Use AI or Not?

Every job is a stack of tasks. AI is only safe on some of them.

That is the idea behind a matrix making the rounds, attached here. It places each task on two questions. How high is the cost of error if it goes wrong, and how much human judgment does it actually need? The answers sort work into four boxes.

1) Low cost of error, low judgment: Automate with AI. Think transcribing calls, formatting files, pulling data from forms. Take call transcription. If the tool mishears a word, you fix it and nothing breaks, so the cost of error is low. Turning speech into text needs accuracy, not taste, so the judgment is low too. This is the one box you can rely on AI to do without causing too many problems.

2) Low cost of error, high judgment. Use AI for first draft. . AI starts, a human refines. Think blog posts, emails, brainstorming. Consider a marketing email. A rough draft costs little because it gets edited before it reaches a customer, so the error cost is low. But the angle and the voice need a human to shape them, so the judgment is high. AI gives you a starting point a person still has to finish.

3) High cost of error, low judgment: Review every time. AI prepares, a human approves. Think financial reports, contract drafts, medical assessments. Look at a quarterly financial report. A wrong figure can mislead investors or trip a regulator, so the cost of error is high. Yet the rules for what goes where are well defined, so the judgment is lower than it looks. AI can assemble it, but a person has to check and sign it.

4) High cost of error, high judgment: Don't use AI. The human decides. Think ethics calls, crisis management, letting someone go. Picture having to fire an employee. Get it wrong and you damage a person's livelihood and expose the company to a lawsuit, so the cost of error is high. It also needs empathy and accountability that AI cannot hold, so the judgment is high. This one stays fully human.

Please note that these boxes sort tasks and process steps, not whole jobs. A single role is spread across all four. Hiring is a good example. Writing the job advert is a first draft AI can start. Screening resumes can be automated, with review for bias. Scheduling interviews and summarizing notes can be automated. The final call, who gets the offer, stays in the box where Humans own the decision making. More than half of hiring can be automated or augmented with AI, yet you can never leave end to end hiring to AI.

In conclusion, only one box, the low-cost, low-judgment one, is truly automatable, and even there you keep a human in the loop and on the loop, checking the output and ready to step in. AI will automate and augment parts of almost any job. But most jobs will continue to require humans to execute it successfully, safely and reliable.