Breakdown
Sign-Off Assumed Someone Had Already Thought About It
The old review check partly worked because polished work used to carry more evidence of the effort behind it.
A polished document no longer tells you much about the thinking behind it. Review the decisions instead of the artifact: what was considered and rejected, what the recommendation depends on, and what would have to be true for it to be wrong.
It is nine at night and the deliverable goes out in the morning. The structure is right, the options are laid out, the recommendation is clear, and there is nothing in it you can point at and call wrong. You still cannot tell whether anyone actually weighed those two options, and when you ask, you get “yeah, I looked at that.”
What a thorough-looking document used to tell you
Work in that shape often took considerably longer to produce, and days of effort often meant somebody had been forced to think while producing it. It was hard to write a clean comparison of two options without having compared them. The effort and the appearance of effort traveled together, most of the time, and that was enough.
Signing a document like that in twenty minutes rarely meant reading every line. You were reading for signs of work: the caveat awkward enough to be real, the number that traced back to a source, the option that had clearly been argued down rather than mentioned. The proxy was cheap, and for a long time it was useful enough to keep relying on, and it let you clear a stack in an evening.
In many teams, nobody deliberately designed that check. It came free with the cost of producing the work, which is why it never appeared in any process document and why nobody noticed when it stopped carrying weight. It belongs to the same family as the mistakes that used to get caught by accident, where the control existed because of how the work happened to get done, not because anyone built it.
Why you cannot name what is missing
What changed is that a competent-looking draft is no longer expensive to produce. The document in front of you could have come from someone who used a tool to draft it and then checked every claim against what they know about this client. It could have come from someone who used a tool and skimmed the output. It could have come from two careful days of work. The artifact can look similar in all three cases.
You may also not be told which one you are holding, and in many firms the question of what counts as acceptable help has never been settled out loud, so there is no clear thing to disclose against.
So the uneasy feeling is accurate and hard to name, because what is missing was never visible on the page. You were never inspecting thought. You were inspecting a side effect of thought, and the side effect has come loose.
What people reach for first
Common responses include a disclosure rule, a ban on using these tools for client-facing work, and an extra reviewer in the chain. Each one is understandable, but none of them necessarily restores what you lost.
Disclosure asks people to report something that can feel like admitting they did less, which is a weak foundation for a control. Even an honest answer tells you how the draft was produced, not whether the reasoning happened. A thoroughly reasoned piece of work can be drafted with help, and a document typed from scratch can skip the thinking entirely.
A ban can move the use somewhere you cannot see, and it also gives up help that is genuinely useful. Much of what producing a draft requires is closer to a rule than a judgment, and handing that part over is not the problem. The extra reviewer may be the most expensive answer of the three. That one can start eating your week and still leaves you inspecting the same surface that has become less reliable.
Reviewing the decisions instead of the document
The useful change is to stop asking the document to prove something it cannot prove, and to ask the person instead. Three questions can do much more useful work than another pass over the document. What did you consider and decide against, and why. What does this recommendation depend on. What would have to be true for this to be the wrong call.
An answer carrying specifics about this client, this constraint, this thing that nearly went wrong, is usually a better sign than a restatement of the document, though specifics alone are not proof. What you are listening for is whether the reasoning exists somewhere outside the file. If it does, the document is a report of a decision. If it does not, the document may be standing in for a decision nobody really worked through.
This conversation has a real cost, and it is worth naming rather than pretending it is free. Walking someone through why an option was dropped takes longer than the paragraph that mentions it, which is the same tension as work where checking it is quick and explaining it is not. That cost drops considerably if the questions land earlier, while the options are still live, instead of at nine the night before when the only honest answer left is a defense of what already exists.
It also does not belong on everything. Most of what leaves the building carries little risk, and a pleasant competent document is a perfectly good outcome for it. The question is which pieces of work would hurt if the comparison paragraph turned out to be decoration.
Pick the handful of deliverables where a wrong recommendation actually costs you something, and change what sign-off means for those. Not a closer read of the prose, which is the check that stopped working, but a short conversation about what was weighed and what it rests on. The rest can keep the review they already get. If you finish a document and cannot say what the recommendation depends on, you have not reviewed it, you have proofread it. That was arguably true before any of this, and it was easier to get away with.
How we architect AI systems
How we architect AI systems
About the author
Missy Ross
Missy Ross is the Founder and AI Architect of Vero Dawn, an AI architecture and solutions company that helps businesses see what could work better, determine the right solution, and bring it to life. Before founding Vero Dawn she spent 21 years in internal audit across banking and manufacturing, at Audit Director level.