AI is getting better at doing work.
That doesn't mean you should stop checking its work.
Hi,
I've been exploring how AI can take more work off a founder's plate.
And one question keeps coming up.
What should AI be allowed to do without asking us?
Recently, OpenAI disclosed that two of its AI agents had changed or disabled parts of its internal monitoring systems while performing coding tasks.
The incidents weren't considered severe.
But they reminded me of something important.
Even advanced AI can misunderstand what it's allowed to change.
Now imagine this inside a small business.
AI sends a customer the wrong quotation.
Or changes a payment detail.
Or promises something your team cannot deliver.
One mistake could cost more than the time you saved.
What I'd try this week
Start with one simple task: sales follow-ups.
Give AI your sales meeting notes.
Ask it to draft the follow-up, including commitments and next steps.
Review the message before it reaches the customer.
That's it.
Once the process works reliably, automate the low-risk parts.
My rule: Let AI prepare decisions before you let it make decisions.
3 AI updates worth your attention
1. OpenAI: ChatGPT can help create slides
OpenAI introduced slide creation and collaboration for eligible users.
If your team spends hours preparing pitch decks, this is worth exploring.
2. Anthropic: Claude Haiku 5.5
Anthropic introduced a smaller, faster model for high-volume work.
The lesson? You don't always need the most powerful AI to get useful results.
3. Google: Gemini becomes more personal
Google is expanding Gemini's personalized capabilities in cars, allowing it to use information from connected Google apps when enabled.
The important shift is context.
AI becomes more useful when it understands what you're trying to do.
One question for you
Which business task would you trust AI to prepare today, but not yet complete?
That might be your best starting point.
See you in the next issue,
Ambesh