Super User Daily: Mar 20, 2026
Today's standout: a user designed custom furniture entirely with Claude Code and handed the PDF directly to a carpenter — zero modifications needed. Meanwhile, Claude's new Dispatch feature is reshaping how power users treat AI as a persistent operator, not a chatbot.
---
@om_patel5
https://x.com/om_patel5/status/2034099391796036082
A user described a full wall unit to Claude Code — wardrobe, drawers, mirror, fragrance display, hidden vault, laundry section — all in one piece. Claude wrote 1,400 lines of Python that generates carpenter-ready technical drawings as a PDF: front elevation, plan view, construction notes, all dimensioned in centimeters. LED lighting positions, ventilation specs, push-to-open latches, soft-close hinges were iteratively added. The PDF was handed directly to the carpenter and built exactly to spec with zero modifications. This is the most compelling non-coding use of Claude Code seen to date.
---
@levelsio
https://x.com/levelsio/status/2034353674613400047
Pieter Levels discovered that running Claude Code locally drains a MacBook Pro M4 from 100% to 0% in about 3 hours, even when idle. His solution: run Claude Code on a VPS server and SSH in via Termius. The laptop then uses 20x less power, essentially becoming a thin client. He later bought a MacBook Neo specifically as a dumb SSH terminal for this workflow — no local dev environment needed anymore.
---
@emollick
https://x.com/emollick/status/2034067677157679379
Ethan Mollick reports that Claude Cowork Dispatch covers 90% of what he was using OpenClaw for, but feels far less likely to accidentally expose his files to a malware site. This is a significant endorsement from one of the most influential AI commentators, signaling that Anthropic's native agent is rapidly closing the gap with open-source alternatives.
---
@L_go_mrk
https://x.com/L_go_mrk/status/2034216492598972855
A detailed walkthrough of making Claude remember your entire business across sessions. The setup: store all company knowledge (revenue model, org chart, pricing, KPIs, sales process, brand voice) as individual markdown files in Obsidian with wikilinks connecting related topics. Then connect Claude Code, which reads files directly from your local machine — no uploads, no cloud, no file size limits. One instruction file teaches Claude how to navigate the knowledge base. Result: every session starts with full context. Three tools total: Obsidian (free), Claude Code (already paid), one instruction file.
---
@kemchenj
https://x.com/kemchenj/status/2034068917614993898
A new engineering leader used Claude Code to upgrade a 10-year-old PHP/Laravel project to the latest version — despite barely knowing PHP. The team was skeptical and publicly doubted it would work. The follow-up post: the upgrade landed smoothly in production. A powerful example of AI enabling non-experts to execute complex legacy modernization tasks.
---
@masahirochaen
https://x.com/masahirochaen/status/2034282341745807765
A power user running 9+ Claude Code projects simultaneously in Cursor with 10+ terminals always active. Key tips: input as much as possible per prompt to reduce back-and-forth rounds, and focus on identifying which parts of your work to delegate to Claude Code. The bottleneck isn't the tool — it's recognizing automation opportunities in your daily workflow. Also notes that consolidating communication to email/Slack (away from LINE/Messenger) dramatically improves AI-assisted workflow efficiency.
---
@wasserstein_rao
https://x.com/wasserstein_rao/status/2034065184285737454
A researcher is using Claude Code to directly control a liquid handling robot in a lab setting. No middleware, no custom integration layer — Claude Code sends commands to the robot hardware. This represents one of the most novel physical-world applications of coding agents: AI-driven laboratory automation.
---
@tonkotsuboy_com
https://x.com/tonkotsuboy_com/status/2034177733375430667
Highlights Anthropic's official /feature-dev skill as a hidden productivity gem. Type /feature-dev followed by a vague feature idea, and Claude Code analyzes your codebase, asks clarifying questions, then implements once requirements are clear. The user calls it a brain-activation tool: when motivation is low, the iterative questioning process forces you to articulate what you actually want, kickstarting both clarity and momentum.
---
@JJEnglert
https://x.com/JJEnglert/status/2034329261960475086
A detailed restructuring guide for Claude Code skills. Instead of one monolithic file, break skills into a folder system: SKILL.md as orchestrator (no rules, just workflow), instructions/ for rules (one per concern), examples/good/ and examples/bad/ for reference, eval/checklist.md for automated pass/fail tests, and eval/advisory-board.md for AI reviewer personas that evaluate drafts in parallel. Includes a ready-to-paste prompt for Claude Code that restructures existing skills in 5 phases.
---
@kazzkiq
https://x.com/kazzkiq/status/2034301953534988466
Shares a prompt that turns Claude Code into a professional code reviewer for your entire codebase. It reviews every file, page, endpoint, and logic path, then outputs a QUESTIONS.md with all architectural, refactoring, and security concerns. You answer the questions in the same file, then prompt Claude again to implement improvements based on your answers. A structured human-in-the-loop code review workflow.
---
@Padday
https://x.com/Padday/status/2034387631149039774
Intercom's VP confirms that how they build software is unrecognizable compared to 12 months ago. The team is fully Claude Code pilled and seeing enormous productivity gains. Enterprise-level validation that Claude Code is transforming professional software engineering workflows at scale.
---
@iwhale
https://x.com/iwhale/status/2034198865117839807
A Thai power user built an OpenClaw-powered affiliate marketing system for Shopee. The agent (using Kimi model as its brain) autonomously finds trending posts on X, analyzes what content to write, generates posts in a preset tone, and manages memory for context continuity. Runs 24/7 on a cheap VPS. Cost: X Premium Basic + AI/VPS fees. Revenue: passive affiliate commissions from Shopee's 7-day cookie window. Key insight: use a main agent (expensive model) to direct sub-agents (cheap models) to minimize costs.
---
@tvytlx
https://x.com/tvytlx/status/2034129973615989040
Identifies a likely profitable business opportunity: customizing OpenClaw for parents to tutor their children. Key features: parents can monitor conversations through separate channels, control the teaching style via soul.md, and design skills for generating quizzes, scoring, and reward systems. Based on a viral post about OpenClaw tutoring that got over a million views.
---
@heygurisingh
https://x.com/heygurisingh/status/2034180608210112824
Highlights stop-slop, an open-source SKILL.md file (809 GitHub stars) that makes AI writing undetectable. It bans 50+ AI filler phrases, kills structural clichés, forces sentence rhythm variation, and scores drafts on a 50-point scale. Works in Claude Code, Cursor, or any system prompt. No SaaS subscription — just one markdown file that outperforms paid humanizer tools.
---
@santtiagom_
https://x.com/santtiagom_/status/2034295951586689484
A cost-saving workflow for Claude Code: use /model opusplan to get a hybrid approach where Opus handles planning and architecture while Sonnet executes the code. You get Opus-quality reasoning where it matters most without burning expensive tokens on code generation. Combined with /plan to separate the thinking and building phases.
---
@masp_007
https://x.com/masp_007/status/2034068866876510607
A user built a high-spec outbound calling tool with Claude Code — through pure daily practice without tutorials or courses — that can obsolete most commercial calling tools on the market. Already deployed internally. A testament to what consistent daily AI coding practice can produce.
---
Eco Products Radar:
OpenClaw (267 mentions) — Open-source autonomous AI agent framework, most-starred on GitHub
Claude Code (257 mentions) — Anthropic's terminal-based AI coding agent
Cursor (28 mentions) — AI-native code editor
Obsidian (12 mentions) — Local-first markdown knowledge base
Vercel (11 mentions) — Frontend deployment platform with AI agent skills plugin
Codex (9 mentions) — OpenAI's coding agent CLI
Gemini (8 mentions) — Google's AI model family
Dispatch (7 mentions) — Claude's new phone-to-desktop remote control feature
Ollama (6 mentions) — Local model runner
MiniMax M2.7 (5 mentions) — Self-improving Chinese AI model
Stitch (5 mentions) — Google's AI-native design tool
n8n (4 mentions) — Open-source workflow automation
Superpowers (3 mentions) — Agentic skills framework by @obra
← Back to all articles
---
@om_patel5
https://x.com/om_patel5/status/2034099391796036082
A user described a full wall unit to Claude Code — wardrobe, drawers, mirror, fragrance display, hidden vault, laundry section — all in one piece. Claude wrote 1,400 lines of Python that generates carpenter-ready technical drawings as a PDF: front elevation, plan view, construction notes, all dimensioned in centimeters. LED lighting positions, ventilation specs, push-to-open latches, soft-close hinges were iteratively added. The PDF was handed directly to the carpenter and built exactly to spec with zero modifications. This is the most compelling non-coding use of Claude Code seen to date.
---
@levelsio
https://x.com/levelsio/status/2034353674613400047
Pieter Levels discovered that running Claude Code locally drains a MacBook Pro M4 from 100% to 0% in about 3 hours, even when idle. His solution: run Claude Code on a VPS server and SSH in via Termius. The laptop then uses 20x less power, essentially becoming a thin client. He later bought a MacBook Neo specifically as a dumb SSH terminal for this workflow — no local dev environment needed anymore.
---
@emollick
https://x.com/emollick/status/2034067677157679379
Ethan Mollick reports that Claude Cowork Dispatch covers 90% of what he was using OpenClaw for, but feels far less likely to accidentally expose his files to a malware site. This is a significant endorsement from one of the most influential AI commentators, signaling that Anthropic's native agent is rapidly closing the gap with open-source alternatives.
---
@L_go_mrk
https://x.com/L_go_mrk/status/2034216492598972855
A detailed walkthrough of making Claude remember your entire business across sessions. The setup: store all company knowledge (revenue model, org chart, pricing, KPIs, sales process, brand voice) as individual markdown files in Obsidian with wikilinks connecting related topics. Then connect Claude Code, which reads files directly from your local machine — no uploads, no cloud, no file size limits. One instruction file teaches Claude how to navigate the knowledge base. Result: every session starts with full context. Three tools total: Obsidian (free), Claude Code (already paid), one instruction file.
---
@kemchenj
https://x.com/kemchenj/status/2034068917614993898
A new engineering leader used Claude Code to upgrade a 10-year-old PHP/Laravel project to the latest version — despite barely knowing PHP. The team was skeptical and publicly doubted it would work. The follow-up post: the upgrade landed smoothly in production. A powerful example of AI enabling non-experts to execute complex legacy modernization tasks.
---
@masahirochaen
https://x.com/masahirochaen/status/2034282341745807765
A power user running 9+ Claude Code projects simultaneously in Cursor with 10+ terminals always active. Key tips: input as much as possible per prompt to reduce back-and-forth rounds, and focus on identifying which parts of your work to delegate to Claude Code. The bottleneck isn't the tool — it's recognizing automation opportunities in your daily workflow. Also notes that consolidating communication to email/Slack (away from LINE/Messenger) dramatically improves AI-assisted workflow efficiency.
---
@wasserstein_rao
https://x.com/wasserstein_rao/status/2034065184285737454
A researcher is using Claude Code to directly control a liquid handling robot in a lab setting. No middleware, no custom integration layer — Claude Code sends commands to the robot hardware. This represents one of the most novel physical-world applications of coding agents: AI-driven laboratory automation.
---
@tonkotsuboy_com
https://x.com/tonkotsuboy_com/status/2034177733375430667
Highlights Anthropic's official /feature-dev skill as a hidden productivity gem. Type /feature-dev followed by a vague feature idea, and Claude Code analyzes your codebase, asks clarifying questions, then implements once requirements are clear. The user calls it a brain-activation tool: when motivation is low, the iterative questioning process forces you to articulate what you actually want, kickstarting both clarity and momentum.
---
@JJEnglert
https://x.com/JJEnglert/status/2034329261960475086
A detailed restructuring guide for Claude Code skills. Instead of one monolithic file, break skills into a folder system: SKILL.md as orchestrator (no rules, just workflow), instructions/ for rules (one per concern), examples/good/ and examples/bad/ for reference, eval/checklist.md for automated pass/fail tests, and eval/advisory-board.md for AI reviewer personas that evaluate drafts in parallel. Includes a ready-to-paste prompt for Claude Code that restructures existing skills in 5 phases.
---
@kazzkiq
https://x.com/kazzkiq/status/2034301953534988466
Shares a prompt that turns Claude Code into a professional code reviewer for your entire codebase. It reviews every file, page, endpoint, and logic path, then outputs a QUESTIONS.md with all architectural, refactoring, and security concerns. You answer the questions in the same file, then prompt Claude again to implement improvements based on your answers. A structured human-in-the-loop code review workflow.
---
@Padday
https://x.com/Padday/status/2034387631149039774
Intercom's VP confirms that how they build software is unrecognizable compared to 12 months ago. The team is fully Claude Code pilled and seeing enormous productivity gains. Enterprise-level validation that Claude Code is transforming professional software engineering workflows at scale.
---
@iwhale
https://x.com/iwhale/status/2034198865117839807
A Thai power user built an OpenClaw-powered affiliate marketing system for Shopee. The agent (using Kimi model as its brain) autonomously finds trending posts on X, analyzes what content to write, generates posts in a preset tone, and manages memory for context continuity. Runs 24/7 on a cheap VPS. Cost: X Premium Basic + AI/VPS fees. Revenue: passive affiliate commissions from Shopee's 7-day cookie window. Key insight: use a main agent (expensive model) to direct sub-agents (cheap models) to minimize costs.
---
@tvytlx
https://x.com/tvytlx/status/2034129973615989040
Identifies a likely profitable business opportunity: customizing OpenClaw for parents to tutor their children. Key features: parents can monitor conversations through separate channels, control the teaching style via soul.md, and design skills for generating quizzes, scoring, and reward systems. Based on a viral post about OpenClaw tutoring that got over a million views.
---
@heygurisingh
https://x.com/heygurisingh/status/2034180608210112824
Highlights stop-slop, an open-source SKILL.md file (809 GitHub stars) that makes AI writing undetectable. It bans 50+ AI filler phrases, kills structural clichés, forces sentence rhythm variation, and scores drafts on a 50-point scale. Works in Claude Code, Cursor, or any system prompt. No SaaS subscription — just one markdown file that outperforms paid humanizer tools.
---
@santtiagom_
https://x.com/santtiagom_/status/2034295951586689484
A cost-saving workflow for Claude Code: use /model opusplan to get a hybrid approach where Opus handles planning and architecture while Sonnet executes the code. You get Opus-quality reasoning where it matters most without burning expensive tokens on code generation. Combined with /plan to separate the thinking and building phases.
---
@masp_007
https://x.com/masp_007/status/2034068866876510607
A user built a high-spec outbound calling tool with Claude Code — through pure daily practice without tutorials or courses — that can obsolete most commercial calling tools on the market. Already deployed internally. A testament to what consistent daily AI coding practice can produce.
---
Eco Products Radar:
OpenClaw (267 mentions) — Open-source autonomous AI agent framework, most-starred on GitHub
Claude Code (257 mentions) — Anthropic's terminal-based AI coding agent
Cursor (28 mentions) — AI-native code editor
Obsidian (12 mentions) — Local-first markdown knowledge base
Vercel (11 mentions) — Frontend deployment platform with AI agent skills plugin
Codex (9 mentions) — OpenAI's coding agent CLI
Gemini (8 mentions) — Google's AI model family
Dispatch (7 mentions) — Claude's new phone-to-desktop remote control feature
Ollama (6 mentions) — Local model runner
MiniMax M2.7 (5 mentions) — Self-improving Chinese AI model
Stitch (5 mentions) — Google's AI-native design tool
n8n (4 mentions) — Open-source workflow automation
Superpowers (3 mentions) — Agentic skills framework by @obra