Indra Prasad Ghatak

I'm giving away a FULL course on how to build a managed AI agent business solo using Hermes Agent, Orgo, Obsidian, Codex, Claude Code etc. Here's everything (47 minutes): 1. The offer: unlimited agents, unlimited usage, all infrastructure and security included. The customer gets a digital employee. They never think about tokens or models. You handle everything. 2. Don't niche down too fast. Try marketing agencies, law firms, insurance, manufacturing, real estate. See where the market pulls you. Then go vertical. Diverge first, converge later. 3. Every executive has the same problems regardless of industry. Too many emails, too many meetings, too many follow-ups, too many open loops. Solve those first. Then layer in vertical-specific skills. 4. The stack: Hermes Agent for the agent harness. Codex or Claude Code desktop to build and configure. Orgo for cloud computers so every agent lives in its own sandbox. Composio for one-click authentication across thousands of apps. Agent Mail to give every agent its own email. Obsidian for the knowledge base. 5. Use agents to build agents. Don't stress about setup. Use Claude Code or Codex to install and configure Hermes inside a VM. Use Perplexity MCP, Context7, and Exa for up-to-date docs. Your agent sets up your customer's agents. 6. GPT 5.5 is the best model right now. Efficient with tool calls. Doesn't eat tokens like Opus 4.7. For cheaper tasks, GLM 5.1 from ZAI is the best open source option. 7. Set up watchdogs for gateway crashes so they auto-restore. Have agents email you when cron jobs break or skills fail. Your customer should never have to tell you something is broken. 8. Get customers through content. If someone jumps on a call and already knows who you are and what you sell, that's the position you want. Content is the most leveraged thing you can do in 2026. 9. Keep scope tight. One to two requests at a time, delivered in under 48 hours. Use Trello for customer-facing project management. Send Loom updates at random hours to show you're always working on their agents. 10. If you can set up Claude Code, Hermes, or OpenClaw, you have a skill that 99% of business owners don't have and would pay $5k/month for. You're probably not giving yourself enough credit. shoutout to @nickvasiles from @orgodotai for coming back on @startupideaspod and sharing the full playbook. tools, stack, fulfillment, everything. this type of episode isn't shared anywhere on the internet. this is the alpha people keep for themselves. i will keep sharing if you keep watching. you could watch netflix or you can watch this (link below) youtu.be/BI-MNjm1tTQ?si… watch

@xynth_m Xynth

Xynth can now scan the stock market for you 24/7 ! Simply describe what you want monitored in plain English. Under the hood, we wire Claude Opus 4.7 + Python to 3,000+ live market endpoints to build your custom alert. The workflow lives in the cloud, hunting your setup the moment it hits. As part of this launch, we're giving free access to the top 5 most profitable alerts built so far. RT + comment "Xynth" below to get access ↓

someone just open-sourced an entire AI company. 147 agents. 12 departments. engineering, design, marketing, sales, finance, QA, support. each agent has a role, a workflow, deliverables. not a chatbot. a team. claude code, cursor, copilot, gemini cli, windsurf. 1 command to deploy everything. started as a reddit thread. 88,000 stars. 14,000 forks later. MIT. free. yours. this should cost $50,000/year

A Stanford lecture just exposed where the ~$650B AI data center spend is actually going and who is quietly capturing the upside They explain why the entire bottleneck has shifted from GPUs to power and what comes next This is a MUST watch for investors:

Andrej Karpathy: "To get the most out of the tools that have become available now, you have to remove yourself as the bottleneck. You cannot be there to prompt the next thing. You need to take yourself outside the loop. You have to arrange things such that they are completely autonomous. The more you can maximize your token throughput and not be in the loop, the better. This is the goal. So, I kind of mentioned that the name of the game now is to increase your leverage. I put in very few tokens just once in a while, and a huge amount of stuff happens on my behalf." --- From @NoPriorsPod YT channel (link in comment)

@DavidCarbutt_ YouTube

YouTube has transformed my life. It allowed me to quit my job, grow a business and give me strong financial security. I’ve also helped a huge amount of content creators & businesses make a lot of money through YouTube. I have dumped all of my YouTube knowledge into an 8 week course, which I will add to continuously. We are opening up to just 20 people who want to learn how to grow and make money on YouTube. If you’re interested: - Follow me (Important so that we can DM you) - comment ‘YouTube’ below.

Que gran articulo, pense que era mucho mas simple armar mi propio harness, habia varias cosas que no habia ni contemplado, es todo un mundo, interesantisimo, mucho mas que los LLMs en si

Here's @rywiggs (VP Mercury) walking through how he built his Claude Code second brain at work: "I pulled together almost 5M words from my five years at Mercury and built that as a knowledge base. At the start of everyday, I get a brief of what's on my calendar, Linear, Slack and at the end of the day it summarizes [everything]." 📌 Watch him talk more about it here: youtu.be/KzqpK1uCczw?si…

OpenClaw is the fastest-growing open source project, but there are no stories of running it safely in production at scale. As we started deploying agents internally at @brexHQ, we couldn’t stop thinking about this question. Agents work, but nobody wants to give them real credentials. Instead of waiting for a solution to emerge, we decided to try a novel approach: using LLMs to judge the network traffic of an AI agent. Today we’re announcing CrabTrap, an open-source proxy that intercepts every outbound request and blocks risky activity using LLMs, before it ever hits an external API. The results are promising; we believe it’s a meaningful step forward in the security of agent harnesses in production environments. Try it out today. (As a side note, it was really fun to work personally on a real systems problem again. And btw, if you want to work at a place where the CEO is building proxies at night, we’re hiring!)

This setup makes Claude 10x smarter and will dramatically improve the quality of your AI outputs. I recently gave Claude perfect memory - and it's been a complete game-changer. My Claude now remembers everything, never drops context, and gets smarter over time. Complete setup:

UI/UX Designers, here are my go-to best design resource sites on the internet you should bookmark: Design Library → curations.supply Landing Pages → landing.love Saas Websites → saaspo.com AI Mobile App Builder → sleek.design/?ref=abrahambw Fonts → uncut.wtf Animation → 60fps.design Mobile Apps → mobbin.com/?via=abraham Brands → rebrand.gallery Icons → hugeicons.com/?via=Abraham Design Systems → component.gallery

Pregnancy is severely underappreciated in our society. Most women don't get the guidance they need from doctors. Self experimentation gives the power back.

🧵 My tips for getting the best results out of Claude Design! I’m on the verticals team at Anthropic which means I serve 7 different products. Claude Design makes it possible! 1. Set up your design system and your core screens. An hour of setup and refinement here is worth it

This post is so so good. We are at this interesting point where we’ve started to really figure out ‘memory’ with these LLM-based systems. And @hwchase17 is totally right - memory is basically just context collected and injected at the right times - it’s probably the most important context, and it *must* be interpretable to you, and portable for you. There is a real danger in adopting systems that won’t allow this for you, or actively prevent you from doing it. I hadn’t put it all together before reading this post, but this is definitely one of the reasons I love building on top of DeepAgents

Most PMs tried Claude Code for a day, didn't get instant magic, and quietly decided it wasn't for them. The PMs pulling ahead are 1500 hours in and still rebuilding their setup every single day. That's the entire gap. Not talent. Not technical background. Just whether you stayed past the awkward week where nothing works yet. Hannah runs product at Anthropic and has the highest documented Claude Code mileage of any PM I've talked to. Her advice for someone with two hours this weekend isn't "build a workflow." It's "find one task to automate so you free up six hours next week to learn." That reframe is the whole game. Most people treat AI learning as something they'll get to after the real work is done. Hannah treats freeing up time to learn AS the real work. Two hours in, six hours out. Next week you reinvest those six into deeper automations that free up fifteen. The compounding only starts if you survive the first week. And almost nobody does, because day-one Claude Code feels mediocre. Your context isn't loaded. Your skills aren't written. Your CLAUDE.md is empty. The tool is guessing about your role, your product, your standards, everything. The PMs at 1500 hours aren't smarter than the ones who quit on day two. They just didn't quit on day two. Every PM interview at a frontier AI company in 2026 is some version of "show me your setup." The honest answer for most people right now is "I tried it once." Build the hour. Then build the loop.

> use Claude Code > no plugins > "I don't need extensions" > install one plugin out of curiosity > it does in 2 seconds what took you 30 minutes > install second plugin > same thing > you mass typed what a plugin does in one click for months

1.Download Obsidian → create a new vault (folder) 2.Download Claude Desktop (Claude Code) 3.Point Claude to the path of your vault 4.Paste the prompt from the article > Just 4 steps and 10 minutes of your time The face I made when I realized I had been suffering all this time, not knowing such simple things

AI marketing in 60 seconds: 1. most marketers use AI to write copy faster, that's level 1. you're doing the same job, just quicker. 90% of people stop here 2. level 2 is using AI as a thinking partner. market research, positioning, competitor analysis, pricing. anthropic's own growth marketer uses it this way. if you're only using it for output you're missing what it's best at 3. level 3 is doing things that weren't possible before: 100 SEO pages in a week, personalized outreach at scale, quality passes on every draft before it ships. ole lehmann took his landing page from 56% to 92% pass rate overnight using an AI optimization loop 4. level 3 agents you can build right now: - SEO agent reads search console + ahrefs daily, finds keyword gaps, writes and publishes pages - search term agent combs every query in your google ads account, finds negatives across tens of thousands of terms - competitor ad spy monitors linkedin and meta, tracks who engages, builds warm lead lists - cold outreach personalization agent researches each prospect, writes custom first lines from their content - site QA agent crawls your site daily for dead URLs, broken UTMs, pixel misfires - trial conversion agent connects to stripe, flags who should convert this week, drafts retention emails - content repurposing agent, one webinar in, blog posts + social clips + email sequences out - campaign performance agent pulls GA4 + meta + google ads into one view, writes a daily digest - positioning agent tracks competitor websites, pricing pages, and messaging weekly, reports shifts before you notice them - review mining agent reads G2, trustpilot, reddit, and app store reviews, pulls objections and testimonials for your copy - tools page agent builds calculators and comparison pages for SEO traffic and link magnets 5. level 4: agent teams. research feeds strategy, strategy feeds content, content feeds distribution. one person directing specialists running in parallel. the output of a full agency from a laptop 6. the agent marketing department pattern: eight functional areas (research, strategy, content, SEO, paid, email, community, analytics), each with two to five agents sharing context through a shared knowledge base. one human oversees each department 7. level 5: agents run campaigns, read the data, adjust spend, and report weekly. one stack doing \$73K/month from agentic social scheduling alone, another doing outbound with zero humans in the loop 8. everyone has the same tools. the gap between L1 and L5 is taste and judgment. AI gets you to 60% instantly. that 60% is the new floor because everyone hits it, the revenue is in the last 40% 9. what taste looks like: knowing which of 20 AI-generated variants to ship, rejecting 80% of the output even when it looks fine, catching the line that sounds like it could be from any company 10. banned word lists miss the point on slop. slop is content with no point of view, no specificity, nothing at stake. if you can swap your brand name for a competitor's and the post still makes sense, you wrote slop 11. distribution is the moat. vibe coding solved building, not shipping. anyone can launch a product in 48 hours. getting 10 paying customers still takes weeks 12. when AI agents start buying from other AI agents, distribution is the only reason they find you first. own the channel, own the data, own the brand 13. the AutoReason pattern: three markdown files. one for goals, one for scoring, one for the experiment log. point an agent at them and let it run variations against your copy, your ads, your onboarding. compound improvement without you touching it 14. rick rubin rule: your job is taste now. deciding what stays, what gets cut, what gets refined

If you don't own the memory, you don't own the agent: - memory is what makes your agent get smarter over time - without it, anyone with the same tools can copy your agent overnight - with it, you build a dataset no competitor can replicate - closed memory = your data on someone else's servers - switch models, lose everything your agent learned - model providers are incentivized to lock you in via memory - the model is easy to replace, memory is not - if you don't own the harness, you don't own the memory - if you don't own the memory, you don't own the agent full story of why this matters and what happens when memory is locked behind someone else's API 👇

I keep telling my team that coding isn't needed to use AI in workflows. But, they do need the 101 basic AI fundamentals. I send them this 1-hour Stanford lecture from an AI program designer and principal ML researcher at GitHub. He walks through everything I've seen actually move the needle for marketers working with AI. Then come back to Claude Cowork, you’ll be ready to delegate.