external tools through a standard interface. If you’ve heard “MCP” and wanted the engineering view rather than the product-marketing view, this is it.
For more depth on MCP, pair it with the DeepLearning.AI MCP course and the existing MCP Course spotlight on this site.
Why it matters: MCP is becoming the standard agent-tool interface; this talk is the orientation.
↔ Watch Building Agents with MCP on YouTube
7. Building an Agent from Scratch
A live-coded build of a working agent, end-to-end, in one video. No slides for the first 45 minutes, just a code editor, a console, and the loop coming together. The kind of tutorial that’s harder to produce than a lecture and more useful once it exists.
Pick this over “LLMs from Scratch” if your goal is specifically an agent, not a model. The two videos are complementary, not redundant.
Why it matters: the lowest-friction way to ship a working agent and read its code the same afternoon.
↔ Watch Building an Agent from Scratch on YouTube
8. Philo Agents (playlist)
A curated YouTube playlist of agent-focused talks and demos. Playlists are usually where quality goes to die; this one is the exception. Each video adds something the previous one didn’t, instead of restating the same intro.
Subscribe to the playlist and use it as a drip-feed, one video a week, alongside whatever you’re building. Treating it as a backlog to marathon is a fast way to retain nothing.
Why it matters: a maintained playlist is rarer than a viral talk, and more useful over a quarter.
↔ Open the Philo Agents playlist on YouTube
How to use this list
Don’t watch all eight in a weekend. Pick the one that matches your gap. If you’re new, watch the LLM Introduction first, then the Stanford Agentic AI Overview. If you’re shipping, watch Building and Evaluating Agents, then Building Effective Agents. If you’re coding along, pick LLMs from Scratch or Building an Agent from Scratch and follow the video with an editor open. The playlist is for ongoing weeks, not for the first session.
Every outbound link opens in a new tab with rel=noopener. No redirects, no affiliate tags. If you want the applied counterparts, the Courses roundup is the structured-graded version and the Repos roundup is the runnable-code version of what these talks cover.
Eight videos, three archetypes. Introductory talks, hands-on builds, and a serious playlist: the YouTube surface of LLM and agent content, filtered to the videos you’d actually rewatch. Bookmark the playlist; treat the standalone talks as depth on demand.