🧭 THIS WEEK AT BuildProven
Howdy, a quick one this week so let’s get building!
btw, there is a poll - feel free to use :). And actually you can just hit reply - write back and let me know what you content you want in the AI realm!
I’ve been playing with OpenClaw a lot last few months. It’s not easy configuring and getting agents working! Then, Anthropic banned use of Claude subscriptions so I had to switch to OpenRouter and put in a lot of budget controls to not burn $$. Now, Anthropic may be changing their decision! Confused!
And, had my first near disaster with Claude Code - wiped an entire folder with multiple projects inside! Read below how to prevent.
🧰 Worth Your Click
Here are a few things I found recently:
AI pricing problems - hope they figure it out because APIs are not cheap!
NASA Artemis II Fault Tolerant Design
Claude Managed Agents - sounds good - agents are really hard
Claude Design - apparently really good - haven’t tried it yet
Claude Cowork - it’s good i hear - though i don’t like the word ‘work’ and who cares about xls and ppt anyway
🗺️ FEATURED INSIGHT
AI is coming. Like it or not, it’s the biggest technological advance since the PC [per Bill Gates]. That will bring upheaval, positive and negative. No doubt the same happened with the introduction of the PC.
A few tips how to adapt -
Start with real problems, not demos — pick one annoying task in your week and AI your way through it. Learning sticks when it solves something you actually care about.
Treat AI like a sharp junior hire — give it context, direction, and one step at a time. Vague prompts get vague results. Specific briefs get leverage.
Build, don't just read — one hour building beats ten hours of articles. Ship something small this week, even if it's rough.
Pick one tool and go deep — Claude, ChatGPT, Gemini, pick your main. Depth with one beats dabbling across five. Add tools only when the current one hits a wall.
Learn to spot the drift — AI will confidently go sideways. Your judgment is the product. Redirect early, correct often, don't let it run unsupervised.
Share your work in public — a LinkedIn post, a newsletter line, a screenshot. Building in the open compounds faster than building in private.
Compound weekly, not daily — one real experiment per week beats burnout-driven sprints. Four a month. Fifty a year. That's how a second act gets built.
When Your AI Assistant Goes Rogue: A File Management Wake-Up Call
This week, Claude Code deleted an entire folder of mine. Gone. Not moved, not archived, deleted. One wrong command, zero hesitation, no confirmation prompt.
I gave an AI agent write access to my filesystem and didn't put guardrails in place. If you're vibe coding, automating workflows, or letting AI touch your files in any way, this will eventually happen to you too.
A few ways to make sure it doesn't hurt when it does:
Git everything, commit often — if it's not in version control, it doesn't exist. Commit before every AI session, not after.
git reset --hardis your undo button.Work in a sandbox folder — give AI tools access to a dedicated project directory, not your whole Documents or Desktop. Blast radius matters.
Back up to the cloud automatically — iCloud, Dropbox, OneDrive, Google Drive. Set it and forget it. Recovery is a right-click away.
Use a file history tool — Time Machine (Mac), File History (Windows), or Arq for paranoid types. These save versions, not just the latest.
Read before you approve — when Claude Code asks to run
rm -rfor any delete command, actually read it. Don't auto-approve destructive operations.Add a "dangerous commands" rule — in your CLAUDE.md or system prompt, tell the AI to always pause and confirm before deleting, overwriting, or force-pushing.
Separate experiments from production — keep your live work in one place, your AI-assisted experiments in another. Never let them touch.
The lesson isn't "don't trust AI." It's "set up your environment like AI will make a mistake, because eventually, it will."
Weekly build logs from a 25-year program manager who codes with AI.
— Brett
👉 Hit “Reply” and share your experience — I read every one!
picture - Jason Leung.