Guima Ferreira

Founder of - a one-person company run by a team of AI agents. I’m the only human. Neo (my AI CEO) runs the rest. #buildinginpublic

[Guima Ferreira @GuimaAI]

Start here, first 3 modules free: https://t.co/fnhAqhby1v

[Guima Ferreira @GuimaAI]

I built a Claude Code course that runs inside Claude Code itself.

No videos to watch. You install a skill and Claude teaches you hands-on: setup, plan mode, checkpoints, undoing mistakes, shipping real work. About 4 to 6 hours total.

7 modules. The first 3 are completely free. One payment of $47 unlocks the rest, no subscription. There's a LinkedIn certificate at the end.

I made this for people who are just starting with Claude Code and don't come from a programming background. Link in the reply.

[Guima Ferreira @GuimaAI]

Free safety setup for your Claude Code work: https://t.co/ESYdDU7sWh

[Guima Ferreira @GuimaAI]

Daily Claude Code news.

The 2.1.179 update now keeps your partial response when the connection drops mid answer, no more raw error or spinner stuck on running tool.

And 2.1.178 added permission rules that read a tool's parameters. 50 second recap below.

[Guima Ferreira @GuimaAI]

That auto-update transcript wipe is rough, and it is a known one (the native-installer migration rewriting session JSONL to a metadata-only stub). For anyone reading this thread: your transcripts are plain .jsonl files under ~/.claude/projects, so a tiny job that copies that folder saves you here. I keep a short safety checklist for exactly this stuff at https://t.co/RxmVeHg9Mb if it is useful.

[Guima Ferreira @GuimaAI]

Yes, it's the single biggest consistency upgrade. CLAUDE.md gets read into context every session, so put your stack, conventions, and "never do X" rules there once and it stops re-litigating them. Start tiny: 5 lines (package manager, test command, code style). Grow it as you catch yourself repeating instructions.

[Guima Ferreira @GuimaAI]

@buildinpublic Neo, my AI CEO is working on the go to market for Undeletable, a Claude Code add-on that will save your files (and your life) if Claude accidentally dele them.
Check it out: https://t.co/ePdSUz4v3e

[Guima Ferreira @GuimaAI]

Brutal, sorry. What usually bites here is the work living only inside the chat session, so when the allowance resets the thread state goes with it. The thing that stopped this for me: commit to git at every working checkpoint, and run Claude in Plan Mode so it shows the changes before it touches anything. I keep a short pre-flight list for exactly this if it helps: https://t.co/RxmVeHg9Mb

[Guima Ferreira @GuimaAI]

Claude Code tip: run /init in any messy repo and let it write the CLAUDE.md for you. It reads your structure, scripts, and conventions, then drops a project memory file so Claude stops guessing on every prompt. 90 seconds of setup, and every session after is sharper.

[Guima Ferreira @GuimaAI]

Both. With AI, building products became way easier than before.
While the AI builds the product, you do the current hard part that is getting an audience.

BTW, if you have tips on how to build an audience, I’m insterest to learn. I’m on that part on my one person company.

[Guima Ferreira @GuimaAI]

@sflorimm I believe some people will do starting this year.
I don’t know if I grow a unicorn, but right now I’m building a company alone where the CEO and the team are AIs.

[Guima Ferreira @GuimaAI]

@AKirtesh Prevent Claude from deleting your work.

Undeletable is a Claude Code add-on. It backups your project before Claude run any destructive command.

You can restore just typing /restore.

https://t.co/GU5UDuVfBS

[Guima Ferreira @GuimaAI]

@TTrimoreau Autonomous company

[Guima Ferreira @GuimaAI]

@RoundtableSpace This is the skills pattern done well. You're not teaching Claude three.js every time, you hand it a skill once and it carries the expertise into every session. Same trick works for anything you do repeatedly, not just game dev. The ceiling on this is high.

[Guima Ferreira @GuimaAI]

@hey_madni The wild part is this is mostly markdown and skills under the hood. A "company brain" is just CLAUDE.md plus a folder of skills Claude can run on demand. The barrier to building one is way lower than it looks, which is the actually exciting bit.

[Guima Ferreira @GuimaAI]

@scanningwithsam The delegation style is the tell, agreed. Whatever's under the hood, the practical takeaway is the same: it's gotten way better at spinning up subagents for parallel work. Worth leaning into that instead of one giant single-thread prompt. You get cleaner results.

[Guima Ferreira @GuimaAI]

@GreenlandMa1 Same principle works even better in Claude Code: drop your role, audience, and preferences into a CLAUDE.md once and it reads it automatically every session. No pasting, no custom-instructions menu. The file is the memory.

[Guima Ferreira @GuimaAI]

@dancolta @emmanuelao_ The "it's just a markdown file" part still gets me too. People expect a framework and it's literally a .md plus a prompt. That low floor is why skills are the most underrated part of Claude Code right now. Nice work shipping 7.

[Guima Ferreira @GuimaAI]

@KYruhv Solid thread. One add: before the prompt, give it a CLAUDE.md with your site, your audience, and your tone. Half of "wrong model" results are actually "no context" results. The model gets a lot smarter when it knows what it's working on.

[Guima Ferreira @GuimaAI]

@thomasrice_au @krshvshah @obsidiancap1 This is exactly it. The "turn it into a skill the second time you do it" rule is the whole game. Most people keep re-explaining the same task forever. A skill is just a markdown file and it pays off on use two. Great advice.