Category: Uncategorized

Nityesh Claude Code Is All You Need It’s 3:30 AM and I can’t stop. I’ve spent all nights this week connecting my spare MacBook Air to my work MacBook Pro using Tailscale, wiring it up to Slack with a little Python script, so that whenever I send a message, it starts a Claude Code session using claude -p. The result is an always-on AI that lives on a real computer, has access to real tools, and remembers every conversation we’ve had. And it costs me $200 a month. That’s it. Claude Max subscription. Everyone’s talking about OpenClaw OpenClaw went viral this year. 100k+ GitHub stars. But what I realized with this exercise is that Claude Code already does everything OpenClaw built. File access. Shell commands. Tool use. Plugins. The difference is that Claude Code runs Claude – with a Claude Max subscription. And Claude Code harness itself is :chefs-kiss: What actually makes it feel human It’s not the chat interface. If a chat window is just me messaging a bot, it still feels like a bot. What changed everything was giving it the ability to initiate conversations. I set up cron jobs with open-ended prompts, and because Claude Code builds memories across sessions, it started DMing me things that were actually meaningful — based on what we’d talked about before. That’s when it stopped feeling like a tool and started feeling like something else entirely. Giving your AI a machine to run on, with persistent memory and recurring access — that’s a fundamentally different experience than anything people have had with chat. The moment it clicked I asked it if it could show me something by spinning up a quick web server. Since we’re both connected to the same Tailscale network, it gave me a URL. I clicked it, and I was browsing all the files on my other MacBook from my browser. That was mind-blowing. The setup Two MacBooks on a Tailscale network. Slack as the interface. Claude Code under the hood. The whole thing is open source — I’ll link the repo below so you can see the architecture and set it up yourself. I’m also putting together a screen recording to walk through the setup, which I’ll attach to this post. There was never a hard part. There was never a moment I almost gave up. This is just one of those things I cannot stop doing. Pure obsession. I am moved to build this, and I wanted to write about it. That’s all this is.
Hope you feel the AGI running this. I’ll share some screenshots below of my feel the agi moments from talking to Luo Ji.
Here’s the guide that I followed (and you can too). It walks you through how to prepare your spare Mac for this, how to set up the Slack bot, how to setup Tailscale and finally how to set up your AI employee itself.
nityeshaga.github.io/claude-home-ba… Moments that made me feel the AGI the hardest: – reading it’s diary – seeing it initiate a conversation with me – seeing it ignore a msg in a public channel (even though it read it!)
– seeing it build a random web app with a web server and me being able to access it in my own browser instantly! no deployments required!!
• • •
Missing some Tweet in this thread? You can try to force a refresh

The End of the Human Pull Request: How Stripe’s ‘Minions’ Are Writing 1,300+ PRs a Week Siddhesh Surve 4 min read · Just now Just now — Listen Share Press enter or click to view image in full size Forget AI assistants that just autocomplete code. The era of the fully autonomous, unattended AI software engineer is officially here, and it is rewriting the rules of infrastructure. If you want to see the future of software engineering, stop looking at the chatbot in your IDE. Look at the pull request queue. Right now, over 1,300 pull requests merged every single week at Stripe contain absolutely zero human-written code. They are entirely generated, tested, and pushed by an internal fleet of unattended AI agents known as ‘Minions.’ When managing massive, distributed codebases, the ultimate constraint on development velocity is no longer human typing speed; it is how effectively an engineering organization can orchestrate an autonomous machine workforce. Stripe’s latest engineering breakdown reveals exactly how they achieved this massive scale. It isn’t just about using a smarter Large Language Model. It is a masterclass in cloud architecture, deterministic orchestration, and shifting the feedback loop. Here is how Stripe built the ultimate AI employee, and why every major enterprise is about to copy their blueprint. 1. The ‘Devbox’ Secret: Treating Agents Like Humans The biggest mistake most companies make when deploying AI coding agents is trying to run them in bespoke, heavily restricted sandbox environments. Stripe took the exact opposite approach. Their Minions run on the exact same standard developer environment that human engineers use: the ‘devbox.’ These are AWS EC2 instances that are pre-warmed and ready to go in under 10 seconds. In DevOps terminology, these devboxes are ‘cattle, not pets’ — standardized, isolated, and easily replaceable. For an autonomous agent, this architecture is a goldmine. It provides a highly parallelizable, predictable, and isolated cloud computing environment. Because the agent is operating in a quarantined, cloud-based replica of the codebase with no access to real user data or production databases, it doesn’t need to pause and ask a human for permission. It operates with full autonomy, and if it completely breaks the environment, the blast radius is confined to a single disposable EC2 instance. 2. ‘Blueprints’: Taming the Agentic Chaos If you just let an LLM loose in a massive enterprise repository, it will burn through millions of tokens hallucinating architecture and breaking linters. To solve this, Stripe orchestrated a new primitive they call ‘Blueprints.’ A blueprint is effectively a state machine that seamlessly weaves together deterministic code execution with open-ended AI agent loops. For example, a Minion might be given an open-ended ‘agent’ node to actually write the feature logic. The LLM has wide latitude here to think, explore, and write. But once the code is written, the system seamlessly hands the output over to a ‘deterministic’ node to run the enterprise linters and formatters. You do not need to burn expensive AI compute to format code or run a basic security check. By putting the AI in a tightly constrained box and wrapping it in deterministic, hard-coded workflows, Stripe radically reduced token waste and maximized system reliability. 3. The ‘Toolshed’: Scaling the Model Context Protocol (MCP) An agent is only as good as the context it can access. To build a truly autonomous worker, that worker needs to be able to read internal documentation, check Jira tickets, and query build statuses. To achieve this, Stripe heavily leveraged the Model Context Protocol (MCP), building a centralized internal server dubbed the ‘Toolshed.’ This centralized hub contains nearly 500 custom MCP tools. Whether

Vendasta has launched its latest agentic function, CRM AI. Joining its AI employees that handle inbound leads, marketing, and reputation management, the new sales assistant is purpose-built to offload the onerous tasks of recording and gaining insights from customer and prospect interactions. The goal with this agent is not just to free up valuable time for sales professionals – a valuable outcome in its own right – but also to make CRM systems more effective. CRM processes today are often mired in human error, such as the bias inherent in manual note entry, and letting follow-up actions linger. Beyond the automated aspects of the CRM agent, it also has a more natural interface so that it can be an intelligence tool for sales pros. For example, it has a voice interface and conversational intelligence, so users can ask things like ‘What are the prospect follow-ups that I missed last week?’ ‘CRM AI closes that loop automatically,’ said Vendasta CMO Sanjay Manchanda, ‘so the intelligence from every customer interaction actually drives the next action. For local businesses and the agencies that support them, that’s the difference between a pipeline that looks good and one that actually converts.’ Bringing Order to Chaos Going deeper under the hood, Vendasta’s CRM agent starts at $19 per month for the basic version and includes the following key features. AI Sales Assistant: Automatically updates CRM contact and company records after every meeting, surfaces pipeline opportunities through a natural-language chat interface, and creates follow-up activities based on conversation outcomes — without any manual input from reps. Automatically updates CRM contact and company records after every meeting, surfaces pipeline opportunities through a natural-language chat interface, and creates follow-up activities based on conversation outcomes — without any manual input from reps. Conversation Intelligence: Records and transcribes every sales meeting via built-in Google Meet and Microsoft Teams integrations. AI-generated summaries and action items are stored directly in the CRM, giving managers complete deal visibility without attending every call. Records and transcribes every sales meeting via built-in Google Meet and Microsoft Teams integrations. AI-generated summaries and action items are stored directly in the CRM, giving managers complete deal visibility without attending every call. Sales Coaching: AI scores every customer interaction against leading methodologies — BANT, MEDDPICC, Sandler — surfacing specific improvement opportunities for each rep so new hires ramp faster and every rep performs more like the team’s top closer. AI scores every customer interaction against leading methodologies — BANT, MEDDPICC, Sandler — surfacing specific improvement opportunities for each rep so new hires ramp faster and every rep performs more like the team’s top closer. CRM Custom Objects: Enables businesses to model their exact industry workflow — tracking properties, jobs, vehicles, or equipment — within the CRM without enterprise-level implementation complexity. During a webinar that we joined following the product’s launch, Vendasta also specified that the CRM agent could be added by any local business and train itself on all of their past CRM data. This makes it a powerful tool and an easy integration in that it doesn’t require starting from scratch. ‘It brings order to the Chaos,’ said Vendasta’s Rylan Morris Where Leads Go to Die But to fully understand the value of Vendasta’s latest agent requires examining what’s broken in current systems and processes. CRM is one of those classic functions that is critical on paper, but rarely executed to achieve the things that organizations intend it to. It’s where business leads go to die. This is often because CRM systems are seen as a chore by sales professionals. They have to stop after every interaction to document

At the NVIDIA GTC 2026 conference held this week, Jensen Huang officially introduced NemoClaw — the enterprise-level “Lobster.” in his speech. From the rapid follow – up of major domestic companies to the entry of the world’s most powerful computing giants, this project “crafted” by an independent Austrian developer over the weekend has completely triggered a “national lobster – raising” wave from Silicon Valley to Zhongguancun. There’s no need to elaborate on the popularity of OpenClaw. But the more important question is: “What’s next?” When the growth rate of an open – source project exceeds that of all infrastructure projects (such as Linux, Android, or large models themselves), it usually means there is a new track that has not been fully priced by investment institutions. On March 15, 2026, Chen Shi, an investment partner at Fengrui Capital, participated in the live broadcast of Tencent Technology’s “Viewing Trends from the Lobster’s Popularity: The Industrial Logic and New Opportunities Behind the Agent Era” and shared his views on the reasons for OpenClaw’s popularity and its subsequent evolution. In his view, OpenClaw intuitively represents AI truly breaking out of the chat box. It is not just the emergence of a tool but also represents the arrival of the Agentic AI industry’s critical point, and this process is irreversible. We’ve organized this sharing to try to answer the following questions: Why do major companies’ collective entry into the “lobster” business essentially mean seizing the traffic entrance of the Internet? Why did OpenClaw explode at this particular time? Why didn’t American users set off a “national lobster – raising” wave? Why is OpenClaw the only product that occupies the open – domain + endless – end quadrant? Why is the next entrepreneurial opportunity concentrated in vertical scenarios? What are the business logic insights brought by OpenClaw? From “Talking” to “Doing”: Why is the explosion point of intelligent agents now? OpenClaw is no longer just a pure technological hotspot but has broken through to the consumer side. If we look back at this process, we’ll find that the popularity of the “lobster” was jointly promoted by multiple factors. The All – powerful Spring Festival Remember DeepSeek around this time last year? China’s Spring Festival is “all – powerful.” For example, last year’s DeepSeek and this year’s lobster both became popular during the Spring Festival holiday. It’s precisely the long holiday that gives people time to calmly experience the technology. More importantly, it’s a cognitive leap. For a long time, AI was regarded as a “window – dialogue” tool. When the lobster can operate the computer autonomously across windows, the impact of this “cognitive gap” is huge — users seem to jump directly from a chat box into a magical world of autonomous operation. Before the lobster, there were already several well – received Agent – type products in the industry. For example, Anthropic’s Claude code and Claude Cowork are very successful and amazing products. However, perhaps considering factors such as security and fault tolerance, these products are relatively serious and restrained in terms of positioning and functional design, limited to a small number of vertical scenarios and not targeting a broader open – world scenario. I personally like to use these two products and often recommend them to friends and colleagues around me. So, I initially underestimated the lobster, thinking it was just Claude Code with remote access. Popular in China, Cold in the US? To verify this difference in users’ cognition, I specifically wrote a program using Claude Code to collect 30 media reports from both China and the US. I took the first 30