Anthropic, the unstoppable
Some companies launch one product per quarter and spend three months hyping it up beforehand. Anthropic decided to ship seven features in a single week and left the internet trying to process everything at once. Everyone’s mental loading screen is spinning right now.
And the most interesting part isn’t any individual feature. It’s what all of them together mean: Claude has stopped being just a chatbot you’d visit when you needed help. It’s becoming an always-on system, distributed across your devices, your apps, and your daily routine — whether you’re paying attention or not.
Which is to say, Skynet is closer than we thought. Fingers crossed!
(Be nice to your Claude.)
Dispatch: controlling Cowork from your phone
You know that magic trick OpenClaw (Cloudbot, whatever — it’s changed names two hundred times…) used to pull off where you’d send a message from your phone and come back to your computer with the work already done? Well…
Dispatch does exactly that. It creates a persistent conversation between the Claude app on your phone and Claude Desktop on your computer. You send the task from the couch, Claude runs it on the desktop, and you come back to finished work.
Setup takes two minutes: open Cowork on the desktop, click Dispatch in the sidebar, scan a QR code with your phone, done. No API keys. No config files. No unnecessary suffering. No security holes (yet).
Channels: Claude Code on Telegram and Discord
This one is for the devs, vibecoders, and everyone in between.
Channels connects your Claude Code session to Telegram or Discord via an MCP (Model Context Protocol) plugin. You send a message to the bot, Claude Code receives it, runs it, and replies in the same conversation.
VentureBeat called it an “OpenClaw killer” — and the comparison is fair. OpenClaw, that infamous tool that went viral earlier in the year, offered something similar but required a dedicated Mac Mini to avoid FOMO, Node.js 22+, a WebSocket gateway, a pentagram drawn on the floor, and a setup that felt like a summoning ritual. Channels requires installing a plugin and scanning a code. That’s it.
The architecture is clean: when you start Claude Code with the --channels flag, it spins up a polling service that monitors the chosen messaging platform. Message comes in, Claude runs it and responds through the same channel.
One limitation: if Claude Code hits a permission prompt while you’re away, the session pauses until you approve it locally.
For fully autonomous use, there’s the --dangerously-skip-permissions flag — but only use it in environments you trust. The flag’s name is already a warning in itself.
Double usage promotion: 2x off-peak capacity through March 28, 2026
Anthropic doubled Claude’s usage capacity during off-peak hours — defined as any time outside 8 AM to 2 PM Eastern time.
No sign-up. No coupon. Works automatically.
The geographic math here is interesting. If you’re in the US on the West Coast, peak hours shift accordingly. Your late-night coding session (or productive procrastination) is covered with double capacity.
The promotion doesn’t count against your weekly limits. It’s genuinely extra capacity, not a redistribution of what you already had.
This is probably Anthropic’s first experiment with time-based dynamic pricing. The “everything for one flat price” model for AI services always seemed too temporary to last, given how high computing costs are. If this works, expect more dynamic pricing in the future.
(Translation: take advantage while it’s free.)
1 million token context window: now for everyone
Opus 4.6 and Sonnet 4.6 now include the full 1 million token context window at the standard price. No multiplier. No premium tier.
To put that in perspective: 1 million tokens is roughly 750,000 words. That’s about ten full novels you haven’t read, or a complete codebase you also haven’t read, or all the emails you’ve sent and received in the past year (including the ones you pretend you didn’t see).
Before this week, the 1 million window was in beta with limited access. Now it’s standard. And the pricing change matters: there’s no extra cost for using the full window. You pay the same whether you use 10,000 tokens or 1 million.
This means fewer compactions — that process where Claude needs to summarize earlier parts of the conversation to free up space. With 1 million context, entire work sessions fit without compaction. Your instructions from the start of the session are still accessible at the end. Nobody will complain about your verbosity anymore.
For Claude Code users, entire repositories fit in a single session. Debugging across dozens of files becomes a continuous conversation instead of a fragmented series of handoffs or hacky workarounds.
Voice mode: talking to Claude Code
This one is rolling out gradually — currently available to about 5% of Claude Code users.
It’s push-to-talk: hold the spacebar, speak, release. Claude transcribes and processes it like typed text.
It doesn’t listen all the time (which is a relief, because my internal monologues are embarrassing). You activate it with the /voice command, hold the key, speak, and release. The transcription supports 20 languages, including English, and was optimized for technical terms and repository names.
Many devs report they can dictate complex requirements faster than they can type them — especially for explaining multi-step workflows or describing bugs.
Makes sense. Devs don’t know how to write: sometimes it’s easier to explain the problem than to write it out.
If you don’t have access yet, update Claude Code to the latest version and try again in a few days.
I used it and didn’t find it particularly useful, but you might. It could genuinely help people with accessibility needs.
Memory for everyone: Claude now remembers you
Until recently, every conversation with Claude started from scratch. No memory of previous discussions. No retained preferences. No context from past work. You’d re-explain yourself every session like someone telling the same joke to the same person every week.
Now Claude retains context and preferences between conversations. Your name, your writing style, your ongoing projects, your preferences — everything persists across sessions.
For those already using Cowork with context files (about-me.md, brand-voice.md, working-style.md), memory adds an extra layer. Your files handle structured, deep knowledge. Memory handles conversational continuity — the small preferences and threads that would be tedious to encode in files.
A nice detail: you can import your memory settings from ChatGPT directly into Claude with one click. For anyone migrating, this removes one of the biggest friction points.
Just go here: https://claude.com/import-memory
And you can view and edit everything Claude remembers about you in Settings. Nothing is hidden. Control is yours.
(Unlike that ex who remembered everything you said in 2016 and used it against you in 2024.)
/loop: recurring tasks in Claude Code
Set an interval and a prompt, and Claude executes it automatically on that schedule. It’s basically a lightweight cron job inside the session.
The syntax is simple:
/loop 5m check the deploy
That tells Claude to check the deploy status every five minutes. Runs as long as the session is open.
Use cases that already work: CI/CD monitoring during deploys, watching log files for specific errors, checking API endpoints at regular intervals, monitoring build status (I use it with Cloudflare), and periodic code quality checks.
It’s not a complete scheduling system since it runs inside the current session and stops when you close it. For persistent scheduled tasks, Cowork’s scheduled tasks feature is more appropriate. But for temporary monitoring during active work, /loop fills a gap that previously required separate tools like Ralph Loops (https://github.com/snarktank/ralph)
.
The elephant in the room
Seven features. One week. None of them, individually, is the technological apocalypse that X.com loves to proclaim. But together, they outline a pattern worth paying attention to.
Claude is no longer a tool you open when you need it. It’s transforming into a system that operates in the background: on your phone, on the desktop, in the terminal, on Telegram, on Discord — with persistent memory, massive context, and the ability to run recurring tasks unsupervised.
It’s the kind of change that doesn’t generate an instant “wow!” moment, but six months from now makes you look back and think “how did I work before this?”
Or not. Maybe in six months we’ll all be hyping the next big thing and pretending this one never existed. (I’ll probably be right there with everyone else.)
Either way, if you use Claude on any plan, it’s worth checking your memory settings, testing Dispatch if you use Cowork, experimenting with Channels if you live in the terminal, and taking advantage of the double usage promotion before it ends.
And if you’re the type who enjoys watching the AI arms race with popcorn and a healthy dose of skepticism — well, Anthropic’s week was a feast.
Just don’t ask Claude to do all your work while you sleep.
Not yet. Well, maybe by July.