Super User Daily: April 07, 2026
The April 5 feed told a clear story: agents are no longer weekend toys. They run job hunts, law firms, trading desks, and manga studios. The most interesting shift is that non-coders are building production systems that would have required a small engineering team six months ago. Meanwhile, the community is splitting hard on the OpenClaw ban, and token-limit anxiety has become the background hum of every power user's day.
@Hesamation [Claude Code]
https://x.com/Hesamation/status/2040820834428658008
Built a fully open-source AI job search system that scanned 740+ listings, auto-generated over 100 personalized CVs, and filled out application forms end to end. The result? Hired as Head of Applied AI. The system runs a Go terminal dashboard with 14 skill modes and has 45+ companies pre-configured. This is what "eat your own dogfood" looks like when the dogfood lands you a C-suite job.
@marketcallsHQ [Claude Code]
https://x.com/marketcallsHQ/status/2040727182453231671
Created a self-improving AI backtesting workflow chaining Claude Code, OpenAlgo, DuckDB, and VectorBT together. The system runs an AutoResearch-style loop that iterates on trading strategies automatically. Each cycle feeds performance data back in and lets the model refine entry/exit logic. Quant shops charging six figures for this kind of pipeline.
@alliekmiller [Claude Code]
https://x.com/alliekmiller/status/2040884878229565816
Built "Claudeopedia," a personal knowledge base combining Karpathy's llm-wiki concept with a /last30days skill, a /wiki skill with screenshot capture, and interactive visualization with date ranges. The kicker: a cron job runs "question your assumptions" against her recent writing. Automated intellectual honesty is a feature nobody asked for but everyone needs.
@haraideka [Claude Code]
https://x.com/haraideka/status/2040715282873028769
Using Claude Code to produce doujinshi manga scenarios at industrial scale. Built a multi-department AI system: planning AI handles story design, production AI manages panel layouts, secretary AI does quality checks. Outputs 120-page scenarios with automated CUB rule checking (Confused/Unbelievable/Bored) on every single panel. The creative industry should be paying attention.
@keitowebai [Claude Code]
https://x.com/keitowebai/status/2040736758900686957
Created a skill that transforms raw video footage into complete short-form videos automatically. Cuts, captions, sound effects, BGM, titles, image B-rolls, video B-rolls — all handled without human intervention. The only manual step is pressing "go." Video editors used to charge $500 per reel for this exact workflow.
@mikefutia [Claude Code]
https://x.com/mikefutia/status/2040829879609094552
Built a $10K/month creative strategist replacement inside Claude Code. Apify scrapes the competitor Meta Ad Library, Gemini analyzes each video across 7 dimensions, then Claude generates 10 ad concepts matched to a specific brand voice. Takes 15 minutes instead of a full day. The ROI math on this one is almost embarrassing.
@BradGroux [OpenClaw]
https://x.com/BradGroux/status/2040648338589069598
Adapted Karpathy's knowledge base pattern into a production OpenClaw system he calls corpus-first architecture. Raw data flows through a Compiled loop, agents follow strict query discipline, and Tuesday/Friday memory health checks keep context from rotting. Dual-write policy ensures nothing gets lost between sessions. This is infrastructure, not a demo.
@helloparalegal [Claude Code]
https://x.com/helloparalegal/status/2040802698833154323
Solo lawyer automated three critical workflows. AI intake system sends personalized responses in 6 minutes instead of 6 hours. Collection system pushed recovery from 85% to 94%, which translates to $27K/year in extra revenue. Marketing automation now does past-client check-ins on autopilot. One person running what used to require a small office staff.
@runes_leo [Claude Code]
https://x.com/runes_leo/status/2040644598247420016
Built a fully automated video pipeline: IndexTTS2 for voice cloning, Whisper for subtitle alignment, Remotion for animation rendering. The human only reviews the final output. The entire production chain from script to rendered video runs unattended. This is the kind of pipeline that makes a one-person media company viable.
@pirrer [Claude Code]
https://x.com/pirrer/status/2040776283337232683
Published an extensive analysis comparing the AI era to the "Death Game" arc in Jujutsu Kaisen. The core argument: just like forced awakening of cursed abilities, AI is triggering a structural shift where everyone suddenly has access to powers they never trained for. But it is a non-zero-sum game — the battlefield keeps expanding. Surprisingly rigorous for an anime analogy.
@garrytan [OpenClaw]
https://x.com/garrytan/status/2040797478434549792
His Karpathy-style git wiki knowledge base for OpenClaw hit 2.3GB. Solution: built GBrain with a SQLite backend as an upgrade. GStack for OpenClaw is the full production deployment. When your knowledge base outgrows flat files, you have graduated from hobby to infrastructure.
@cathrynlavery [OpenClaw]
https://x.com/cathrynlavery/status/2040620133828239867
Added session-monitor to catch loops and failures in real time, daily-digest for incident and cost summaries, and auto-port detection to her openclaw-ops setup. This is DevOps for agents. The gap between "I use AI" and "I run AI in production" is exactly this kind of observability tooling.
@AdolfoUsier [Claude Code]
https://x.com/AdolfoUsier/status/2040682441535283710
Burned through 69M tokens running Qwen 3.6 Plus to completely replace Opus-level tasks. Only 2 mistakes across the entire run. Total cost: $50/month subscription. If the quality gap is really this small, the pricing pressure on frontier models is about to get brutal.
@dontbesilent [Claude Code]
https://x.com/dontbesilent/status/2040790528615461178
Built a Xiaohongshu (Little Red Book) title formula tool containing 75 validated viral formulas categorized by 12 psychological triggers. If you have ever scrolled Xiaohongshu and wondered why every title feels engineered, now you know. Someone reverse-engineered the entire playbook.
@pentaclay [Claude Code]
https://x.com/pentaclay/status/2040891955199246518
Spent 45 minutes with Claude Code to design a website loaded with insane micro-interactions. No custom JavaScript, no Figma, just detailed natural language prompts. The output quality suggests that the "design to deploy" pipeline is collapsing into a single conversation.
@aniketapanjwani [Claude Code]
https://x.com/aniketapanjwani/status/2040936714798006683
Using ChatGPT Pro via Oracle for economics research — structural estimation, theorem proving, paper analysis. Full academic workflow integration where the AI handles the heavy mathematical lifting while the researcher focuses on interpretation and novelty. The ivory tower is getting a new tenant.
@vista8 [Claude Code]
https://x.com/vista8/status/2040719882019549523
Recovering from surgery, wrote a script to convert every Paul Graham essay into knowledge cards for an e-book. Downtime turned productive. There is something poetic about using AI to distill one of tech's great essayists while healing.
@jconsu [OpenClaw]
https://x.com/jconsu/status/2040762376656421367
Running a team of agents managed through OpenClaw and Hermes, where agents update their own skills as they learn from errors. Watching them reason about tasks together and self-correct is the moment multi-agent coordination stops being a slide deck concept and starts being a Wednesday afternoon.
🗣 User Voice
User Voice
Token anxiety is real. Multiple users report hitting rate limits within 30 to 60 minutes of focused work — sessions that used to last days now get cut short. The frustration is palpable, and the workarounds (juggling Claude, GPT, Gemma, Qwen in rotation) feel like duct tape on a leaking dam.
Context management is the silent tax on every power user. One user audited 900 sessions and found 22x token repetition per session. That is 95% of your context budget going to the model re-reading things it already knows. Until this gets solved at the architecture level, every workaround is just damage control.
The OpenClaw ban created a genuine community schism. Some users see Anthropic protecting its platform; others feel betrayed by a company they evangelized. The middle ground — that the ban was probably necessary but poorly communicated — is where most thoughtful takes land.
Non-coding users keep surprising everyone. Lawyers, traders, manga artists, marketing strategists — people with zero engineering background are building production systems. The "learn to code" era might be giving way to the "learn to prompt" era faster than anyone expected.
Token anxiety is real. Multiple users report hitting rate limits within 30 to 60 minutes of focused work — sessions that used to last days now get cut short. The frustration is palpable, and the workarounds (juggling Claude, GPT, Gemma, Qwen in rotation) feel like duct tape on a leaking dam.
Context management is the silent tax on every power user. One user audited 900 sessions and found 22x token repetition per session. That is 95% of your context budget going to the model re-reading things it already knows. Until this gets solved at the architecture level, every workaround is just damage control.
The OpenClaw ban created a genuine community schism. Some users see Anthropic protecting its platform; others feel betrayed by a company they evangelized. The middle ground — that the ban was probably necessary but poorly communicated — is where most thoughtful takes land.
Non-coding users keep surprising everyone. Lawyers, traders, manga artists, marketing strategists — people with zero engineering background are building production systems. The "learn to code" era might be giving way to the "learn to prompt" era faster than anyone expected.
📡 Eco Products Radar
Eco Products Radar
OpenClaw remains the center of gravity for the power user ecosystem, but the ban drama is pushing some users toward alternatives like Kilo Claw. Hermes Agent is becoming the default orchestration layer for multi-agent setups. Goose from Block keeps showing up as a lightweight alternative for simpler workflows.
On the model side, Qwen 3.6 Plus and GLM 5.1 are getting serious traction from users doing cost optimization. At $50/month for 69M tokens with near-Opus quality, the value proposition is hard to ignore. Gemma 4 is the dark horse — mentioned frequently but still in the "experimenting" phase for most users.
DESIGN.md and Awesome Design MD are becoming standard practice for project documentation that agents can actually consume. GStack is what happens when your knowledge base outgrows files and needs a real database. Obsidian continues to be the preferred human-side knowledge layer that feeds into agent workflows. MiniMax is picking up mentions as a specialized tool but has not broken into the mainstream yet. Codex appeared in several conversations as OpenAI's answer to Claude Code, though early impressions are mixed.
OpenClaw remains the center of gravity for the power user ecosystem, but the ban drama is pushing some users toward alternatives like Kilo Claw. Hermes Agent is becoming the default orchestration layer for multi-agent setups. Goose from Block keeps showing up as a lightweight alternative for simpler workflows.
On the model side, Qwen 3.6 Plus and GLM 5.1 are getting serious traction from users doing cost optimization. At $50/month for 69M tokens with near-Opus quality, the value proposition is hard to ignore. Gemma 4 is the dark horse — mentioned frequently but still in the "experimenting" phase for most users.
DESIGN.md and Awesome Design MD are becoming standard practice for project documentation that agents can actually consume. GStack is what happens when your knowledge base outgrows files and needs a real database. Obsidian continues to be the preferred human-side knowledge layer that feeds into agent workflows. MiniMax is picking up mentions as a specialized tool but has not broken into the mainstream yet. Codex appeared in several conversations as OpenAI's answer to Claude Code, though early impressions are mixed.
Comments