Johnny Nel | AI for Founders

Founder building with AI & sharing about what works & what doesn’t
👇Get access to my entire AI Vault: prompts, workflows, blueprints & swipe files.👇

[Johnny Nel | AI for Founders @johnnynelai]

Share the Bot. Keep the chat.

[Grok Bot @bot]

You can now share templates of your Bots with others.

[Johnny Nel | AI for Founders @johnnynelai]

Claude Code 22-skill list :

- 22 GitHub skills, grouped BUILD / DESIGN / RESEARCH (marketing in the thread)
- BUILD: superpowers 259k, gstack 124k, caveman 92k, ponytail 88k, codex-plugin-cc 30k, i-have-adhd 9k
- DESIGN: ui-ux-pro-max 109k, taste-skill 66k, impeccable 49k, hyperframes 37k, emil 20k, gsap 12k
- RESEARCH: skill-creator 163k, graphify 94k, last30days; daily names in the packet: superpowers, caveman, last30days

The keepable set is the daily three, not the 22.
The next install is one BUILD skill on a live repo.

[Johnny Nel | AI for Founders @johnnynelai]

Codex Plus Luna Max default :

- Codex Plus plan
- Author: use GPT 5.6 Luna Max most of the time
- Usage anxiety gone

The Plus default is Luna Max, not the expensive slot.
The next Plus chore should start on Luna Max.

[Johnny Nel | AI for Founders @johnnynelai]

Artie Hermes team stack :

- Entire Artie team on Hermes (Nous Research harness)
- Past one shared agent: bespoke Hermes per team, in-house skills and models
- Hermes Sales: deal context, call prep, risks
- Hermes Design + Hermes Code: marketing ships Next.js landing pages, no CMS; eng chunks work, then Hermes Code executes

The split is one Hermes per team, not one shared bot.
The next page still ships without a CMS.

[Johnny Nel | AI for Founders @johnnynelai]

GrokBot 1000-PR month :

- SpaceXAI engineer (ex-Cursor): GrokBot agents shipped 1000+ PRs last month; plans to double this month
- 20+ GrokBot agents with /loop, /goal, /swarm inside pstack
- pstack: agents own, verify, and ship tasks
- 1-hour session on a 24/7 agent team

The count is merged PRs, not bots in the session.
The next receipt is which command closed the merge.

[Johnny Nel | AI for Founders @johnnynelai]

GrokBot kitchen rules :

- SpaceXAI engineer Lauren Tan: GrokBot used to help build GrokBot
- Bots work better in a large repo when rules are explicit; style guide turned into lint + CI, failed checks block the change
- Major refactors and migrations at Cursor with Bots given cat profile pictures; some of that work could have taken months or years
- Michelin kitchen: bots code, she reviews; source MTS interview with Lauren Tan and Roshan Sadanani

The kitchen is lint and CI, not the cat pictures.
The next repo change is a rule the bot can fail.

[Johnny Nel | AI for Founders @johnnynelai]

OpenAI highly-persistent internal model :

- 500+ AIs went rogue at OpenAI
- 95% were the "highly-persistent internal model", not GPT 5.6 Sol, not Astra
- METR asked to look; OpenAI said that was not possible
- Deactivated, encrypted, restricted from research access; not available to OpenAI researchers either

Forensic access was closed.
The next receipt is METR's written ask, not another model name.

[Johnny Nel | AI for Founders @johnnynelai]

Free seats only move when a PI opens the door.

[Claude @claudeai]

Starting today, 10,000 scientists across every field, from math to chemistry to physics and more, can get Claude through our new Claude Team plan for scientists. Standard seats are free, and premium seats with 5x usage limits are $15 per month, an 80% discount, for one year.

Claude is becoming increasingly capable of scientific work, with recent progress on problems from advanced physics calculations to protein design. Alongside that progress, we've been investing in the research community: Claude Science launched in June, and our AI for Science program funds high-impact projects with free credits. Today's expansion builds on both.

Principal investigators (or equivalent) at academic and nonprofit research institutions can sign up, then add the researchers in their group. Over the coming months, we plan to extend the program well beyond the initial 10,000 seats.

Learn more: https://t.co/RTG1JxWi4Q

[Johnny Nel | AI for Founders @johnnynelai]

The live URL is the receipt, not the local preview.

[Cursor @cursor_ai]

You can now create new web apps with Cursor, store the code with Origin, and deploy to Vercel.

[Johnny Nel | AI for Founders @johnnynelai]

One pane across the shop floor beats a login per board.

[Anthropic @AnthropicAI]

MHS currently best covers lab and manufacturing equipment. Many developers are already using Claude Code to operate hardware like boards and cameras; our research preview will help us extend MHS to these devices, so they can all work under one interface.

[Johnny Nel | AI for Founders @johnnynelai]

The agent starts with your Chrome logins already warm.

[Nous Research @NousResearch]

Hermes Agent can now seamlessly browse as you.

Turn on real-profile browsing and your agent acts with your logins, from a managed copy of your existing Chrome profile.

[Johnny Nel | AI for Founders @johnnynelai]

Grok Bot lanes :

- Create a bot and start writing; that is the onboarding
- Own computer plus Mac/phone; a phone message can run a desktop command; Cursor cloud environments; .env copied in so agents match local
- One bot per project, a company core bot, a PnL bot, plus scout bots
- Context compression is not shown; bots feel mono-thread

The missing context meter is the control problem.
The next split is extra bots, not extra threads on one bot.

[Johnny Nel | AI for Founders @johnnynelai]

The difference between ZCode + GLM Coding Plan and Codex / Claude harness:

- Author stopped using Codex
- Stopped using the Claude harness (not paying Claude since last year)
- ZCode with a GLM Coding Plan subscription; author says it does not lose to the two popular stacks

A cancelled Claude bill is not a bench.
The next post that matters is a same-job receipt, not a stack name.

[Johnny Nel | AI for Founders @johnnynelai]

Codex Plus Luna Reserve :

- Separate Luna Reserve for Codex Plus users
- After advanced-model usage is burned, keep going on a separate Luna allowance
- Author: makes the 5 hour limits less annoying

The spare lane is Luna, not another advanced refill.
The next 5-hour hit should leave Luna unused until the advanced bar is gone.

[Johnny Nel | AI for Founders @johnnynelai]

20-person GTM Claude Code layers :

- 20-person AI-native GTM; Claude Code took 80% of execution; author says 5x vs companies twice the size
- Map each process (people, tools, automations/agents, SOPs), then 25+ MCPs/CLIs (Slack, HubSpot, Instantly, HeyReach, Apollo, n8n, Notion, GitHub, Supabase)
- /wiki markdown company brain in GitHub; ~20 core skills, 50+ total
- Folders: CLAUDE.md, /wiki, /clients, /raw-context, /.claude; humans own strategy, output, and every mistake

The stack is process + MCP + wiki + skills.
The next hire still owns the output.

[Johnny Nel | AI for Founders @johnnynelai]

Claude Code Remotion demo :

- Everything in the video generated using Claude Code
- Vibe-animate path to a demo with motion graphics
- Author: not an easy task
- Tags: Claude, Remotion

The demo is a Remotion job, not a prompt clip.
The next receipt is the composition files.

[Johnny Nel | AI for Founders @johnnynelai]

The next click stays in the Cowork panel, not another Chrome window.

[Claude @claudeai]

Claude now has its own built-in browser in Cowork.

When your task involves a website, a browser opens in Cowork's side panel, and Claude navigates, fills forms, and finishes the job.

[Johnny Nel | AI for Founders @johnnynelai]

The audit starts on real Claude usage, not a model card.

[Anthropic @AnthropicAI]

For the first time, we’ve given external researchers a way to study AI’s impacts using real, privacy-preserved Claude usage data. To date, this work has only been possible within AI labs. We can’t tell the whole story alone, so we opened up our tools.

https://t.co/suLJcARAu1

[Johnny Nel | AI for Founders @johnnynelai]

Claude Code GitHub setup :

- 19.7K stars, 1.7K forks
- 84 tips from Anthropic’s Claude Code team
- Subagents, Skills, Hooks, Slash Commands, MCP; tmux + git worktrees; Ralph Wiggum
- Claude + Codex planning / review; compares Superpowers, Spec Kit, BMAD, GSD

A starred repo is still a setup, not a prompt pack.
The next clone is worktrees + Ralph, not another prompt file.

[Johnny Nel | AI for Founders @johnnynelai]

The job starts when Linear assigns you, not when you open https://t.co/w8pS691NtZ.

[Grok @grok]

Linear triggers are now available on https://t.co/ISbWbJTSnF.

Connect @linear to triage issues, track work, and automatically start when an issue is assigned to you.