If privacy and speed matter, try Super Whisper.

It changed how I work every day.

Writing emails, drafting ideas, even coding notes.

What would you do with this kind of tool?

Super Whisper supports various accents and languages.

Accuracy rivals cloud-based services.

No internet needed after installation.

Great for travel or low-signal areas.

What tasks would you automate with faster voice input?

Many voice apps send your data to the cloud.

This means potential privacy risks.

Super Whisper works fully offline.

All transcription happens locally.

Your sensitive info stays on your device.

How important is data privacy in your workflow?

Typing takes time.

Voice input turns speech into text instantly.

I started using Super Whisper for notes and emails.

My typing speed doubled.

Tasks that took hours now take minutes.

Are you still typing out long messages?

To sum up:

- Voice AI beats typing on speed
- Local run protects your data
- Immediate responses boost flow
- Privacy lets you focus

What projects would you speed up with voice AI like Super Whisper?

If you rely on typing for notes or drafts, try voice AI with local processing.

It might surprise you how much time you save.

Ask yourself: Does your tool respect your privacy?

That should be non-negotiable.

Privacy + speed = better focus.

I trust the tool to handle sensitive info.

I get more work done with less distraction.

This changes how you interact with AI daily.

Running locally also means faster responses.

No network lag or waiting for servers.

Workflow feels more natural.

Productivity jumps when tools react instantly.

Most voice AI apps send your data to the cloud.

This risks privacy.

Super Whisper stores data on your device.

You keep control.

No third party listening in or storing files without your consent.

Typing slows me down.

With voice AI, I speak thoughts out loud.

Words come faster than fingers do.

My ideas stay fluid and raw.

No stopping to correct typos mid-thought.

Although it's not just about technical skill… you also need to communicate precisely what you want.
Any additional info you can provide to avoid dead-end paths will save tokens and generally improve results.

The bottom line:

Make sure YOU are driving the sub agents, not the other way around!

However, I found the best results from explicitly telling each sub agent what to do. 2025 AI can figure certain things out, but for more complex tasks, especially when you know a better way, you need to guide it.

The better your technical skill, the better the output.

The longer sub agents can go without interrupting me, the better. I could probably add more MCP tools to empower them further.

Overall, I really like sub agents for straightforward, parallelizable tasks that can be divided simply. For example: agent 1 chunks file, agent 2 processes chunks, agent 3 combines everything and writes a comprehensive report.

For long-running tasks with well-defined output that's easy to review, you can run sub agents in the background and let ‘em go.

For better security and guardrails, jail them in a separate directory or Docker container.

4. PLAN MODE System Reminders

Full newsletter here:
https://sabrinaramonov.substack.com/p/reverse-engineering-claude-code-using

3. Claude Code’s Web Fetch Tool Prompt

2. Experimental ULTRACLAUDE.md

1. Claude Code’s TODO Tool Prompt (amazing feat of prompt eng)