Artificial Intelligence (AI)
Your First AI Pilot Needs One Job
A good first AI pilot has one job: a task you can judge, an owner, and a finish line.
“Use more AI” leaves all three for someone else to figure out.
I want tools that help us do more and do it well. I look at AI the way an operator looks at any new tool. A focused trial gives a team a way to find that value without changing the whole business at once.
Pick one task where you can inspect the input and check the output before anyone relies on it. For a first test, that might be turning approved meeting notes into a draft action list.
Set the Standard Before the Test
In a team meeting, we were talking about asking an AI assistant to make new tasks from the notes. I told the team I wanted someone to review those tasks before they were posted.
I usually ask for priorities and due dates, too. Those need checking. Putting a person’s name and a date beside a task can make it look like everyone agreed to it.
For a first test, use notes from a meeting you attended. Read the draft list against the notes and check what was agreed to, who owns it and when it is due. I don’t want an AI-generated task to become someone’s responsibility just because it appeared on a list.
Your instruction to the tool could be:
Turn these notes into an action list. Keep the stated task, owner, and date. Flag missing details. Do not turn a possible idea into an agreed task. Do not invent a person or a deadline.
Keep a Person on the Review
AI can produce a clear, confident answer that is wrong. That is why the check belongs in the task from the start.
One word in the instructions Claude had drafted for a client’s website caught my attention: “only.” It had called the client the only luxury community builder.
I was walking a teammate through the work, and I said I wasn’t sure we could claim that. So I removed the word and kept going.
That’s an easy thing to miss when you’re reading a draft that sounds good. Before I use it, I need to know which claims I can support. In this case, I couldn’t support “only.”
Follow the company’s rules for the tool and the information you put into it. Keep a person responsible for the review.
Count the Whole Job
Record the time spent preparing the input, running the task, and checking or fixing the result. Compare that full effort with the current way of doing the job.
Fast output is useful only in the context of what comes next. If someone has to untangle invented tasks, that time belongs in the review.
Track quality, too. Note missing items, wrong owners, and dates the source never gave. A trial can also help someone spot gaps they would have missed. Keep that benefit separate from any measured claim about time saved.
Use more than one input. Include an untidy one with missing details and a suggestion nobody agreed to act on. The tidy demo should not be the only case the tool sees.
Decide What Earns a Place
When the trial gives you enough evidence, decide what to keep, change, or stop. If it works, save the instruction, a checked example, and the review steps so someone else can repeat it.
Choose one task this week. Give it an owner and a quality check before the first run. Compare the whole job, including the cleanup.
Let the next use of AI earn its place in the work.
Common Questions
What makes a good first AI pilot?
One task where you can inspect the input and check the output before anyone relies on it. Turning approved meeting notes into a draft action list is a good example: you can see what went in and judge what came out.
How do you know if the pilot worked?
Compare the whole job with the current way of doing it. Count the time to prepare the input, run the task, and check or fix the result. Note the errors you caught. Keep any benefit separate from a measured claim about time saved until you have the evidence.
Source for AI accuracy and checking claims: Claude Help Center.
