13. Agent Platform Landscape
Note: The video covers material not in the guide below — please watch in full.
Training Guide
You've experienced an agent platform deeply. You've configured it, created documents, managed tokens, saved your work, spun up sub-agents. You know what it feels like to work this way.
Now let's zoom out. Claude Code isn't the only agent platform. Understanding the landscape — what else is out there and how it's evolving — gives you the bigger picture.
(Let's start with the major players)
The Agent Platform Landscape
Every major AI company is building their own agent platform. Here are the ones worth knowing:
Claude Code (Anthropic) — what you're using. It runs inside VS Code, directly on your machine, in your file system. It's built for knowledge work — writing, research, project management, document creation. It's the one we chose because the underlying model is the strongest for conversational, professional work.
Co-Work (Anthropic) — think of it as Claude Code that runs in the cloud instead of on your machine. You give it a task, it works in the background, and you come back to check the results. Good for long-running tasks where you don't need to watch it work in real time.
Codex (OpenAI) — OpenAI's agent platform. Similar concept to Claude Code — an AI that can read files, write code, and execute tasks. Built on GPT models. More engineering-focused than knowledge-work-focused, but the gap narrows with every release.
Gemini / Anti-Gravity (Google) — Google's approach to agent platforms. Different paradigm, strong on reasoning benchmarks. The ecosystem is newer but Google has the resources to move fast.
The landscape changes fast. New platforms launch, existing ones add features, capabilities leapfrog each other. But here's the thing that matters: the skills you're learning transfer. The concepts — file system access, token management, context windows, sub-agents, chain-linking — those aren't Claude Code features. They're agent platform features. If you switched to Codex tomorrow, the mental models are the same.
We chose Claude Code because the underlying model is the strongest for knowledge work — writing, research, strategy, professional communication — and the agent features are the most mature for what we do at LeanScale. That could change as platforms evolve, but right now it's the best fit. You're not learning a tool — you're learning a way of working.
(Now here's something that gives you context on how fast this is all moving)
How Claude Code Got Here
Claude Code started as a developer experiment. It was a tool that lived inside the terminal — the raw command line that developers use. No visual interface. No sidebar. Just a blinking cursor and text. Developers loved it for writing code, but it wasn't built for people like us (knowledge workers).
Also: Claude's underlying model turned out to be exceptionally good at knowledge work — writing, research, project management, document creation. Not just code. People who weren't developers started using it. And Anthropic noticed.
So they built the VS Code extension — a more visual, friendlier way to use Claude Code. Same power as the terminal, but wrapped in an interface you can actually see and interact with. That's what you've been using.
Then came Co-Work — Claude Code that runs in the cloud. You give it a task, walk away, and come back to check the results. Less hands-on, less visibility, but useful for background work.
And of course there's Claude in the browser — Claude.ai. The consumer app we talked about in the earlier trainings. The one where you type a question and get text back.
(Here's the important part — these aren't just different products. They're a spectrum)
The Spectrum of AI
Think of it as a slider. On the left, you have maximum power and visibility. On the right, you have maximum simplicity but less control.
Terminal → VS Code Extension → Co-Work → Browser (Claude.ai)
As you move from left to right, you lose features. The terminal gives you everything — every setting, every flag, every power-user option. The VS Code extension gives you almost everything, wrapped in a visual interface that's easier to navigate. Co-Work strips away the real-time visibility — you can't watch the work happen, you just get the results. The browser is back to the consumer app world — type, get text, copy-paste.
Right now, at the time of recording, the VS Code extension is the sweet spot. It has all the power-user features you need for day-to-day work at LeanScale — file system access, token management, sub-agents, diffs, task tracking. There's more functionality if you drop down to the raw terminal, but for your role, the extension gives you everything.
Co-Work is useful for specific situations — long-running background tasks where you don't need to supervise. But for the hands-on, interactive work you'll be doing most of the time, it's underpowered.
And the browser? That's the consumer app. You already know why that's a different world.
(Here's the thing to remember about all of this)
This Is the Cutting Edge
Everything you've just learned about — the platforms, the spectrum of control, the evolution from terminal to browser — this is all very, very new. A year ago, most of this didn't exist. Six months from now, parts of it will look different.
The landscape will keep shifting. New platforms will launch. Features will move around. The spectrum might compress as tools get better. What's in Co-Work today might end up in the browser tomorrow.
But the fundamentals — agent architecture, token management, context windows, sub-agents, chain-linking — those aren't going anywhere. The concepts stay. The specific tools are just the current best way to access them.
You're on the cutting edge of this. LeanScale is investing in staying ahead of it. And when things change — and they will — you'll already understand the underlying concepts well enough to adapt.
(Next up: how your work gets saved and synced with GitHub)
Comment in Slack
Post your answer in your onboarding channel.
What was your biggest takeaway(s) from this training?