Time for AI honesty

Let me just clarify something right from the start.

Using AI is not cheating. 

Using it and pretending you did not is. 

Here follows a practical scale for being honest about what the tool did.

A note, in the spirit of the thing. Parts of this essay were generated by AI from my own notes and a long back-and-forth. 

The thinking is mine; the drafting is not all mine. By the scale I propose below, this piece sits in the middle: my thoughts, written with a machine. It would be a strange essay to write any other way.

The thing nobody admits

When the first good chatbots arrived, something odd happened. Almost everyone started using them, and almost nobody said so. 

It became the Ozempic of knowledge work: a quiet, widespread habit that people were faintly ashamed to confess. Writers used it and called the output their own. Students used it and filed it as homework. Engineers used it and pasted it in with a straight face.

Then, to hide the seam, we invented a whole genre of self-deception: the “humanise” pass. You generate the slop, then you run another tool over it to scrub the tells, to rough up the edges until a detector, or a colleague, cannot quite prove what made it. 

The result is the worst of both worlds. It is not your voice, and it is not the machine’s competence either. It is laundered. It is garbage with the fingerprints filed off.

I read a document recently that someone had posted as their own work. Above the first heading, still visible, was a line that read: “Sure, I will draft the document for you.” The rest was generic AI prose with a few sentences edited in by hand, tonally mismatched. The author had not even scrolled up far enough to delete the instruction.

The strange part is that the use was never the problem. The hiding is.

Say what the tool did

Here is what I think honesty actually looks like, and it is not a confession. It is a label. Say how much of this was AI, and at what level. You can give a percentage if you like, but the more honest signal is the kind of involvement, because “thirty percent AI” means nothing if you do not say which thirty percent.

There is a simple scale, and it works for prose and for code:

High. The idea itself came from the machine. You prompted, it produced, you picked. The originating thought was not yours. This is not inherently wrong. If the goal is to refactor 800 repositories as quickly as possible, automate it into a repeatable workflow rather than hand-changing each one and pretending you did it yourself. The dishonesty is in the pretending, not in the automation.

Medium. The thinking is yours; the drafting is the machine’s. A long discussion, a pile of your own notes, a position you already held, turned into clean prose or working code by the tool. This essay sits here.

Low. The work is yours; the tool tidied it. Spellcheck, a proofread, a rename, a second pair of eyes.

For code it reads the same way. Did the AI design the architecture, implement a design you specified, or just review what you wrote and catch a bug? Those are very different claims, and collapsing them into a silent “I built this” is where the dishonesty lives.

None of these levels is shameful. The label is not an apology. It is there to stop you claiming a kind of authorship you did not have, and to let the reader weigh the work for what it is. 

And the detail you owe scales with the stakes: a throwaway message needs nothing, a published piece or a result someone will credit to you deserves the label. The rule is simple. Never claim unaided authorship you did not earn.

It is not cheating

Using the tool is not cheating. 

Using it and pretending you did not is.

I would not fell a tree with a chainsaw and then tell you I did it with an axe. 

Not because the axe is nobler, but because the claim would be a lie, and a pointless one. 

The chainsaw is a tool. It extends my hands. It does not make the tree any less felled, or the choice of which tree, or where it lands, any less mine. 

AI is the same. It extends the reach of your thinking into prose, into code, into things that used to take a week and now take an afternoon. 

The judgement is still yours. The taste is still yours. 

What you must not do is launder the help out of the story.

Tobias Lekman, Cloud & Security Architect