Hello hello hello,
As you've likely realized, I'm writing to you a day late. I landed in Denver late Thursday night (I'm spending the next couple of days here), and it's been a little hectic finding the time to sit down and write out my thoughts on everything that happened this week.
However, I've got a hot cup of coffee in my hand and my laptop is open, so let's not waste any time...
3, 2, 1...
3 things worth your attention
OpenAI's GPT-5.6 is here, and DAMN!
OpenAI officially released GPT-5.6 two days ago, and the AI world is going crazy.
To preface, I talked a few weeks back about how new AI models are becoming less and less important because they are all so freaking good now, and while the three models released with GPT-5.6 are no exception, they are changing the way AI experts are using AI (and you should start to think about this too).
There are actually three new models inside GPT-5.6, each built for a different kind of job:
Sol is the heavy hitter. The flagship you pull out for your hardest, most complex problems.
Terra is the daily driver. Balanced, fast, and great at just about everything. This is the one most people will reach for most of the time.
Luna is the budget speed demon. Crazy cheap and crazy fast, but still really high quality. Built for the high-volume stuff you run over and over.
Different tools for different jobs, all under one roof.
The team over at Every has been running it internally for about 30 days across coding, writing, design, and knowledge work. Here's how it stacks up:
Coding: An A-tier coder (Dan Shipper, their CEO, scored it an 8 out of 10). It's still not Fable, but for a fast, effective solution, it's really, really good
Writing: They're calling it the best writer of any model out right now, noticing that it's clearer and more concise than Fable or Opus 4.8
Design: Fable and Opus are arguably still better here, but this is the best design OpenAI has ever shipped
Knowledge work: This is where there's a massive leap (more on that below)...
For the past few years, AI has helped you with individual tasks. Write this email. Summarize this doc. You get the point.
5.6 is the first model people are trusting to run entire loops of knowledge work on its own. Processing your whole inbox. Pulling decisions out of your meetings and Slack. Even scanning Facebook Marketplace for furniture. It's less "AI helps me do my work" and more "AI does the work while I manage the system that does it."
That shift is the thing to pay attention to, and no one explained it better than Dan Shipper himself (CEO of Every) in a 50-minute video with Greg Isenberg, walking through how he runs his entire work and personal life on OpenAI's Codex Desktop + 5.6 (find that video later in the newsletter).
Here's what stuck with me:
It's an operating system, not a chatbot. Email, research, planning, building software, all from one surface. Dan basically lives in it now
Everything becomes a card. Each morning it sweeps his inbox and turns every email into a card with a summary + a drafted reply + a next action. Same for his Slack and meeting notes
Context is the multiplier. It's wired into his whole computer and the web, so it actually knows enough to be useful
The mindset shift is the whole point. In Dan's words, "It's not about any individual email. It's about the system that does my email."
Quick TLDR: GPT-5.6 is fast, cheap, and incredibly good at everyday work. It should become your new daily driver for pretty much everything. If you use multiple models, keep a heavy hitter like Fable reserved for the really big coding and workflow-automation jobs, and let 5.6 handle the other 95% of your day. That happy medium (the right model for the right job) is where all of this is going, and it's a really good place to be.
Claude Cowork can work 24/7? (and it ate Claude Chat?!)
It wouldn't be a weekly edition of The Filter if Anthropic didn't have some new, insane update to talk about.
On the menu this week: Claude Cowork now works on the web, and iOS + Android devices.
There are a couple important layers to this update that you should be aware of.
First, and most importantly, this update allows Cowork to work for you after you close your laptop. The session keeps running in the cloud, making Cowork closer to a 24/7 AI employee than ever before.
Scheduled tasks can now run completely on their own too, even when none of your devices are online. Set one up in the morning, close your laptop, and come back to finished work (it'll even ping your phone when it needs you to approve something).
Second, and this one caught me off guard: Chat and Cowork just became one thing. The separate Cowork tab is gone. Now you kick off both from the same message box you'd use for a normal conversation, and just toggle between "Chat" and "Cowork" depending on whether you want a quick answer or a full task done for you. That same session then follows you across your laptop, phone, and the web.
One quick catch: if your task needs your local files or your browser, you'll still want the desktop app open on that computer (the cloud can't reach your hard drive on its own). But for research, writing, and anything web-based? Fully hands-off.
I've already used this for a few tasks, and it's worked great (not to mention, my computer battery is loving it).
(Here's a quick video Anthropic made explaining the update, if you care to watch that too: https://x.com/claudeai/status/2074525815820169320?s=20)
One more day with Fable 5...
Anthropic lied to us.
Last week they said that Fable 5 was going away after July 7th, and guess what? It's July 11th and Fable 5 is still available in normal Claude plans lol.
They announced the extension earlier in the week, and are claiming it will be taken down after July 12th now (tomorrow night).
I'm choosing to believe Anthropic this time, which means you have a little over 24 hours left with Fable 5 (for the time being).
In case you've been living under a rock (don't sweat it): Fable 5 is the most capable AI model ever released, built to take on long, complicated jobs and run them almost entirely on its own. I went deep on it in last week's edition, so I won't rehash the whole thing here.
But if you want to actually put it to work before it (maybe) disappears tomorrow, I built out the 15 best prompts to run with Fable 5 for when you're not sure where to start.
2 videos
ChatGPT Work
GPT-5.6 is grabbing everyone's attention, but don't let OpenAI's other release slip past you - ChatGPT Work.
From what I've seen and used it with, ChatGPT Work appears to be OpenAI's version of Claude Cowork (only they're about 6 months late lol), giving ChatGPT the ability to finally connect to all of your external tools & use them on your behalf.
If you're an OpenAI power user, this 5-minute video will be worth the watch, but if you've been using Claude Cowork for a while, this is nothing groundbreaking or new for the AI space.
GPT 5.6 SOL IS HERE! How to use it.
I referenced this video already in the newsletter, so I won't talk much more about it here.
Dan Shipper (the Every CEO that has been testing GPT-5.6 for a month) and Greg Isenberg go over the best use cases for GPT-5.6, how to use Codex with it, and more...
This will be the most valuable 50 minutes of your week (besides this newsletter, of course) ;)
1 workflow worth your time
As much as I love making content, posting my content across all my different platforms is the annoying part (not to mention it's incredibly time-consuming).
I make one video, and then I have to go post it everywhere... Instagram, TikTok, then finally YouTube Shorts. Upload, write the caption, switch apps, do it all again.
It's not hard, it's just tedious. And when you're cross-posting the same video to a bunch of platforms like I am, it eats up WAY more time than it should.
So I found a way to hand the whole job to Claude.
Now, the second a video is done, I drag it into my Claude Chat, tell Claude to write the video caption in my voice, and then tell Claude which platforms it needs to post it on, and it posts everywhere for me. That's it. The tedious part just... disappeared.
It takes about 5 minutes to set up, one time. You need 3 things:
An OmniSocials account. This is the piece that actually lets Claude connect to and publish to your accounts. It's free to try for 14 days (after that, it's $10/month), and if you have any questions, the founder answers them himself
A prompt that teaches Claude your voice. OmniSocials reads your old posts and writes it for you, then you paste it into Claude's settings. Don't skip this one. It's the difference between sounding like you and sounding like every other AI post in the feed.
Connect the two. In Claude: Settings → Connectors → Add custom connector → paste
https://mcp.omnisocials.com→ Connect. Two minutes and you're done.
To make sure it worked, open a fresh chat and ask "what social accounts do I have connected?" If Claude lists them back to you, you're good to go.
After that, you just talk to it like an assistant who already knows your voice. Drop in the video/image, say the word, and it's live everywhere.
If you're posting to more than one platform, I cannot recommend this enough.
That does it for this week's edition of The Filter.
The AI world is moving fast, and the best way to learn is by doing.
Get your hands dirty with these AI tools, fail with them, learn through the process, and don't forget to touch grass every once in a while ;)
Talk next Friday (for real this time)
P.S. I haven't talked about this anywhere else, but I have a really cool video being released within the next week. It's a behind the scenes look at a client AI automation project I've been working on. I think you're gonna like it, so be on the lookout on Instagram :)