Computer use in Claude Code is a huge unlock.
The biggest bottleneck in AI coding is it can't “see” what it built. Computer use gives Claude Code eyes.
It can now run this closed loop: “write the code, compile it, launch the app, click through it, find the bug, fix it, and verify the fix.”
The fact that coding is inherently closed-loop verification makes it easier and more reliable than open-ended computer use tasks like “do my taxes” or “book me a flight.”
After two months in development, today I’m excited to announce the beta launch of Alfred Desktop Web 🚀
It’s like OpenClaw (long live the 🦞) - but radically simpler
If you’ve wanted to try OpenClaw but have been too intimidated, Alfred is for you
Alfred makes a deliberate tradeoff:
~70% of OpenClaw’s power
at ~1% of the complexity
The guiding principle:
Simple Defaults. Infinite Customization.
What does that actually mean?
🔌 Apps
✅ Simple Default
Out of the box, you can connect Slack, Gmail, Calendar, Notion, Asana — and 800+ more apps. Alfred uses a 3rd party OAuth/credential management provider (Composio).
You log in.
Grant Alfred scoped access.
Done.
♾ Infinite Customization
You can still connect to other APIs/services.
Alfred includes a Vault:
•Store private keys and secrets
•Encrypted
•Not stored on your VM
•Only readable by your agent at runtime
🧠 Memory
✅ Simple Default
A basic, built-in, note based memory system.
♾ Infinite Customization
Bring your own context system. Build a better memory system from scratch. That's a rabbit hole you can still go down later if you'd like.
👤 Usage Mode
✅ Simple Default
Single-player.
Your apps only work for you.
Your Vault secrets are private.
Nothing is shared by default.
♾ Infinite Customization
Multi-player
Later, you can share app permissions and Vault secrets with other Alfred users in your network, enabling coordination across multiple users' emails, calendars, apps - whatever.
⚙ Infrastructure
✅ Simple Default
Connect Stripe
Alfred comes provisioned with a default Fly VM (shared 4x CPU/8GB RAM)
You’re live in ~30 seconds
Pay-as-you-go LLM API and VM usage billing
♾ Infinite Customization
Later, you can:
Import/Export your data
Resize your VM
Use your own LLM provider API keys
*What Alfred Doesn’t Do*
❌ No Default Heartbeat
You can schedule future or reoccurring agent sessions (Agent Cron Jobs).
But no background tasks runs automatically out of the box.
❌ No SOUL dot md/ Personality System
Alfred doesn’t come with a predefined identity layer.
You could build one using memory. But I left it out because it makes starting slightly easier.
❌ Platform Scope
Alfred is desktop web only.
No WhatsApp, Telegram, iMessage, etc. This one hurts, even for me.
But it also means you can get fully set up in literally 30 seconds, in the familiar cocoon of a laptop browser session.
❌ No Built-In Browser Agent
Alfred avoids this entirely.
Yes — you could build one. But if you're use cases need this, then you won't get much out of Alfred anyways. Remember, Alfred uses a 3rd party provider (Composio) for Terms of Service compliant app integrations as the default - not agentic web browsing.
In Conclusion:
- OpenClaw is a jet-powered motorbike🏍️🔥
- Alfred is a tricycle 🐢
I built this and am sharing because I think there might be some tricyclists out there…
#BuildInPublic
@thepatwalls If I'm ever on the show at some point, I'll happily receive the paper.
Made my kids some paper valentines and I can confirm that paper is still a thing.
I’m testing a daily word deduction game and trying to see if someone can complete one puzzle without instructions.
If you like Wordle/Jotto/Mastermind and want to try it out before it's released, DM me.
Using Opus 4.5 @claudeai a lot lately
Noticed it outputs VERY differently ❤️ from other LLMs:
- It's not ready to move on bc it still has questions
- It disagrees and pushes back
- It outputs EDITED and concise content, not word barf
I'm doing the same🤔 & getting SOLID outputs
I'm trying to figure out what to care about next. I joined Anthropic 4+ years ago, motivated by the dream of building AGI. I was convinced from studying philosophy of mind that we're approaching sufficient scale and that anything that can be learned can be learned in an RL env.
@Persist_Change@gregisenberg Yeah nah. That's just mindset. I'm 42 and have been in the kids' discord servers for over a year now.. Greg's right about it all. Take a deep breath and embrace this branch of tech. It's often freeing.
Gotta stay frosty when marking words to keep or ditch in a very long spreadsheet...
deawy - ditch
debag - ditch
debar - ditch
debit - wait... this isn't de-bit, it's deh-bit. Keep!
If you were me, would you include "buxom" and "busty" in your word list for your new word game?
I'm leaning toward 'no'... along with "boner" and "b!tch".
Curating a list of words to use in my new word game is kind of a fun way to spend time. Mindless, yes, but not the worst data entry I've ever done.
The worst? Subpoena data
@mobilecraft_xyz A bunch of the time wasn't on the interface but instead was on the psychology and order of the feelings I'm trying to create within the user.
Uploaded the full list of screen descs and @claudeai just batched it in sets of ~8 so I could review, commit.
It's crazy how good the LLM coding asst's are these days.
I've seen a massive difference in just the past 12 months.
This morning in 2.5 hours - made ALL of my 26 onboarding screens.
One of my favorite things to do is to talk with people working wherever I'm buying something. Restaurant? Talk w/a server or host. Convenience store? Clerk.
Bonus I found out recently: it's apparently proven to make you a happier person!
Everyone should visit New York City at least once in their life.
If you can, live in New York (any borough but Staten Island) for at least 4 months.
It's just ALIVE y'all. Visiting my brother atm and I miss living here.
When your wife handles the sick kids for 6 hours straight and then does dinner and bedtime on a Saturday while you bust out the majority of your app's product pivot.
None of us are self-made, my friends. And that's ok.