Three Things to Decide First When Handing Work to AI Feels Scary: Reversible Operations and Parallel Running

I still often hear people say, "Leaving everything to AI is scary." And many companies have stopped implementing automation altogether for that very reason. Being careful is not bad in itself. But my answer is clear. Accept that now is the time to take risks. In exchange, make sure operations can be rolled back. And run the old way and the new way in parallel to give yourself a trial period. This article lays out these three points as decision criteria.

On October 18 I talked about competing on the number of attempts, because AI makes mistakes. On October 20 I talked about how a company's speed is set not by AI but by the round-trips of human decisions. This piece continues from there: when you actually reach the stage of handing work over, what should you decide first so that you can take the step?

Scared, so you don't move; not moving, so you can't see

The most common thinking is, "Handing this to AI is scary, so let's wait and see a little longer." It looks careful and correct. But with automation, you cannot see what will happen until you actually build it and run it. Where it makes mistakes, and where it is safe to leave it alone, only become clear when you look at the real thing in operation. While you are waiting and seeing, there is nothing to look at in the first place.

Scared, so you don't move. Not moving, so you can't see the real thing. Unable to see it, so you stay scared. No number of meetings and no amount of research gets you out of this loop. The only way out is for someone to decide to take the risk. In a company, that is usually the president or the person responsible for the business. Telling frontline staff "it's fine to fail" does not help if they cannot carry the responsibility.

Criterion 1: Accept that now is the time to take risks

The first is to decide in advance, as a period, rather than facing the fear decision by decision. Look at decisions one by one and you will find a reason to oppose every one of them. But with a shared premise of "this is the time to move forward, so let's try it first," people can move ahead without asking for permission every time.

That is exactly what happened at my company. We decided first that "now is the time for risk-taking," and then pushed automation forward quickly. If we had debated "is it safe to hand this over?" one item at a time, I think we would still be stuck on the first one. As a result, almost all of our internal systems are now connected, and marketing runs with about three minutes of my own hands-on time. That state was not designed from the start; we built it while running the real thing.

Criterion 2: Make operations reversible

Deciding to take a risk is different from jumping in with no preparation. The second criterion is to make operations reversible. Concretely: keep a record of what was changed and when. Decide in advance how to return to the old way if the new way does not work. Decide where results will be checked. With these three in place, the company will not stop even if something fails.

Preparing to roll back may sound like a brake, but I see it as an accelerator. If you know you cannot go back, anyone becomes scared and cautious. If you know you can, you can try boldly. Caution is needed for decisions that cannot be undone. For reversible ones, decide quickly, try, and roll back if it does not work. Changing everything at once in a way that cannot be undone is close to gambling. Bold decisions to move forward combined with reversible operations make for a way of taking risk that you can repeat as many times as needed.

Your first rollback plan can be simple. If you automate a routine report, for example: "If it doesn't work, the person in charge makes it by hand as before, and the original template and procedure stay in place." That alone is a proper rollback plan. A surprisingly common mistake is clearing away the old procedures and templates as soon as you switch, which leaves you unable to go back even when you want to.

Criterion 3: Take a trial period by running in parallel

The third is to run in parallel and take a trial period. When you introduce new automation, do not stop the old way immediately; run both for a while. Process the same work both ways and compare: are the outputs different, how many errors occur, where were human hands needed? Even if the new way has problems, the old way keeps running, so the work does not stop.

Running both temporarily increases the workload. Even so, compared with the effort of recovering when a sudden switch stops the work and causes trouble for customers and colleagues, the cost of running double for a while is usually small, a price paid for safety. Set the period in advance and decide the final check date when you start. Parallel running is not meant to last forever; it is a period for switching over with confidence.

And verify risks generously. Deciding to move forward does not mean checking less. A company that has stopped has nothing real to verify. Because you are running it, you can verify it. Taking risks to move forward and checking risks generously point in the same direction.

Daily upkeep to resident agents; whether to continue or roll back to people

Automating and integrating is not the finish line. That is where daily upkeep begins: fixing small defects, adjusting procedures, reflecting new findings. This is an area resident agents can take on quite well, because each task is small and easy to record, so it can be rolled back if it does not work. What remains for people is the decision of whether to continue the current way, roll back, or widen the scope. With records, what matters is deciding who looks at them and when, for example a set person reviewing the comparison once a week.

While you stand still, other companies move ahead

Thinking that standing still costs nothing as long as things stay as they are is, I believe, a big misunderstanding. While your company stands still, other companies keep trying, failing, fixing and trying again. The difference in the number of attempts becomes a difference in experience, and it is hard to make up later. Not moving carries a clear risk too. And trying something in a reversible way and rolling it back is not a failure; it is a result, because you learned something. Not blaming the person who tried is also an important job for a president who has declared a period of risk-taking.

What you can do this week

Pick one automation that has stalled because it feels scary. Work that recurs daily or weekly and whose results are easy to check by eye is a good fit. Next, write down the steps to roll it back. Then decide the period for running it alongside your current way and the final check date. When you tell your team, I suggest giving what they may do and what they must protect as a set: "We are entering a period of taking risks and moving forward. In exchange, always proceed in a way that can be rolled back. For a while, run it alongside the current way and compare the results."

▶ Watch the full video (Japanese)
https://www.youtube.com/watch?v=kVcRCXfevYY