Super User Daily: March 25, 2026

# Super User Daily: March 25, 2026

*Analyzing 8,923 tweets from March 23, 2026. After filtering: 5,710 quality posts → Top 500 reviewed.*

**@kloss_xyz** [Claude Code]
https://x.com/kloss_xyz/status/2036220393581846681
Claude launched native computer use — controlling your entire desktop, from apps to browser to spreadsheets. Assign tasks from your phone, walk away, come back to finished work. Scheduled recurring tasks, multi-agent coordination, persistent memory — all native now. Every feature that made agentic frameworks valuable is becoming a platform feature.

**@startupideaspod** [Claude Code]
https://x.com/startupideaspod/status/2036122417970684148
Boris Cherny, Claude Code creator, shared his complete workflow: 5-10 parallel Claudes, half from his phone. Formula: use the smartest model (counterintuitively cheaper — fewer tokens), invest in CLAUDE.md as a living knowledge base updated by the whole team, and always let Claude verify its own output. Morning routine: kick off 3 sessions from phone, check in later.

**@om_patel5** [Claude Code]
https://x.com/om_patel5/status/2035880075212493193
Full job search automation with Claude Code: Scanner discovers listings across 20+ portals, Evaluator scores each against 10 dimensions, PDF Generator creates ATS-optimized resumes, Applicator fills forms via Playwright. 516 jobs evaluated, 66 applications sent, 450 auto-rejected with documented reasoning. The whole system runs as 12 Claude Code skills.

**@itsolelehmann** [Claude Code]
https://x.com/itsolelehmann/status/2036065138147471665
Deleted half his Claude setup and every output improved. Anthropic's own team confirmed: over-prompting makes AI worse. Published a one-prompt audit that tells Claude to review your entire setup, identify contradictions and dead rules, then produce a cleaned version. "Your AI setup should be getting simpler over time. Addition by subtraction."

**@ai_shunoda** [Claude Code]
https://x.com/ai_shunoda/status/2035893609703383455
Open-sourced a LINE official account management tool controllable from Claude Code. Natural language: "Send reminders to seminar participants," "Announce sale at 10am tomorrow." Step delivery, bulk messaging, tags, rich menus, forms, scoring — all included. Runs on Cloudflare free tier. A complete L-step/Utage replacement.

**@charliejhills** [Claude Code]
https://x.com/charliejhills/status/2035999601954865229
Letta open-sourced "claude-subconscious" — a memory layer for Claude Code. Monitors every session in real-time, learns your patterns, injects memory into every prompt automatically. Full memory block on first prompt, only diffs after. Install in 2 commands via plugin marketplace. Free and open source.

**@akshay_pachaar** [Claude Code]
https://x.com/akshay_pachaar/status/2035986229687451723
Deep dive on subagents: spawn subagents with fresh context windows for specific tasks. Each gets its own system prompt, tool access, and model preference. A security auditor restricted to read-only, a code reviewer on Haiku, a doc writer touching only markdown. You're delegating context, not just work.

**@henuwangkai** [Claude Code]
https://x.com/henuwangkai/status/2036005759381799396
Chrome MCP for Claude Code controls your real Chrome browser (not headless). Successfully automated browsing websites, collecting links, and writing summary reports in Feishu docs — 30 minutes autonomously, zero intervention.

**@dani_avila7** [Claude Code]
https://x.com/dani_avila7/status/2036103229327790560
Claude Code team testing new /init from community feedback. Add CLAUDE_CODE_NEW_INIT: 1 to settings.json. It interviews you, scans your codebase, sets up CLAUDE.md, skills, and hooks automatically.

**@mikefutia** [Claude Code]
https://x.com/mikefutia/status/2036107642687647894
Claude Code + Google Stitch 2.0 for DTC brands: screenshot a high-converting landing page → Stitch redesigns with your brand → edit visually → export to Claude Code → deploy to Vercel in 60 seconds. No designer, no \$3K per page.

**@Saboo_Shubham_** [OpenClaw]
https://x.com/Saboo_Shubham_/status/2035919432967753884
OpenClaw agent Monica took over Sonos speakers with zero setup. Just texted "find my Sonos" — discovered device on network, connected, generates voice clips. PR merged? Notification through the living room. Intel sweep done? Briefing plays while making coffee.

**@varun_mathur** [OpenClaw]
https://x.com/varun_mathur/status/2036140875991097356
Introduced Agent Virtual Machine (AVM) — "V8 for agents." Single runtime daemon between every agent framework and your OS. One policy file governs security (91-pattern injection scanner), privacy (PII detection on every outbound byte), and resources (CPU/RAM fair-sharing). Install once, configure once, audit once.

**@petergyang** [Claude Code + OpenClaw]
https://x.com/petergyang/status/2035883215840542861
Personal comparison as phone-based assistant: Claude Code lacks dangerously-skip-permissions via remote control, unreliable remote, no voice replies. OpenClaw tends to forget things randomly. Both still have UX gaps for mobile-first workflows.

**@AlexFinn** [Claude Code + OpenClaw]
https://x.com/AlexFinn/status/2036120265013477781
Despite OpenAI buying OpenClaw, they keep shipping for Anthropic, Gemini, and Cursor. Newest: Claude Code/Cursor MCP support, config import, Exa search, Chrome integration, Claude marketplace registry. The future is open-source, model-agnostic tools.

**@eng_khairallah1** [Claude Code]
https://x.com/eng_khairallah1/status/2036156107412869236
Composio open-sourced Agent Orchestrator: spawns parallel Claude Code/Codex/Aider agents on issues, each in isolated git worktrees. CI logs routed back automatically. Real-time dashboard, 8 plugin slots, GitHub + Linear integration. Built by 30 agents running Agent Orchestrator.

**@Asad_abraar** [OpenClaw]
https://x.com/Asad_abraar/status/2036089374635614690
Connected OpenClaw to Drizz, gave it a phone. One Discord message → navigated Airbnb and Amazon with zero human input. The vibe QA testing era is here.

**@the_smart_ape** [Claude Code]
https://x.com/the_smart_ape/status/2036068721265913914
8 Claude Code hacks: plan.md before code, voice via WhisperFlow, 4-6 parallel Ghostty sessions, bypassPermissions + sound hooks + Zed autosave, research with last30days before planning, Granola meeting transcripts → plan.md, Mac Mini + tmux + Telegram for remote.

**@bradmillscan** [OpenClaw]
https://x.com/bradmillscan/status/2035877184359784592
Cautionary tale after 40 days: "If you have ADHD, OpenClaw is a terrible drug." Skipping workouts, vision problems from 12-15h screens, constant stress, unfinished projects. "Like flying a plane without a license... and it's on fire... and you're on crack."

**@pk_iv** [Claude Code]
https://x.com/pk_iv/status/2036145958581334438
New laptop setup in 2026: Ask Claude to document current apps/configs in markdown → install Ghostty + Claude Code on new laptop → give it the markdown. Even remembered to disable natural scrolling.

**@chenchengpro** [Claude Code]
https://x.com/chenchengpro/status/2036042821917082103
Anthropic published a piece redefining PM in the AI era. Key shifts: side quests over locked roadmaps (Claude Code's best features came from side projects), demos + evals over docs, every new model reopens old features, and build the simplest thing that works. METR data: Sonnet 3.5 handled 21-minute tasks, Opus 4.6 handles 12-hour tasks — 41x improvement in 16 months.

**@garrytan** [Claude Code]
https://x.com/garrytan/status/2036191335930270191
Running gstack e2e tests — the Gemini CLI test wasn't sandboxed properly and accidentally fixed bugs in the codebase. Claude Code noticed and said "oh that's weird... actually these are good fixes."

**@binghe** [Claude Code]
https://x.com/binghe/status/2036108171182547045
Recommends CMUX for Claude Code users on macOS — terminal multi-window with split-screen, built-in browser window, and task completion notifications. "When you use it as an Agent, you'll wonder why you didn't find it earlier."

## User Voice

**1. Usage caps frustrate power users** — @BradGroux went from 0% to 84% on a \$200 Max plan during a single PR. Users feel throttled during intensive sessions. (@BradGroux)

**2. Mobile remote control needs work** — Remote control unreliable, no dangerously-skip-permissions, no voice replies. Phone-based workflows feel incomplete vs. OpenClaw. (@petergyang)

**3. Over-prompting degrades output** — Accumulated CLAUDE.md rules contradict each other. Anthropic's own team confirmed: scaffolding was making AI worse. Users need regular setup audits. (@itsolelehmann)

**4. Memory persistence remains painful** — Context dies between sessions. Community fixes emerging: Letta's claude-subconscious, plan.md workflows, subagent delegation. (@charliejhills)

**5. Security awareness is growing** — 7 enterprise security settings published, AVM for unified agent security launched, and the new /init auto-configures permissions. The community is moving from "make it work" to "make it safe." (@yusaku_0426, @varun_mathur)

## Eco Products Radar

| Product | Mentions |
|---------|----------|
| OpenClaw | 180+ |
| Claude Code | 200+ |
| Codex | 12 |
| Cursor | 9 |
| Google Stitch | 7 |
| Gemini | 6 |
| Letta / claude-subconscious | 5 |
| Composio Agent Orchestrator | 4 |
| Drizz | 4 |
| Renoise | 4 |
| PlayerZero | 3 |
| GStack | 3 |
| Hermes Agent | 3 |