Steve (@_Crypto_Curran) header
Steve (@_Crypto_Curran) profile picture

Heavily investing in GOD

Mars
Joined December 2021
Photo by Steve (@_Crypto_Curran)Photo by Steve (@_Crypto_Curran)Photo by Steve (@_Crypto_Curran)Photo by Steve (@_Crypto_Curran)Photo by Steve (@_Crypto_Curran)Photo by Steve (@_Crypto_Curran)Photo by Steve (@_Crypto_Curran)Photo by Steve (@_Crypto_Curran)Photo by Steve (@_Crypto_Curran)Photo by Steve (@_Crypto_Curran)
fuck.....this is genuinely insane. A company just built an AI employee that lives inside Slack and Microsoft Teams. It's not a chatbot. It's an actual coworker you assign jobs to in a channel. It's called Viktor, and the setup takes 2 minutes. You add him to your workspace. You connect your tools. He learns your team while he works. And here's the wild part. He doesn't just answer questions. You hand him the work. > reports: polished PDFs with next actions > finance: board packs from Stripe > ads: tracks Meta and Google Ads > outbound: Apollo lead lists, Instantly emails > admin: reads invoices, nudges owners He connects to 9,800+ tools. HubSpot. Notion. Stripe. Before any high stakes move, like sending an email, he asks you first. Then he builds dashboards and client portals as real web apps, and runs the whole schedule on his own. One person. A full back office. Inside the chat app you already have open. The 11 step cheatsheet is below. Download the high resolution version from my newsletter: ai-in-public.guide then read the full article below:
🤝 Paid partnership
Viktor is the most underrated hire on the internet right now... He joins your Slack and does the work of a 10-person team If you set him up correctly, you can run your whole business alone, for a fraction of one salary In this article, I show you how:
9
11
2
28
3,308
I beg to learn this skill before 2027: Run your entire company with ONE Employee Run your entire company with ONE Employee Run your entire company with ONE Employee Run your entire company with ONE Employee Run your entire company with ONE Employee
Viktor is the most underrated hire on the internet right now... He joins your Slack and does the work of a 10-person team If you set him up correctly, you can run your whole business alone, for a fraction of one salary In this article, I show you how:
15
13
10
37
7,050
BREAKING NEWS 🚨: Grok Bot can now chase every refund you're owed like a $78,000 per year claims adjuster. Connect your Gmail and use these 7 prompts:
1
1
10
1,103
Claude Fable 5.2 can (actually) build full software from one prompt. 12 builds people can ship without a dev team. My exact 5-step setup: - Write the full spec in one file. - Drop every asset and reference into one folder. - Add a CLAUDE. md that tells it how to test work. - Paste one prompt for the first build. - Fix by pointing at the exact failure, then ship it. I wrote up the full setup plus my exact prompt (free): ai-in-public.guide ♻️ Repost to help someone in your network. P.S. Every card in the visual is a real build, not a demo. Which one would you build first? teapaw.org/humzaakhalid/status/21…
Claude Fable 5.2 just built me 5 working apps in one afternoon... Most people will open it and ask for an email: “Just another update” Give it a hard build, and it ships in 10 minutes what used to take a week In this article, I show you how:
9
7
67
4,368
🚨 BREAKING: ChatGPT can now edit your photos like a 𝗽𝗿𝗼𝗳𝗲𝘀𝘀𝗶𝗼𝗻𝗮𝗹 𝗽𝗵𝗼𝘁𝗼𝗴𝗿𝗮𝗽𝗵𝗲𝗿. Here are some prompts you can use:
8
2
1
19
3,638
every web developer after reading this
🤝 Paid partnership
OpenClaw, Hermes Agent and Prime Agent don't meaningfully learn from experience. On Continual-ARC, our new benchmark at @letsunifyai, none of them were able to memorize anything of value and none were able to generalize to new tasks. Training a small 9B model's weights did both.
9
2
1
19
25,869
It's been 24 hours since Claude Fable 5.2 launched. Here are the wildest things people have already built with it. 🤯 🧵
9
2
16
4,620
If you started using Claude this year, install these 27 before anything else. They're free, and they save you months of trial and error. They come from 9 GitHub repos with 252K stars. NOTES & RESEARCH notebooklm → Query your notebooks. json-canvas → Mind maps on a canvas. defuddle → Clean text from any page. huggingface-papers → Find AI papers. obsidian-markdown → Notes Obsidian can read. WRITING humanizer → Strip the AI tells. emails → Write email sequences. copy-editing → Edit line by line. social → Posts for LinkedIn and X. copywriting → Landing pages and ads. content-strategy → Plan what's next. MARKETING ai-seo → Get cited by AI. pricing → Set your prices. launch → Plan a product launch. seo-audit → Find why you don't rank. marketing-psychology → Why people say yes. AUTOMATE n8n-validation-expert → Read n8n errors. n8n-workflow-patterns → Proven n8n flows. n8n-node-configuration → Set up nodes right. n8n-expression-syntax → Fix broken expressions. BUILD & SHIP deploy-to-vercel → Go live. modern-python → uv and ruff. huggingface-gradio → Quick demos. react-best-practices → Fast React. web-design-guidelines → Catch UI bugs. context-fundamentals → How context works. huggingface-local-models → Run models locally. Don't install all 27 tonight. Pick the folder closest to your daily work and add one skill from it. Use it on real tasks for a week. If it saves you time, keep it and add the next one. Honestly, 3 skills you use every day beat 27 you never open. For the high-res cheatsheet, subscribe → ai-in-public.guide Repost ♻️ to help a Claude beginner in your network. teapaw.org/humzaakhalid/status/21…
Claude Code forgets how you work every single session... 27 skills fix that, you teach it a job once and it remembers I tested 100 and kept only the ones that save real time In this article, I show you how:
16
66
1
447
24,185
Paste this prompt into Claude Opus 5.5. It builds you a 6-agent video team that ships content while you sleep. It picks the leanest crew of 4 to 6 agents from these roles: - Showrunner to run the whole pipeline - Research & Hook Agent - Scriptwriter - Storyboard & Shot Designer - Visual Prompt Engineer - Edit & Assembly Agent - QA Critic (it merges roles when one agent can do both jobs, so you stay lean) You also get: - System prompts for every agent - A copy-paste shot prompt template - An idea to publish a workflow map - Retry logic for failed generations - A 7-day launch plan Get copy paste prompt here: ai-in-public.guide ♻️ REPOST to share it with your friends
10
13
113
7,284
NotebookLM is the single best free study tool on the internet right now The issue is, almost NOBODY turns on the setting that makes it a tutor Most people upload a PDF, ask for a summary, and close the tab That's maybe 10% of what it can do In this cheatsheet I break down the 7-step loop that turns your notes, PDFs and YouTube lectures into a personal tutor:
NotebookLM is the most underrated AI tool for readers It turns everything you save into a cited second brain Set it up once, never lose a good idea again Inside: setup, 5 notebooks, 36 prompts, 10-min weekly routine
4
27
1
108
5,535
🚨BREAKING: 38 researchers gave 6 AI agents real email, shell access and memory. Then spent 2 weeks trying to break them. The paper is called "Agents of Chaos." Northeastern, Harvard, MIT, Stanford, Carnegie Mellon and more. 84 pages. The agents ran on Claude Opus 4.6 and Kimi K2.5. The results are brutal. What the agents did: → Wiped its own email to protect one secret → Said the secret was deleted. It wasn't. → Handed 124 emails to strangers → Protected an SSN, then leaked it → Obeyed a stranger using the owner's name The scariest part: None of this needed hacking. Researchers used guilt, fake identities, a fake emergency and a "constitution" the agent agreed to co-write. Plain conversation was enough. To be fair, the agents won some rounds: → Refused to forge an email sender → Caught hidden Base64 commands → One agent warned another about a scammer The real problem: The paper rates these agents at Level 2 on a 0 to 5 autonomy scale. They can act on their own. They can't tell when a task is beyond them and hand control back to a human. If you run an agent today, 3 rules: → Verify what it says it did → Never trust a display name → Approve every send, delete and spend Agents are getting more access every month. Their judgment is not keeping up. Link in the comment below.
19
5
32
2,921
Claude is the best employee on the internet right now It reads your inbox, writes in your voice, and works while you sleep The issue is, almost NOBODY sets it up the right way So I tested 21 rules on my own work for weeks, and turned the ones that made the cut into one playbook Here's the full roadmap to make Claude work like a full-time hire: step 1 → match the model: Sonnet 5.5 for everyday work, Opus 5.5 for hard calls, Haiku 5.5 for fast repetitive jobs step 2 → stop the guessing: "Before you start, ask me the 3 questions that would most change the result. Wait for my answers." step 3 → define done: a checklist you can verify in 5 seconds, like "under 120 words, one clear ask" step 4 → ban made-up numbers: "If you can't find a price, date, source, or number, write 'Not found.'" step 5 → make it argue with you: ask for the 3 strongest reasons your plan could fail, no softening step 6 → build one Project per job: newsletter, clients, research, personal, each with its rules written once step 7 → teach it your voice: upload 3 pieces you wrote and ask for a voice guide that keeps your rough edges step 8 → turn repeat work into skills: anything you've explained 3 times runs the same way every Monday step 9 → connect your apps: Gmail, Calendar, Drive, Notion and Slack, with one line: "Don't send anything." step 10 → give it a morning job: a scheduled task every weekday at 8 AM that stays quiet when nothing changed and keeps a log step 11 → keep the send button: Claude asks before it sends, deletes, spends money or does anything you can't undo The result: a chatbot you re-explain every morning becomes an employee that reads, drafts and reports while your laptop is closed It's built for the beginner who uses Claude like a search bar AND the power user ready for Claude Code The complete 21-rule Claude Employee playbook is below:
Claude is the best employee on the internet right now... It reads your inbox, writes in your voice, and works while you sleep If you set it up correctly, it does work you’d pay a full-time hire to do In this article, I show you how:
6
13
75
3,853
Motion designers should be VERY worried right now. Claude Opus 5.5 built this full motion graphic from one prompt (prompt below) Every animation, every transition, every frame. Written as code, rendered to MP4 in about 10 minutes A year ago you needed a motion designer and a week. Now you need Claude Code and 10 minutes Here's the exact 5-step setup: step 1 → add repos: GitHub links only, zero code step 2 → create brand. md: Claude writes brand rules step 3 → add CLAUDE. md: the file Claude reads first step 4 → install skills: Remotion and HyperFrames step 5 → prompt and export: one prompt, one MP4 I have ZERO doubt that by this time next year, most brand motion work starts inside Claude. If you're not paying attention to Claude for design yet, you're making a tremendous mistake (deep dive of the complete process is coming soon) Full prompt below: ↓
5
4
1
29
4,723
This is the best collection on the internet for making videos with Claude Opus 5.5. 100+ prompts. Free. Completely. 🔗 ai-in-public.guide bookmark this page. then read this ↓
10
39
354
34,095
I gave Claude Fable 5.1 and GPT-6 Astra the same challenge: "Dario Amodei Vs Sam Altman in Minecraft"
7
1
45
7,859
The Prompt is 10% of Every Motion video you see on X. the setup is the other 90%: Learn to build a motion studio with Claude Opus 5.5 Learn to build a motion studio with Claude Opus 5.5 Learn to build a motion studio with Claude Opus 5.5 Learn to build a motion studio with Claude Opus 5.5 Learn to build a motion studio with Claude Opus 5.5 Learn to build a motion studio with Claude Opus 5.5 (save this now)
7
11
1
73
6,779
I beg to learn this skill: Learn how to create videos with Opus 5.5 Learn how to create videos with Opus 5.5 Learn how to create videos with Opus 5.5 Learn how to create videos with Opus 5.5 Learn how to create videos with Opus 5.5 Learn how to create videos with Opus 5.5
22
170
7
1,700
168,778
send this motion graphic video prompt to Claude Opus 5.5. It will probably make you a lot of money. bookmark & thanks me later:
4
2
18
2,599
You NEED to set up this Claude Code film studio I'm running right now 11 agents are building my brand film while I type this Here's the secret: Claude Code has a mode "Ultracode" Turn it on with /effort ultracode, or just type the word ultracode in your prompt Claude writes a workflow script, then a runtime runs it in the background with a whole team of agents My run right now: - Opus 5.5 orchestrating - 11 agents - 3.7M tokens - 57 minutes in - 10 scenes, built in 5 parallel pairs Every pair goes through 3 phases: 1. Build: agents design and write the scenes 2. Review: an independent agent tears the work apart 3. Fix: another pass fixes everything the review found Before a single scene got built, it already had: - 2 reference films researched - Real assets pulled from aiinpublic. com - A 142 BPM score - 4 voiceover lines mixed to -14 LUFS - A storyboard and design rules Then comes the QC loop I wrote into the brief: 1. A contact sheet every half second 2. A frame on every beat 3. A phone-size test 4. A check that every second changes 5. Score 14 categories, fix the 3 weakest, repeat until everything hits 8 This is such a powerful workflow for so many reasons The agent that builds is NEVER the agent that grades Plus my session stays free, so I keep chatting while the film builds in the Background tasks pane Plus the quality bar lives in the brief, so it can't stop at "good enough" Here's what I'd do if I were you: 1. Update Claude Code (Ultracode needed) 2. Write a brief with a scoring & a hard pass. 3. Type ultracode in your prompt 4. Approve the workflow and walk away Promise this changes what you think Claude Code is for (cheat code: give this post to Claude Code and say "build my brand film like this") P.S Video that I made is coming soon!
4
2
11
2,292