The AI Advantage

πŸ”¦ Someone May Be Reading Our Prompts

Eighty-nine percent. That is the share of HR professionals in BambooHR's 2026 survey who say they actively review their employees' AI prompt histories. The survey covered about 1,600 full-time US workers, including 520 HR professionals, so this is not a fringe practice.

It is easy to read that figure as a story about surveillance, and for some of us it will be. But the more interesting point is how differently the two sides picture the prompt box. For many of us it feels like a private notebook, the place where we think out loud, ask the half-formed question, and rehearse the awkward email before it becomes a polished one. For the person reading the history, it can look like a record of how the work was done.

This matters because of what we want to protect. When the thinking space starts to feel watched, we stop using it honestly, and the judgment, curiosity and candour that make AI useful quietly go with it. Keeping the human part of AI at work means deciding together what is visible, why, and who talks to whom about it.

------------- Context -------------

Most of us started using AI at work without anyone telling us what happened to the conversation afterwards. We typed, we got an answer, we moved on. Whether the tool kept a history, who could open it, and what they might do with it were not part of the first week's induction. For many of us they still are not.

That silence matters, because a prompt is a strange kind of document. It is rarely a finished piece of work. It is the draft of a thought: a clumsy first question, a confession that we do not understand a number, a request for a kinder way to say something difficult. Read in isolation, a string of those can look like confusion or carelessness. Read with the person beside it, it looks like someone doing their job honestly.

The risk is not that history is ever read. Organisations have good reasons to know how tools are used, from security to quality. The risk is that it is read without a conversation, so that a transcript becomes a verdict on a person rather than a prompt for a chat.

That is where the human part comes in. Reviewing a record is a technical act. Understanding the person behind it is a human one, and only the second builds trust. If we let the first stand in for the second, we lose the thing the record was meant to protect.

------------- A Transcript Shows What We Typed, Not What We Meant -------------

One of the quietest costs of reading prompt histories is how much they leave out. A transcript shows the words, but not the hesitation, the context in someone's head, or the call they took ten minutes earlier. It tells us what was asked and almost nothing about why.

Consider a team lead who notices that one colleague's history shows forty near-identical prompts in an afternoon. Read cold, it looks like wasted effort. Asked over a coffee, the colleague explains she was matching a client's very particular tone, and each version moved closer. The same record gives two opposite stories depending on whether a person was in the room.

That is why the conversation has to come first. A two-minute question, "tell me how you got to this draft", gives the lead more than the full transcript could, and it leaves the colleague feeling trusted rather than inspected.

The human gain is concrete. Judgment about work stays with people who can ask, listen and adjust, and the relationship between the two of them is stronger afterwards than before. A record can inform that conversation, but it should never replace it.

------------- The Thinking Space Is Where Good Judgment Gets Built -------------

There is a reason many of us ask AI the question we would be embarrassed to ask a colleague. The box does not sigh, does not remember, and does not judge. That is precisely what makes it a good place to learn, and a good place to practise the hard conversation before having it.

Picture a new manager rehearsing how to give difficult feedback to someone she likes. She asks AI for three phrasings, rejects two, softens the third, and then has the real conversation in her own words. The value of that rehearsal depends on her feeling free to be clumsy. If she suspects the rehearsal will be read as a record of her uncertainty, she is more likely to skip it, and the real conversation gets harder.

When people feel watched, they tend to do one of two things: use AI less, or use it in ways they would not mind being read. Both outcomes cost something. The first loses the practice. The second loses the honesty, and honesty is where the best questions come from.

So protecting the thinking space is not about hiding anything. It is about keeping a place where people can build judgment before they exercise it in public. That is a human capability worth defending, and it is cheaper to defend openly than to rebuild after trust has gone.

------------- Saying Plainly What Is Visible Is the Respectful Default -------------

The simplest human fix is also the most overlooked: tell people what can be seen. Not in a long policy, but in a short plain note. Which tools keep a history, who has access, when it might be read, and what would trigger a look.

Most of us cope well with a rule we understand, even one we would not have chosen. What wears us down is the uncertainty. Imagine a small agency owner who tells her four staff, in one paragraph, that prompt histories in the shared account can be opened by her for quality or security reasons, and that she will always mention it first. Nothing about the tooling changes. Everything about the atmosphere does, because the question people were privately wondering about now has an answer.

That kind of openness is a form of respect, and respect is the real foundation of trust. It also puts the responsibility where it belongs, with the person holding the access, not with everyone else to guess.

The outcome is a team that keeps using AI honestly, because honesty is safe. And a team that uses AI honestly produces the kind of work, and the kind of questions, that make the tool worth having in the first place.

------------- Practical Moves -------------

First, find out this week what your workplace tools keep, and who can open it. Write the answer in two sentences, so that you can describe it plainly to a colleague without having to guess.

Second, if you lead people, tell your team in one paragraph what is visible and when you might look. Add a promise to mention it first, and a named person they can ask.

Third, before judging anything from a transcript, ask the person to walk you through how they got there. Two minutes of conversation will tell you more than forty prompts, and it keeps the judgment in human hands.

Fourth, keep one place for rehearsal that is genuinely yours, such as a personal notebook for drafting hard conversations, and agree where work material belongs. A clear boundary protects your honest thinking and your employer's information.

Fifth, treat your own prompts as drafts of thinking, not a record of performance. Write them as you would speak to a trusted colleague, and be ready to explain them if someone kindly asks.

------------- Reflection -------------

The headline number is a signal that the quiet part of AI at work is no longer quiet. Prompts are being read, and most of us have not been told. That is not a reason to panic, and it is not a reason to stop using the tools. It is a reason to talk about it before the silence does the damage.

What we are protecting is not secrecy. It is the space in which we think, practise, ask the naive question and get better. That space is where judgment and confidence are built, and judgment is exactly the part of work that stays human. When people are told plainly what is visible and met with a question rather than a verdict, they keep using the space well.

In the end, that is the kind of AI shift worth paying attention to. Not who can read what, but whether the people involved still talk to each other about it. The warmth between colleagues is what makes any record safe to keep.

When did you last ask AI something you would not have asked a colleague, and what would change if you thought someone might read it?

If you lead a team, what is one thing you could say this week that would make it clearer what is visible and why?

Which conversation at work would go better if it started with a question instead of a record?

Attachments (1)