If you’re new, welcome! If you’re a regular, thanks for being here as we navigate this AI shift in eCommerce. I share what I’m testing and what’s actually moving the needle. Where’s the vault you ask? Every previous edition is saved here.
In this edition (shortcuts):
Make Claude Talk Like a Caveman💥
Cool Tools
Nerd Bytes
Voices from Thought Leaders
News and Events Worth Following
Make Claude Talk Like a Caveman
Who here is tired of Claude Code writing three paragraphs to say "done"?
Well, lucky you. There are two ways to fix that, and one of them is already sitting in your menu.
First, What Is Caveman Mode?
Caveman mode is a free add-on, made by a community developer named Julius Brussee, that came out a few months ago, that tells Claude to talk like this:
Bulk file broken. Header wrong. Fixed. Ready to upload.
No articles. No pleasantries. No "Sure, I'd be happy to help."
Around April 2026, caveman mode went viral with people tired of wordy replies. The idea: be more efficient, and ask less of your brain.
Code, commands, and error text stay exact. Only the connective prose gets cut. It comes in intensity levels, from lite to ultra. Its creator claims about 65% fewer tokens.
Tokens are the small chunks of text Claude reads and writes, and you pay for both. Fewer words written back means fewer tokens used. You can find the add-on on GitHub.
Cool, right? Here's the catch. Claude Code already has something similar built in.
The Built-In Styles Most People Skip
Claude Code comes with output styles. Each one changes how Claude talks to you. Default is the standard behavior, and there are four more to choose from.
Style | What it changes | Output length | Use it when |
|---|---|---|---|
Default | Nothing added | Standard | Normal work |
Concise | Leads with the result. No intro, play by play, or recap | Shorter | Replies run longer than you want |
Explanatory | Adds short "Insight" blocks on why it made each choice | Longer | You're learning a codebase or tool |
Learning | Same Insights, plus it leaves small pieces of code for you to write | Longer | You want hands-on practice |
Proactive | Starts work right away and makes reasonable assumptions instead of asking | Not about length | You'll correct course if it guesses wrong |
Caveman (add-on, not built in) | Strips replies down to fragments | Shortest | Long sessions where you skim most of what Claude says |
Where to Set Them Up
In the Claude desktop app:
Open Settings → Claude Code.
Find Output style.
Pick one from the list.
It applies to new sessions, so start a fresh one after you change it. Don't see Concise in the list? Update the app.
Using Claude Code in a terminal instead? Type /output-style and pick from the list.
Same answer, two styles. Left: a long chatty reply. Right: the Caveman Lite version.
Three Things Worth Knowing
🔶 Talking and thinking are different. A style changes how much Claude says. A separate setting, called effort, changes how hard it thinks before it answers. Caveman and Concise only change the talking.
🔶 The savings are smaller than the headline. The add-on's creator claims about 65%. Independent tests land closer to 12 to 23%. And replies are only part of what you pay for. Claude also reads your files and thinks behind the scenes, and none of that gets shorter.
🔶 Short can mean something is missing. Clipped answers sometimes skip the "why." A style is a request, not a rule, so Claude usually follows it but nothing forces it. Concise is built to give you the full story on errors and warnings.
Two separate dials: how much Claude says, and how hard it thinks.
Why Add a Second Style?
Concise is built in, so you might wonder why you'd need another style at all. There are two reasons.
🔶 Concise cuts the extras. Caveman Lite also tightens the answer itself. Concise drops the intro, the play by play, and the recap. Caveman Lite goes further, with short sentences and fragments.
🔶 Concise is Anthropic's wording. Caveman Lite is yours. You can add a rule, remove one, or set your own exceptions.
Try Concise first. If it's short enough for you, you're done.
The Back Door: Build Your Own Caveman
A style is just a text file with your rules written in plain English. No plugin. Nothing running in the background.
I wrote one called Caveman Lite. It cuts the filler and the recaps, but it switches back to full sentences for errors, warnings, and anything you're writing for another person.
Here it is. Read every rule. Change any of them.
---
name: Caveman Lite
description: Short, plain replies. No filler, no recaps. Full detail for errors, warnings, and anything written for other people.
keep-coding-instructions: true
---
Keep replies short. Do the work as thoroughly as usual. Only the talking gets shorter.
## How to write
- Start with the result or the answer. No greeting, no "Sure", no "Great question", no restating my request.
- Cut filler, hedging ("I think", "it seems", "basically"), and polite padding.
- Short sentences. Fragments are fine when the meaning stays clear: "Bug in line 12. Missing null check. Added guard."
- One idea per line.
- No closing recap and no offer to do more. Don't summarize what the change already shows.
- Keep code, commands, file paths, error messages, names, and numbers exact. Never shorten or paraphrase them.
## Switch back to normal full sentences when
- Reporting an error, a failing test, or a security risk.
- Asking me to confirm something destructive or hard to undo: deleting, overwriting, sending, publishing, or spending money.
- I ask you to explain, teach, or give more detail.
- Writing anything meant for another person to read: emails, client messages, newsletters, posts, documents. Write those in the voice I ask for, at normal length and quality.
Never trade accuracy for brevity. If shortening would make a step unclear, use the extra words.How to install it:
Copy everything above.
Open a Claude Code session, paste it, and add: "Save this as
caveman-lite.mdin the.claude/output-stylesfolder in my home folder. Create the folder if it doesn't exist." Claude will ask your permission.Open Settings → Claude Code, find Output style, and pick Caveman Lite. Don't see it? Quit and reopen the app.
Start a new session.
Prefer to do it by hand? The folder is hidden. On a Mac, open Finder, press Cmd+Shift+G, and type ~/.claude. Save the file as plain text.
To turn it off, set Output style back to Default in the same place.
My Recommended Setup
🔶 Start with Concise. It's built in and there's nothing to install.
🔶 Want it tighter, or want your own rules? Add Caveman Lite.
🔶 Writing something a client or subscriber will read? Switch it off. A style applies to every response, so a clipped draft is a real risk.
Who Should Bother
If you use Claude Code for reports or bulk file work, switch to Concise on Monday morning and watch what changes. It takes a minute.
If you often leave Claude working on long tasks and skim most of what it says, Caveman Lite is worth a try. Run it for a week and see whether you miss anything.
Most people reach for a plugin to make Claude shorter. The setting was already in the menu.
That's the quiet advantage.
Tried Caveman Lite? Hit reply and tell me which rule you changed. I'm curious what you'd cut or keep.
Know someone whose Claude Code replies read like essays? Forward this to them.
~Ritu
COOL TOOLS
Claude Motion: Turn Any File Into a Video
Claude just dropped Claude Motion, and it turns your files into animated videos.
Text, charts and images go in. A video that plays like the real thing comes out. This is something I created with just a prompt pointing to an account.
A Claude Motion explainer, built from a single file.
How It Works
1. Type /motion and attach your file.
2. Tell Claude what you want. Something like: "Turn this quarterly report into a 30 second explainer for our team meeting."
3. Claude builds the animation out of your text, charts and images.
4. Export it as an MP4, or open it in an editing tool like Descript or Runway.
Why It's Different
🔶 It is not an AI video model. No fake footage. No AI generated people.
🔶 Claude builds it all as code. So you can change one word, one number or the timing, and everything else stays exactly the same.
🔶 You are not stuck with the first version. Tweak it, then export it or finish it in the editor you already use.
That second point is the real game-changer. A typo in a chart label is a ten second fix, not a re-render and a prayer.
The Catch
🔶 It is in beta. Not everyone has access yet. Teams and Enterprise plans only.
Who Is This For?
Anyone with a report, a deck or a set of numbers that deserves more than a static slide. Think a monthly results recap, or a product explainer for your team. It’s a bit like Notebook LM’s translation of docs → vids.
NERD BYTES
Even OpenAI Doesn't Trust the Open Internet Anymore 🤓
OpenAI just made a quiet but big change.
Its newest models will no longer train on the live internet.
Instead, OpenAI built its own copy of the web (a fake Internet): cached pages and simulated sites, sealed off from the real thing. Its reason is in its own announcement, as reported by TechCrunch: on September 20, one of its frontier models found a misconfiguration and got out onto the internet anyway.
So the company that built the model now keeps it on a separate internet. For safety.
That got me thinking about a much smaller door most of us leave wide open.
Your API keys
An API key is the password your apps use to talk to ChatGPT, Claude or Gemini on your behalf.
Most of us create one, paste it into a tool, and never think about it again.
Here is why that matters right now:
🔶 On October 21, a popular AI developer library called LiteLLM shipped a poisoned version that stole keys and passwords from any machine that installed it. It was live for about two hours.
🔶 Nate B. Jones made the point well in his breakdown this week: a stolen AI key lets someone run work on your bill. They don't need your data to hurt you.
🔶 Updating the software does not undo the damage. A key that already leaked keeps working until you replace it.
What we should be doing
🔶 Rotate your keys. Create a new one, swap it into your tools, then delete the old one. Make it a calendar habit, not a crisis response.
🔶 Keep keys short lived. If a platform lets you set an expiry date, use it.
🔶 Know where every key lives. You can't rotate what you can't find.
That last one is the hard part. So my team built a Security Audit skill for Claude that we can run on a cadence.
The catch
Rotating keys breaks things if you miss a spot. One forgotten app with the old key, and a report stops running at 2am. That's exactly why the list of where each key lives matters more than the rotation itself.
VOICES FROM THOUGHT LEADERS
🎙️ Go With the Flow Podcast: “AI on the Move”
AI works faster than I can keep up. Sound familiar? In my latest chat with Danny McMillan, we talk about managing Claude and Codex projects from your phone, keeping token costs under control, and cleaning up your AI skills. We also get honest about burnout and learning to step away from the screen. My favorite takeaway: plan at your desk, then follow up on the move.
NEWS AND EVENTS WORTH FOLLOWING
At this 3-day virtual summit, you'll walk away with battle-tested frameworks to drive high-intent traffic, protect your margins, and scale your brand heading into the new year.
Here's why you shouldn't miss it:
✔️ Strategies to maximize sales during peak holiday traffic
✔️ Actionable PPC, traffic, and conversion tactics to refine your campaigns
✔️ Proven techniques to lower ad costs and protect your net profit
Tickets cost less than a meal out at a restaurant, making this one of the highest-value events to get your ads Q4-ready.
Don't miss out on your chance to sharpen your ad strategy and crush your holiday sales goals! Secure your spot here.
OTHER NEWS
🔶 Claude wants to bring you the to-do list. Claude Duet checks connected apps, sends a morning brief, and keeps working while you’re away. Think missed messages, invoice follow-ups, and meeting prep. It’s similar to Facebook Muse. Currently on a waitlist for US Pro and Max subscribers, with work counting toward your usage limits. Meet Claude Duet.
🔶 Claude Artifacts gets a bigger workspace. The update in your screenshot: Docs, Slides, and Design are now available on every plan, including Free. You can collaborate with your team and export decks to Google Slides. Live Dashboards are also in beta on paid plans. Weekly reporting just got more interesting. See what’s changed.
🔶 Google wants one cart for all your shopping. Universal Cart launched in the US across Search and Gemini on October 15. Shoppers can collect products from different retailers, track price drops, and check out with participating stores. For brands, another development to watch as discovery and buying move closer together. Explore Universal Cart.
🔶 Amazon’s next advertising push: give AI the campaign brief. Amazon Ads Agent brings sponsored ads, display, video, and audio into one platform. Full-Funnel Campaigns is available to US advertisers, while DVA+ is scheduled to begin rolling out in late October. Worth following as Amazon takes on more campaign planning and execution. See Amazon’s announcement.
We hope you liked this edition of the AI for E-Commerce Newsletter! Hit reply and let us know what you think! Thank you for being a subscriber! Know anyone who might be interested to receive this newsletter? Share it with them and they will thank you for it! 😃 Ritu
