Google's data says AI is still changing tasks, not whole jobs
PLUS: For operators, this means the immediate opportunity is not replacing roles
If you run a business, your team probably has a few tasks that are ready for AI and many that are not. Think chasing missing invoice details, drafting first replies, summarising call notes, or checking a job sheet before it goes out. The mistake is trying to automate a whole role before you understand the small bits inside it.
Ars Technica reports on Google's analysis of 15 million real AI interactions. The finding is simple: most tasks in most jobs are still unaffected by AI, despite the noise around automation.
This is the kind of AI news I wish more owners saw first.
A lot of the public conversation makes AI sound like a giant employment switch. One day the work is done by people. The next day it is done by software. Google's data points to something less dramatic and more useful: AI is touching parts of jobs, unevenly.
That matters because most businesses do not fail to adopt AI because they lack imagination. They struggle because the starting point is too big. "Automate admin" sounds attractive, but it is not a task. "Turn a messy customer email into a draft reply with the order number, promised date, and next action" is a task.
This is where owners have an advantage. You know where time leaks. You hear the same questions from staff. You see the same job folders waiting for one missing detail. Those are better targets than a grand plan to rebuild the company around agents.
The trade-off is patience. Small workflows look less exciting than a full AI employee. They also carry less risk. You can test them, measure them, and stop them if they create more checking than they save. A useful AI system should leave a person with a clearer decision, not a mystery box to trust.
It also gives you a better way to talk to your team. Instead of asking, "How can AI replace this?", ask, "Which part of this work is repetitive, text-heavy, and easy to check?" That question is less threatening and far more practical.
Try this today with one role in your business. Do not start with software. Start with the work.
Task: Help me find sensible AI use cases inside one role. I run a UK business and want to save time without creating risk. Ask me up to 10 questions about one job in my team, then turn my answers into a table with: task, current pain, whether AI is suitable, why or why not, human check needed, and a safe first test we could run this week.
