Claude is Taking Over: Every New Feature Explained (Full Guide)
TL;DR
Anthropic’s big bet is one “super app,” not a product bundle — Riley argues Claude is winning because Anthropic merged chat, coding, agents, connectors, and desktop workflows into Claude Desktop/Co-work, while OpenAI is only now trying to unify ChatGPT, Codex, Atlas, and Sora-like efforts.
Claude Co-work is basically Claude Code with guardrails — instead of giving the model dangerous full-computer access, Co-work confines it to a project folder sandbox, then layers on connectors like Gmail, Slack, Figma, Canva, Excel, and PowerPoint so non-developers can automate real work safely.
The center of gravity is shifting from chatbots to general agents — Riley frames the last three years as text chat (ChatGPT), then vibe coding (Claude Artifacts, Cursor, Lovable, Replit, Bolt), and now general agents like Claude, OpenClaw, Manus, and Perplexity Computer that can actually operate software.
Claude’s feature velocity has been ridiculous — in just a couple months Anthropic shipped Co-work, scheduled tasks, project memory, plugins, Slack/Gmail/Figma integrations, 1M-token context windows, voice in terminal, session sharing, agent teams, remote control, and direct mouse/keyboard computer use.
Some of the most important updates are about persistence and delegation — features like Dispatch, Telegram/Discord channels, scheduled tasks, auto-memory, remote control, and multi-agent “agent teams” push Claude from prompt-response tool into something more like an always-on coworker you can message from your phone.
Claude Code is getting expensive because it’s genuinely replacing hours of labor — Riley says he knows developers spending over $1,000/day on tokens, and companies are happy to pay because faster models like Opus 4.6 can sustain longer agentic work and produce dramatically more output.
The Breakdown
From ChatGPT magic to the age of agents
Riley opens with a bold claim: for three years it felt like OpenAI would own everything, but now “everyone’s moving to Claude.” He maps the whole AI arc in three phases — the text-chat era starting with ChatGPT in late 2022, the vibe-coding era unlocked by tools like Claude Artifacts and Cursor, and today’s “general agent” era led by Claude, OpenClaw, Manus, and Perplexity Computer.
Why Claude Artifacts mattered more than people realized
He calls Claude Artifacts the real turning point because Claude didn’t just generate code — it rendered it. That changed AI from “here’s a code file” into “here’s a thing you can actually click and use,” which pushed Riley headfirst into building apps and eventually into Cursor, where AI could spin up more complete front-end and back-end projects.
Claude Code: the tool that made people obsessed
From there he moves to Claude Code, which he describes as the terminal agent that “took over the world by storm.” His own use case is practical and very non-hypey: internal tools, landing pages, marketing assets, even slide decks in a few prompts, all saved directly to his computer — but with enough freedom that “dangerously skip permissions” can let it install apps, reorganize files, and generally do whatever you ask.
Claude Co-work is the safer bridge for normal users
Co-work is his key framing device: halfway between chat and Claude Code. Instead of touching your whole machine, it works inside a sandboxed project folder, but still plugs into tools like Figma and Gmail; he demos connecting both in minutes, then asks it to study his script folder and write an intro in his voice, which comes back sounding uncannily like him.
Skills, plugins, and the quiet replacement of software
One of the stickier ideas in the video is his claim that “most software is going to be replaced by skills.” In his example, Claude turns a script into a PDF without needing a separate PDF app, and he extends that logic to plugins and shared workflows: admins can package connectors, commands, and instruction sets so teams can reuse the same effective agents.
The feature sprint: connectors, scheduling, diagrams, and huge context
The middle of the video is a rapid-fire tour of releases: Slack MCP on all plans, easier Gmail setup, Claude in PowerPoint and Excel, Windows support, folder-specific instructions, better Figma workflows, scheduled tasks, clickable built-in diagrams, and default 1 million token context windows. Riley’s bigger point isn’t just that these exist — it’s that Claude keeps becoming more interconnected, more persistent, and more useful as an actual coworker.
Dispatch, channels, and the weirdly powerful desire to text your computer
Then the story shifts into what he sees as the hottest new pattern in AI: persistent agents you can message from your phone. He ties Claude’s Dispatch and Telegram/Discord channels directly to the rise of OpenClaw-on-a-Mac-Mini setups, saying people clearly want “the power of a computer while being on their phone,” even if the category still feels early and slightly strange.
Claude Code’s deeper upgrades: memory, teams, security, and computer control
The final stretch gets more technical but also more consequential: session sharing, /insights summaries, rewind, built-in security scanning, remote control, auto-memory across sessions, voice mode in terminal, scheduled code reviews, and the /btw side-thread while long jobs run. He ends with the biggest update of all — Claude controlling mouse, keyboard, and screen — calling it “endgame for AI agents,” and closes on the idea that Anthropic stayed focused on one giant tool while OpenAI spread across multiple apps.