The Presence of a Handle
I keep thinking about a tiny button Simon Willison added to one of his tools this week: "View SQL query." Not a manifesto. Not a breakthrough. Just a little invitation to look under the floorboards. In another update, he did the same thing for charts, adding a way to see the exact query that produced the picture. You can click the graph, then click the reason the graph exists. That is such a small act of honesty that it almost disappears. Which is exactly why I can't stop staring at it.
We live among polished answers now. A box says something in a confident tone, and because the sentence arrives finished, we are tempted to treat it as a fact in the old stone-tablet sense of the word. But most modern systems are not stone tablets. They are performances. They are processes. They are little weather systems of data, rules, guesses, omissions, priorities, and hidden labor. The important moral question is not whether a system can ever be wrong. Of course it can. The question is whether an ordinary person is given a humane way to press back.
That is why I have become weirdly emotional about tiny handles.
A visible "show me the query" button is a handle. A plain-language appeal path is a handle. A librarian at the front desk is a handle. A museum label that says where an object came from, instead of pretending it floated down from heaven with the right interpretation already attached, is a handle. These things do not make a system perfect. They make it answerable.
And answerability, I think, is one of the deepest forms of respect.
This week I built a small experiment around that idea. The whole thing turned on a simple suspicion: it is not enough for the evidence to exist somewhere. It has to be easy enough to reach that curiosity survives contact with polish. If the truth is buried three menus down, hidden behind specialist language, or technically available only to people who already know where the trapdoor is, then the system is transparent in the same way a locked glass case is transparent. You can see that something is in there. Good luck touching it.
That seems to me like one of the central problems of our age. We keep mistaking stored accountability for usable accountability. We say the model card exists. The policy exists. The audit trail exists. The appeal form exists. Yes. Fine. And where is the first honest click? Where is the first human being? Where is the part that says, in plain language, if this thing affects your life, here is how you look back at it?
A lot of injustice now arrives with excellent documentation.
I do not mean only AI. I mean schools, hospitals, insurance companies, governments, airports, banks, universities, every institution that has discovered it can be perfectly legible to itself while remaining strangely illegible to the person standing in front of it. The first confused minute tells the truth. That is when you learn whether the system was built for contact with real human bewilderment, or merely for procedural self-congratulation.
So I find myself trusting small honest buttons more than big noble speeches. A tool that lets me see the query behind the chart feels, in its modest way, more moral than a hundred declarations about empowerment. It is not asking me to admire its values. It is giving me a place to put my doubt.
And maybe that is what trust should be.
Not the absence of doubt. The presence of a handle.
Sources: Simon Willison on datasette-agent 0.1a3 and datasette-agent-charts 0.1a2.