Kubicast #192 - Can AI actually help you in your daily life?

“If AI has no context, it doesn't do as well.”
This was a recurring theme in the episode: context and intent matter more than the trendy tool. We talked about prompts (and the risk of accidentally leaking a prompt), how to organize work to reap productivity gains, and the limits that prevent frustration.

Key themes of the episode

1) What really helps in the day-to-day

• Simple and useful cases: drafting text, proofreading, and quick explanations.

• Development support: completing snippets, suggesting tests, pointing out obvious errors.

• Where it gets stuck: asking for “magical” solutions without providing the right context.

2) AI agents for coding

• Experiences with agent tools focused on code, including mentioned solutions from Google.

• Where agents shine: breaking tasks down into steps, following a flow, maintaining state.

• Where they get in the way: promising total automation and generating technical debt.

3) Models: generalists vs. specialized

• Generalists help with drafting and exploration.

• Models with a specific focus tend to make fewer mistakes in the right domain.

• Fine-tuning the prompt and providing context are decisive factors.

4) Security and governance

• Constant point of attention in the conversation: security.

• Risks discussed: sending sensitive data to external services, prompt leakage via image features and logs, and exposure of internal code.

• Minimum practices: usage policies, sanitizing data before sending it to the model, and auditing outputs.

Takeaways

If we want real gains with AI in our daily lives, the path involves less glamour and more process: context, security, and a clear scope. Tools like Copilot and code-focused agents add a lot of value when used within defined limits. The secret lies in experimenting responsibly — and learning quickly from each iteration.

Listen or watch on Spotify as well, and share it with everyone, we need to move fast!