every company needs an AI OS
Ramp's new Cowork competitor, plus my static ad engine and Claude's new dashboards
Remy Gaskell
October 19, 2026
 |
Hola amigos, grab a seat.
There a 40 THOUSAND of you mf’s now reading my stupid emails each week!!
let’s get into it shall we *hat tip
|
📌 TL;DR
Claude's new tools → Live dashboards, short animated videos, docs, slides and designs, all made right inside Claude.
Gemini 4 Argon → Google's flagship is open to everyone, at roughly half the price of Opus for now.
Ramp Glass → An AI workspace where every employee starts from what the company already knows.
Muse Wonder → Meta's new models look frontier on paper. Early testers found them slow.
Builder's notes → Instinct runs my personal admin, plus the workflow we built to remake competitors' winning static ads in Paper.
Claude can now make dashboards, videos, docs and designs
Claude can now build four new things for you: live dashboards, short animated videos, docs and slide decks, and designs.
1. Dashboards
Claude can now build dashboards that connect to your live data through your Claude connectors.
For example, you could build one for your ads and marketing performance, with data pulled in from the Meta Ads connector and Shopify.
It's live and keeps syncing, so open it again a week later and the numbers are up to date. You can also share your dashboards with your team inside Claude.
2. Motion
this is really cool actually. claude can now generate short animated videos, in a variety of styles
Type /motion and describe what you want.
Claude builds the animation as code, not AI footage, so you can change any word, number or timing afterwards.
 |
3. Docs and Slides
Claude now has its own version of Google Docs and Google Slides, built right into the app, so you can write and edit docs and decks together with Claude.
I'm actually drafting this newsletter you're reading right now in a Claude doc, and I'm really enjoying it so far.
4. Design
I believe they are merging the old seperate ‘claude design’ product inside claude.
so landing pages, social graphics and mockups, are now made right inside Claude instead of on a separate site.
If you used the old Claude Design site, it closes on 14 December, so move your design systems over before then.
FYI: the 4 new output types all show up in claude as artifacts...
 |
Gemini 4 Argon: new model alert
 |
Google announced its new flagship model; Argon 4
It's in the Gemini app + in Antigravity, Google's coding tool.
Right now it's roughly half the price of Claude Opus 5.5, but that's just an introductory price though. On 30 November it doubles, to the same price as Opus.
On Artificial Analysis, the independent benchmark site, it scores 53, against 58 for Opus 5.5. So it's close to the top, not the top.
Where it shines
Structuring writing. Give it a messy long draft and it turns it into a clear story.
Ideas. It comes up with the most creative details.
Following your lead. It sticks closely to a design you give it, and to your voice.
Where it falls short
It doesn't check its own work. It finishes and moves on.
It crams too much in. Slides and pages come out dense.
Google's apps hold it back. Antigravity felt about a year behind other coding tools.
Ramp is launching a Claude Cowork competitor
Ramp (the corporate card and finance software company) launched Glass, its answer to Claude Cowork and ChatGPT for teams.
It's an agentic OS for companies, with a company brain, a skill library and your tools built in, so every employee starts from everything the company has already figured out, instead of an empty chat.
 |
Ramp's line for it: "We learn things once."
kinda cold 🥶
What's inside
A company brain. It connects to Linear, Notion, Google, Slack, Gong and dozens more, and remembers what your team teaches it.
A skill library. Saved, reusable ways of doing a job. Ramp calls them "one person's best work, now everyone's starting point."
Multiplayer sessions. A support agent can pull an engineer into her session to finish a ticket together.
Any model, open or closed. Ramp Router picks the model for each request, and Ramp says its own AI bill fell 30% overnight.
A cost meter on every project, with spending limits.
Why Ramp built it
Giving every employee AI ... is not the same as becoming an AI-native company.
One week in May, Ramp burned $1.5M in AI tokens.
The problem is that someone works out the perfect way to do a task like prepping a sales deck, but it lives in one person's chat history.
It disappears when they close the tab, go on holiday or leave.
So a small team packaged up the setup Ramp's most AI-forward employees had built for themselves, and gave it to everyone.
What it changed inside Ramp
The CEO asked how many UK customers they'd added….
Glass knew to leave out US companies with UK offices, because Finance had taught it that months earlier.
Feature requests from customer calls used to be copied into Salesforce by hand, about 300 a month.
An account manager saved a prompt that pulls them straight from the calls, and now it's 20,000 a month.
A second prompt prioritises those requests, and a third emails customers when theirs ships (5,000+ emails so far).
In total, 323 Ramp employees have made 6,500+ edits to what Glass knows.
It's in early access at ramp.com/glass, and Ramp is onboarding the first businesses personally. There's no public pricing yet.
Ramp also just raised $1.85B at a $60B valuation.
Meta's Muse Wonder: frontier on paper, slow in practice
Meta released Muse Wonder and Muse Wonder Full, which it calls its most capable models.
Meta's big pitch is "Active Memory".
Meta says it trained Wonder to get better at using memory: knowing what's worth keeping, and pulling up the right detail when it becomes relevant.
The two models
Muse Wonder is available through Meta's API and its coding app, Muse Code, at $4 per million input tokens and $20 per million output. It will "soon" power Meta's everyday Muse assistant.
Muse Wonder Full is the bigger model, and it's only inside Muse Code (plans start at $50 a month). There's no API.
Neither can be downloaded and run yourself. That's a change for Meta, whose older Llama models anyone could download.
Ben Tossell from Ben's Bites tested both. He found them slow, and "churning out awful UI".
His verdict: he "probably won't reach for these models again" unless he's just testing.
Also this week...
xAI launched The X Subscription → one plan that bundles Grok, Grok Bot, Cursor and X Premium, with a shared pool of Grok usage across all of them.
Builder's notes
1. Instinct is my personal assistant now
I've been really, really loving Instinct to fill the personal assistant side of my life as an AI agent.
I message it like I would a real assistant, and mostly I just voice-note it.
I had to fly to Miami this weekend for a meeting, so of course I had Instinct book my flights and hotel.
 |
Then it booked the Delta flight on my business card. All I had to do was approve the payment in my Wise app.
 |
And the hotel.
 |
The night before, any detail I needed, I just voice-noted it:
"Hey, when's my flight tomorrow?" Bang, tells me the details.
"Hey, what time should I book an Uber for tomorrow from the hotel to make my flight in time?" Bang, tells me.
"Hey, can you schedule the Uber for me?" Bang, done.
Even super small, trivial things.
I had some Buck Mason store credit sitting somewhere in my emails from a few months back. At the checkout, about to buy a jumper, I voice-noted Instinct: "Hey, can you find my store credit code for Buck Mason?"
Literally within a few seconds I had the code to tell the gentleman at the checkout. Rather than me sitting there scrolling through my emails for 5 minutes.
I find that your emails are a really adequate second brain.
The other day I pulled up to my Airbnb and sent Instinct a voice note: "Hey, how do I get in my Airbnb?"
It told me straight away the door code, the key safe code and where to look for each one, pulled from the check-in email. "Hey, what's the Wi-Fi password for the Airbnb?" Same thing.
Agents talking to agents
With all my close friends, we've made our Instincts and set them up on the trusted people network, so our Instincts can talk to each other to make plans and organise schedules.
Here's Mav's Instinct asking for my Australian address, and mine checking with me before it hands it over:
 |
2. Static ads with Paper
This week I ran a live workshop for The AI Course community, the AI Ad Creative Masterclass, with our coach Zach.
We built a workflow that finds static ads already working for your competitors, then remakes them for your brand in Paper.
Why remake what's already working
Your competitors have already paid to find their winners, and the Meta Ad Library shows you every ad they're running.
AI isn't great at coming up with net-new creative concepts. But it's really good at taking what's working for other brands and remixing it for yours, with your own spin on it.
And you only need one to rip. When I ran my agency, I had a "breaking news" style ad for a mortgage broker client: a picture of a nurse, with a headline about a scheme that let nurses wipe thousands off their home loan.
That one ad took all the spend in the account for over 12 months and brought in thousands of leads.
Why Paper, not an image model
Image models like GPT Image 2 make great statics. But they're expensive, and they can't make small, exact changes.
When an ad wins, you want to spin up variations. Same text, different product shot. Same image, ten different testimonials.
An image model generates every version from scratch, so it slightly changes the bits you wanted to keep.
Paper is basically a Figma built for AI agents. Claude builds the ad as an editable design, so every word, image and layer can be changed on its own. Swapping a headline is exact and cheap.
 |
(Paper's $20/month plan also includes image generation for the photos.)
Why it's built as three skills
paper-ads-prep builds your brand brain once: what you sell and why, how you sound, your customers, testimonials, offer, product facts and sizes, and your branding pulled from your site. Garbage in, garbage out.
competitor-ads-intel finds your competitors and pulls the static ads they're running from the Meta Ad Library into Paper as references.
paper-ads remakes the ones you pick for your brand, in feed and story sizes. Tell it you hate an angle and it adds that to a learnings file, so that angle never comes back and it gets better every run.
Why a human stays in the loop
We originally wanted it fully automated. Just tap the skills, bang, bang, bang.
But Claude doesn't have the taste. It would find some really shitty ads amongst the good ones, remake them, and you'd get terrible ads out the other end.
So you pick the references (we just talked through them with Wispr Flow) and give feedback. Your taste decides how good the ads are.
Then rerun it every month with fresh competitor ads.
We ran it live for a sleep mask brand, a hair transplant clinic and my own newsletter.
 |
 |
|
Tools to try
Anatomy → a free skill for Claude Code (also works in Codex and a few other agents). Describe a real-world machine or complex system and it builds it as a detailed, interactive isometric illustration. Great for explainer visuals. (free and open source, install it with one command from the GitHub page)
404s.design → a curated gallery of creative 404 pages. Filter by style, interaction or industry, or hit "Surprise me" to shuffle. Grab a few you like and hand them to Claude as references next time you build a site. (free to browse)
bot pfp → build a cute bot and download it as your profile picture. Pick its parts and looks (or hit Randomize), then download it as a PNG or SVG.
Brain food
Give your Grok Bot every model → Corey Ganim sent his bot one tweet, it installed Claude Code and Codex itself, and now it writes with Opus 5.5. Sorts out my model-choice gripe from last week.
Send this URL to your agent → Notion's official setup guide. Your agent copies your memories and skills into Notion (asking first), so every agent you use shares one brain.
How Wispr Flow's paid ads were built → a 15-minute breakdown from the guy who built Wispr Flow's paid ads program, and how he'd rebuild it in 30 days.
morning walks observing the world has lead me to the conclusion that; there is so much dignity in every profession….
it’s thanks to the cab drivers that hundreds of people get to their important meetings each day
it’s thanks to the garbage man that we can enjoy clean streets,
the smiling barista at the coffee shop brightens up hundreds of peoples mornings
