🪪 Standing Behind Our Work Is Becoming the Human Part of AI
For years, the hard part of knowledge work was producing something worth sending. Writing the first draft, building the analysis, finding the right words took most of the effort, and the effort itself acted as a quiet guarantee. If we had spent three hours on it, we probably understood it.
Now the first draft arrives in a minute, and that guarantee is gone. Glean's Work AI Index 2026, a survey of 6,000 full-time digital workers, found that 69% of AI users admit to sending work they have not verified, do not fully understand, or could not confidently stand behind. The words are on the page, but the person behind them may not be.
That matters for what we want to protect. If the writing can be automated, the part that cannot be handed off is the willingness to put our name to it. That willingness is what colleagues, clients and customers are really trusting, and it is the most human thing in the whole process.
------------- Context -------------
Most of us picture the risk of AI at work as a mistake. A wrong number, an invented quote, a confident paragraph that turns out to be false. Those risks are real, and checking for them is sensible. But there is a quieter risk underneath, and it is about ownership rather than accuracy.
Work that nobody has really read can be perfectly accurate and still be a problem. If a colleague asks "Why did we choose this option?" and the sender cannot answer, something has been lost, even if every figure was right. The page was correct. The person was absent.
The same Glean research found that 87% of workers use AI and say it saves them around 11 hours a week, yet only 13% say their organisation is performing significantly better because of it. It also found that people spend roughly 6.4 hours a week on what the researchers call botsitting: feeding context, fixing mistakes and tidying outputs. Much of the promised gain is being spent on the human work of making AI output defensible.
So the human part is not disappearing. It is moving. It used to sit in the writing. Now it sits in the standing behind. And a team that does not notice that move will find its trust thinning without ever finding a single obvious error.
------------- A Polished Page Can Hide an Absent Author -------------
One of the quietest costs of fast output is that it looks finished before anyone has made it theirs. A tidy summary, a well-formatted proposal, a smooth email: all of them signal care. When the care is not actually there, the signal becomes misleading, and the person receiving it pays for the difference.
Picture an operations manager who asks an AI tool to summarise a supplier contract and sends the summary to her director before a meeting. It reads well. In the meeting, the director asks about the termination clause, and she realises she never checked that section. She spends the next ten minutes scrolling the original while three people wait.
Nothing in the summary was false. What the meeting exposed was that no human judgement had been applied to it. The director leaves with a small, lasting adjustment: next time, he will read the contract himself. That is a relationship cost, and it arrives long before any visible mistake.
For Stay Human, this is the point worth underlining. The thing our colleagues rely on is not that the words are correct. It is that someone who understands them is accountable for them. When that someone is missing, trust has nowhere to rest.
------------- Standing Behind Work Is a Practical Skill, Not a Slogan -------------
It is easy to say we should take ownership of AI output. It is harder to say what that looks like on a Tuesday afternoon with 40 unread messages. So it helps to make it small enough to do.
One test that travels well is simply this: could I explain this out loud, to a colleague who disagrees, without reading from the page? If the answer is yes, we can stand behind the work. If the answer is no, we have found exactly where our attention is needed, and it usually takes a few minutes to close the gap.
Consider a consultant who uses AI to draft a client recommendation. Before sending, she spends 12 minutes doing three things. She checks the two figures the whole argument rests on. She reads the recommendation aloud once. And she rewrites the single sentence she would not have said herself. The document is still mostly AI-assisted, but the argument is now hers.
That is what a person staying in the loop looks like in practice. It is not a heavy review process. It is a short, honest moment where a human decides whether this is something they would defend if challenged.
The outcome we are protecting here is judgement. The consultant keeps her credibility with the client, and the client keeps a reason to trust that a real adviser is behind the page.
------------- Accountability Is Becoming the Part of Work We Cannot Delegate -------------
Teams are already sorting tasks into what AI does and what people do. The less obvious question is what only people can do, and accountability sits near the top of that list. A tool can produce a recommendation. It cannot answer for it in a room.
This shows up in how people treat each other's work. If a colleague sends something and we sense nobody has really looked at it, we double-check everything they send afterwards. That second look is a tax on the relationship. It is small each time, but it compounds, and it makes collaboration slower and a little colder.
By contrast, a colleague who says "I checked the numbers and the timeline, but I am less sure about the second option" earns more trust than one who sends something flawless without comment. The honesty about limits is the signal. It tells us a real person looked, and it tells us where to look next.
Picture a team of six where everyone adds a one-line note to AI-assisted work: what I checked, and what I am unsure about. After a month, reviewers stop re-reading everything and start reading the flagged parts. The human judgement has become visible, and the trust follows.
For our working relationships, this is a clear gain. Visible ownership is warmer than polished anonymity, and it keeps the people on a team feeling like colleagues rather than editors of each other's machine output.
------------- Leaders Set the Standard for What Counts as Finished -------------
None of this works if the culture quietly rewards speed over ownership. If the fastest sender is praised and the person who paused to check is seen as slow, people learn quickly what "finished" means. The Glean figure of 69% is partly a story about incentives, not about carelessness.
Leaders have an outsized role here because they decide what gets noticed. A manager who says, in a meeting, "I read this properly, and here is the part I changed" gives the team permission to do the same. A manager who forwards unread AI output teaches the opposite, however good the intentions.
Suppose a team lead starts opening her weekly update with one sentence: "Two things I checked, one thing I am not sure about." It takes her 30 seconds. Within a few weeks, her team begins doing the same without being asked, and conversations in review shift from "Is this right?" to "What did you decide?"
That shift is the human upside. People are not being asked to slow down. They are being asked to be present in the work, and that is a more satisfying way to work than being an invisible relay between a tool and an inbox.
For trust and quality, this is the culture worth building. Finished does not mean generated. It means somebody has made it theirs.
------------- Practical Moves -------------
First, before sending any AI-assisted work this week, ask whether you could explain it aloud to a colleague who disagrees. If not, spend ten minutes closing that gap before it goes out.
Second, add a one-line note to AI-assisted work: what you checked, and what you are unsure about. It takes under a minute and tells the reader where their own attention is best spent.
Third, choose the one or two facts that carry the weight of each document and verify those by hand. You do not need to check everything, only the parts that you would be embarrassed to get wrong.
Fourth, if you lead a team, say out loud when you have read something properly and what you changed. Your habits set the working definition of finished for everyone watching.
Fifth, when a colleague flags what they are unsure about, thank them. Honest limits are the most useful thing a person can add, and thanking them keeps the habit alive.
------------- Reflection -------------
The 69% figure is not an accusation. It describes a very human situation: work arrives faster than we can fully absorb it, and the page looks done before we have made it ours. The question is what we do with that gap.
What we are seeing is that AI is moving the human contribution, not removing it. The effort that used to go into writing now goes into judging, and the willingness to say "this is mine, and here is where I am unsure" has become the clearest sign of a real person at work.
In the end, that is the kind of AI shift worth paying attention to. Not how much a tool can write, but whether the people reading our work can still tell that someone is behind it. When they can, trust grows. When they cannot, even perfect output feels hollow.
When did you last send something you could not fully explain, and what would have made it easier to say so?
Which kind of work in your team would most benefit from a visible "what I checked, what I am unsure about" note?
What does a finished piece of work look like to you now, and has that changed since you started using AI?
34
10 comments
Igor Pogany
8
🪪 Standing Behind Our Work Is Becoming the Human Part of AI
The AI Advantage
skool.com/the-ai-advantage
Founded by Tony Robbins, Dean Graziosi & Igor Pogany - AI Advantage is your go-to hub to simplify AI and confidently unlock real & repeatable results
Leaderboard (30-day)
Powered by