November 10, 2026AI Needs a Performance Review Before You Let It Do More in 2027

We review employees, budgets, vendors, subscriptions, and strategies. We even ask whether the software we bought in February was worth the money. But most of us have spent this year quietly adding AI to our lives without ever reviewing the relationship.

Maybe AI writes your first drafts now. It may research restaurants, summarize long documents, help with presentations, and fix a spreadsheet that has been broken since spring. It may give you ten better ways to say the thing you were already trying to say. All of that is useful, and I’m not here to argue otherwise.

But before we head into 2027 and let AI do even more, I think we owe ourselves a harder question. Did AI make me better this year, or just faster? Those two things are not automatically the same, and the gap between them is where the interesting work sits. I have been thinking about that gap for weeks.

What am I happy to hand over? What deserves a second look? And what should stay mine? Those three questions turned into a kind of year-end review for me. A few guests from our interview series shaped how I answer them, and I want to share what stayed with me.

Why AI deserves the same review as everything else

When a new person joins a team, we check in. We ask what is going well, what is confusing, and where they need support. We do it at thirty, sixty, and ninety days. AI got none of that. It showed up one small task at a time. First it fixed a typo. Then it wrote an email. Then it summarized a meeting, and before long it was shaping how we approach the work itself.

McKinsey describes AI guardrails as the boundaries that keep AI aligned with an organization’s policies, values, and risk tolerance. That makes sense for a company, which has legal teams, risk teams, and written rules. But it got me thinking about my own guardrails. Most of us don’t have policies. We have habits, and habits are easy to miss because they never feel like decisions.

What strikes me is how rarely we look at those habits. Nobody wakes up and decides that a machine should write every first draft. It happens because the first draft is hard and the shortcut is right there. The easy thing becomes the normal thing, and the normal thing stops being questioned. A review is really a way to notice what became normal without our permission.

Speed is simple to measure because you can feel it. Better is harder to judge. Better asks whether the work is more thoughtful, more accurate, or more like you. I think many of us confused the first for the second this year, and I include myself in that. I’m not suggesting we pull back. I’m suggesting we act like the manager in this relationship instead of a passenger in it.

Don’t hand the first thought to a machine

AI is great at giving you a place to begin. That is its biggest strength, and I think it is also its biggest risk. The danger is letting it decide what you think before you have thought anything yourself. A blank page is uncomfortable, so it is tempting to ask a tool to fill it. But that discomfort is often the place where your real perspective starts to show up.

For work that matters, I’m trying to leave room for the messy first thought. What do I think about this? What feels interesting? What bothers me? What am I noticing that maybe somebody else isn’t? Those answers come out rough, and they are not always polished. They are still mine. Then I bring in the tool. I let it organize my thinking, challenge it, research it, pressure-test it, and show me what I missed. The order matters more than I expected.

When the machine goes first, the result tends to drift toward the average. When I go first, the machine becomes a sharp colleague who makes my idea stronger. It is the same tool with a very different outcome, and the only thing that changed was who spoke first. I’ve seen this pattern in my own drafts more than once. The writing that sounds most like everyone else is usually the writing where I skipped my own thinking.

Hall of Fame Speaker Scott McKain has spent his career talking about distinction, which is what makes a person, a company, or an experience impossible to confuse with everybody else.

Scott McKain

His work keeps coming back to me in this context. If a thousand people ask AI the same question, they will get answers that look alike. Distinction can’t come from there. AI can make something more efficient, and it can even help make it better. But it can’t decide what makes you worth remembering. That is a human job, and it starts with the first thought, which is why it belongs in any honest AI review.

🎧 Watch and listen to the full conversation with Scott McKain here.

Let the stakes decide how much you trust AI

I use AI differently depending on what I’m asking it to do. If I’m brainstorming restaurant ideas and one of the suggestions closed six months ago, that’s annoying. If I’m making a financial, medical, legal, hiring, or reputational decision, that is a very different kind of mistake. The tool is the same in both cases. The cost of being wrong is not.

My rule is simple: the higher the stakes, the higher the verification. AI can sound remarkably confident while being wrong. That doesn’t mean you should stop using it. It means you should know what kind of conversation you’re having. A confident tone is not the same as an accurate fact, and these tools are very good at the first one.

For low-stakes brainstorming, play. Let it surprise you. For consequential decisions, ask for sources. Then open them. Check the date. Look for what is missing, because the gaps are often more telling than the answer. I’ve found that this habit takes about two extra minutes, and those two minutes are cheap insurance against a mistake that could take months to repair.

I also think about this in terms of who pays when the answer is wrong. If the mistake lands on me, I can absorb it. If it lands on a client, a patient, a job candidate, or a colleague, the standard has to be higher. That is a useful lens because it moves the question away from the tool and toward the people affected by the decision. AI doesn’t carry that responsibility. I do.

Convenience makes us careless

The same judgment applies to what we put into these tools. We casually paste in internal documents, client conversations, financial information, meeting transcripts, strategy notes, and all kinds of personal details. It feels harmless because it happens in a quiet window on a screen. Nobody is watching, so it feels private. But feeling private and being private are two different things.

Before handing over something sensitive, I try to know where I’m putting it and what protections actually apply. That is a boring question, and it is an important one. A simple test helps me: if I wouldn’t casually forward it outside the company, I shouldn’t casually paste it somewhere I haven’t thought about either. Convenience has a way of making us less careful, and it does that gently, which is why it works so well.

Guardrails don’t always have to be complicated. Sometimes they just make us pause. A pause of three seconds before pasting a client email can be the whole policy. I think leaders in particular should say this out loud to their teams, because people rarely slow down on their own when a tool makes things fast. When a leader models the pause, it becomes part of how the team works with AI instead of a rule nobody remembers.

Keep a human-only list that AI never touches

This may be my favorite guardrail. I think everybody should have a few things they simply don’t outsource. Mine would include a meaningful apology. Difficult feedback to someone I care about. A condolence note. An important personal message. The final decision about another person’s career. None of these are hard because of the writing. They are hard because of the caring, and the caring is the point.

Could AI help me organize my thoughts before one of those moments? Absolutely. I’d be fine with that. But I don’t want it impersonating the caring. There is a difference between getting help with the words and handing over the feeling. A person who receives an apology can usually tell whether someone sat with the discomfort of writing it. Even when they can’t tell, I can.

AI keynote speaker Kate O’Neill has a phrase I keep coming back to: the future isn’t tech or human. It’s tech with human. That distinction matters more each month. The old framing sets people against machines and forces us to choose a side. Kate’s framing is more useful because it puts the question on how the two work together. It asks which part of a moment needs the machine and which part needs a person.

Kate O’Neill

I find this useful for teams too. When I picture a group using AI well, I don’t picture the one that has automated the most. I picture the one that can explain, without hesitation, which parts of the work are machine work and which parts are people work. That clarity builds trust. Clients feel it, and so do colleagues, because they know the human parts are still being done by humans.

Just because we can automate a human moment doesn’t mean the moment gets better when we do. I’d go one step further. Some moments only exist because a human chose to show up for them. If AI shows up in your place, the moment may still happen, but its meaning shrinks. That is why I think the human-only list is worth writing down before you need it.

🎧 Watch and listen to the full conversation with Kate O’Neill here.

AI saved you an hour. Where did the hour go?

This may be the question I’m most interested in. Let’s say AI saved you an hour this week. What did you do with it? Did you spend it thinking? Talking to a client? Mentoring someone? Doing work that needed your experience? Going home earlier? Or did the newly empty space fill with more work almost immediately?

I ask because I’ve watched how this tends to go. A task that took an hour now takes ten minutes, and nobody announces the saved fifty. The calendar simply absorbs it. Someone adds another meeting or another deliverable, and the gain disappears before anyone can enjoy it. We call it productivity, but it can look a lot like running faster on the same treadmill.

I don’t think this is a flaw in people. It’s a feature of how work gets organized. Unless someone names the time as protected, it belongs to whoever asks for it first. That is why I think the question has to be personal. If you don’t decide what the hour is for, the system will decide for you, and the system usually wants more output.

Greg Verdino talks about AI as an opportunity to move people toward more meaningful work: relationships, creativity, innovation, and empathy. He doesn’t frame it as a way to accelerate everything already on the list. That feels like the bigger promise to me. Efficiency isn’t particularly interesting if all it does is create more room for busyness.

Greg Verdino

What I find interesting is a different possibility. Machines can take some repetitive work off our plates and give us more time for the things that still need a human being. That might be a hard conversation with a teammate, a slow lunch with a client, or an hour of quiet thinking before a big decision. Those things rarely show up on a dashboard, but they are often what people remember. The leaders who get this right will decide on purpose where the saved time goes. The rest will find it already spent.

🎧 Watch and listen to the full conversation with Greg Verdino here.

Start with the problem, not the platform

One more thing before 2027 arrives: you probably don’t need seven AI platforms. You don’t need to become fluent in every new tool that launches or feel behind every time LinkedIn tells you somebody built an agent that runs their entire company while they sleep. That kind of story creates pressure, and pressure is a poor way to choose tools. It makes people collect software the way some people collect gym memberships.

Start with the problem. Where do you repeatedly lose time, attention, or information? Maybe it’s research. Maybe it’s digging through your own documents. Maybe it’s starting every project from scratch. Maybe it’s taking something you already know and turning it into a usable first draft. Find the tool that helps with that one thing. A narrow fit that you actually use beats a broad platform that you open twice.

I’d also give each tool a short trial period, much like a probation review. Use it on the real problem for two weeks. Then ask whether it saved time, improved the work, or just felt impressive. If you can’t point to a clear answer, let it go. There’s no shame in dropping a tool that didn’t earn its place.

AI keynote speaker Shawn Kanungo is someone I enjoy talking with about this topic because he pushes people to experiment. Try the weird thing. See what the technology can do. Don’t let fear make the decision for you. I’m all for that. Curiosity is a real advantage, and people who stay frozen will fall behind people who play.

Shawn Kanungo

But experimentation and surrendering your judgment are two different things. You can try ten tools and still keep your standards. I suspect the people who use AI best won’t necessarily be the ones who use it most. They’ll be the ones who know when to challenge it. That skill is quiet, and it doesn’t make for a good social media post. It is probably the one that matters most.

🎧 Watch and listen to the full conversation with Shawn Kanungo here.

A year-end review you can finish in an afternoon

If all of this sounds like a lot, it doesn’t have to be. You can run this review in an afternoon with a notebook and a cup of coffee. Start by writing down every place AI shows up in your week. Be honest and be specific. Include the small stuff, like the quick rewrite of a subject line or the summary you skim instead of reading. Most people are surprised by how long the list gets.

Next, sort each item into one of three groups. Some things you are happy to hand over, such as formatting, scheduling, and first-pass research. Some things deserve a second look, such as anything with a number, a name, or a date that someone will rely on. And some things should stay yours, which is the human-only list from earlier. The sorting itself is the useful part. It forces you to decide on purpose, which is exactly what the quiet drift of the past year never asked of us.

Finally, pick one change. Only one. Maybe you will start with your own notes before opening a chat window. Maybe you will add a verification step to anything that leaves your desk. Maybe you will protect an hour a week for the work AI can’t do. Small changes tend to stick, and big plans usually fade by February. A review that leads to one real habit is worth more than a review that produces a long document nobody reads.

Leave a few things stubbornly human, even as AI improves

I wouldn’t end 2026 by making a list of every AI feature you haven’t tried. I’d ask something simpler. What did AI genuinely make better this year? Where did you become too dependent on it? What deserves more protection? Which decisions still need your judgment? And what parts of your life and work do you want to stay unmistakably yours?

Then there is the question I would really sit with. What will I do with the time AI gives back to me? That may be the most important guardrail of all, because it points past the tool and toward the life around it. Every other guardrail protects something, like accuracy, privacy, or trust. This one protects your attention, which is the only resource you can’t get more of.

The goal for next year shouldn’t be to use more AI simply because we can. Use it where it makes you more capable. Question it when the stakes rise. Keep responsibility for the outcome. Protect what’s private. And leave a few things stubbornly, wonderfully human. If you do that, your review won’t end with a score. It will end with a clearer sense of who is in charge, and I think that clarity is the real gift of a year-end review.

🤖 Want to book an expert who can help your team review this year’s new tech without the overwhelm? Let’s talk before your lineup is set.

📅 Send us your early 2027 meeting dates at info@thekeynotecurators.com.

📬 Find out how to handle digital changes safely; subscribe to our weekly newsletter today.