In casual conversations about AI and in conversations with
leaders across organizations, I keep noticing the same thing: people aren’t
using AI to its full potential.
They ask for a summary, a draft, five ideas, or an answer to
a question. That’s a little like hiring an exceptionally capable senior
professional and limiting them to whatever you happen to think of.
The better analogy is management. More specifically, being
the CEO. A CEO doesn’t need to know more than the CFO about finance or more
than general counsel about law. Their job is to set direction, provide context,
establish boundaries, review important work, make decisions, and own the
result. That’s exactly how you should be using AI.
Here are 7 ways to improve what you’re doing.
#1 Delegate the work, not the accountability. AI can
already research, analyze, draft, calculate, critique, recommend, and
increasingly act. That’s great … until something goes wrong. At that point,
“ChatGPT told me to” remains a horrible excuse. If your name is on the work,
you own it. If an AI agent sends something because you gave it permission, you
own that too. AI can take on more of the workload, but it should NEVER take on
your accountability.
Have you ever worked for someone who takes the credit when
things go well and shifts blame when they don’t? Don’t be that person when it
comes to AI. Maybe this point is about leadership in general, but I digress.
#2 Give direction, not just instructions. There’s
been a lot of emphasis on writing better prompts, and I teach prompting, so I’m
certainly not arguing against that. But sometimes we confuse detailed
instructions with good direction. If I hired an experienced senior employee, I
probably wouldn’t (shouldn’t) explain an assignment and then dictate every
individual step they should take.
AI generally performs better when it understands what you’re
trying to accomplish, who the work is for, what matters, what constraints
exist, and what a good result looks like. Sometimes it needs detailed
instructions. Sometimes it needs room to figure out how to approach the work.
Knowing the difference is where amateur AI users lag and exceptional AI
managers excel.
#3 Give specific feedback. “Make it better” is not
useful feedback. Neither is “I don’t like it” or “try again.” I’ve worked for
people like that and I understand, personally, how little that helps – human or
machine.
Better feedback is: “This assumes too much knowledge from
the reader.” “You buried the recommendation.” “This doesn’t sound like our
organization.” Once you explain what missed the mark, the AI has something
useful to work with. We sometimes treat iteration as evidence that the original
prompt failed. Poor leaders don’t give actionable feedback, poor AI users do
the same.
#4 Use checkpoints instead of micromanaging. If AI is
helping with something complicated, review the direction before you review the
finished product.
For a long report, look at the structure first. For a
recommendation, review the assumptions. For a presentation, agree on the
argument before polishing the slides.
Just like managing people, if the underlying direction is
wrong or incomplete, you’re better off catching that early before having your
team or AI write 2,000 words. Catching a bad direction early is much more
efficient and less frustrating than fixing it after all the work has been done.
#5 Define decision rights. As AI moves into agents
and connected systems, this matters a lot more. Good managers decide what
people can handle independently, what requires consultation, and what needs
approval. AI needs the same clarity. Research vendors? Fine. Compare them? Kind
of fine. Recommend one? Probably not fine. Sign a contract because it found a
promotional rate? Definitely not fine.
Just because you “empower” an employee or AI to do something
doesn’t mean you should. You need to think through the ripple effects of
decision-making. Capability and authority are not the same thing.
#6 Build context over time. A lot of people still
start from scratch with AI every day. New chat. New task. No context. Imagine
managing a senior employee who forgot everything about you and or organization every
morning. You’d spend most of your day reintroducing yourself. In reality, that
would never happen. There would be a baseline that you build upon with that employee.
You should be working with AI the same way.
Most AI platforms have memory that you can curate over time.
That is akin to having an employee learn the context of an organization and of
YOU. The more AI understands your standards, preferences, audience,
terminology, and examples of “good” work, the more useful it becomes.
#7 Ask it to challenge you. If every person around a
CEO tells them every idea is brilliant, that CEO has a problem. Hopefully you
remember the children’s story The Emperor’s New Clothes. The same applies to
AI.
Too often, we use it to strengthen whatever we already
think. We give it our idea and ask it to make the argument better. If you’re a
good manager or AI user, you’re asking: What am I missing? Where is my
reasoning weakest? What would someone who disagrees with me say? Is there a
better approach?
We have access to systems trained on an absurd amount of
human knowledge, and one of our favorite uses is asking them to reassure us
that the idea we already had was excellent. That is a waste. A useful thinking
partner should sometimes make you uncomfortable.
One very important caveat. There is one important
limit to the analogy of treating an AI like a senior employee. It actually
isn’t an employee. It doesn’t have lived experience, professional
accountability, or any particular concern about what Mark in finance thinks.
That’s why human involvement (your involvement) really matters.
As AI gets more capable, our job starts to look less like
operating software and more like directing capability. You provide the context,
set the boundaries, review the important work, decide what authority it gets,
and remain accountable for the result.