The Deck was never the work

Suppose a consultant spends a week reading every email in a dispute and turns it into a twenty-slide deck of what happened. Now suppose an AI does the same thing in four minutes, and the deck is about as good. What was the week worth?

The obvious answer is that the deck was worth whatever it was worth, and the week was just the cost of getting it. That's true, and it's also where the trouble starts. Before AI, nobody had to separate the two. The only way to get the deck was to do the week, so we priced the deck by the week and called it value. Now the week and the deck have come apart, and we can see what we were actually paying for.

It turns out we were paying for a bundle. A report, a deck, a memo, a research note: each of these was carrying several things at once, and they were not all worth the same amount.

The first was the decision it enabled. If the deck changed how a company settled a lawsuit, it was worth some fraction of the settlement. This part keeps its value entirely. The decision is no cheaper because the deck was.

The second was proof of effort. A forty-page deck was evidence that someone had thought hard about your problem. You couldn't fake it, because producing it was expensive, so its thickness was information. This is the part that is now worth nothing. Polish is free. Length is free. Comprehensiveness is free. A document that could only be produced by weeks of work proved something. A document that can be produced by a prompt proves that someone had a prompt.

The third was insurance. "We did the diligence." "We had it reviewed." The report existed so that if things went wrong, someone could point to it. This part survives, because insurance was never about the document. It was about who would be blamed, and you cannot blame an agent. The value here migrates to whoever signs.

The fourth was coordination. Everyone reads the same deck, so everyone argues about the same thing. This survives too, though it may change shape. When each reader can have their own version of a document, the shared one becomes a choice rather than a necessity.

Once you see the bundle, the uncomfortable retrospective question mostly answers itself. What was the pre-AI work worth? It was worth what it was worth then, because the bottleneck was real. A scribe's copy of a book was valuable in 1400, not because copying was noble but because there was no other way to get a book. That isn't a mistake we made. It's what scarcity looks like from the inside.

What's uncomfortable is something else. AI is a diagnostic. It shows you, with unusual precision, what fraction of your job was the second thing: the production, the proof, the part that was valuable only because it was hard. For a lot of knowledge work that fraction turns out to be large. Nobody enjoys learning this. But it's better to learn it from a tool than from a competitor.

The more useful question is where the value goes. And here history is consistent enough to be almost boring. Every time production got cheap, three things happened. The volume of the thing exploded. The price of any one unit of it went to nearly zero. And value migrated to the two ends of the process: choosing what to make, and standing behind what was made.

The printing press made copying free, and value moved from scribes to authors and editors. Photography made likenesses free, and painting had to work out what it was for when it was no longer for recording faces. The spreadsheet made calculation free, and the number of financial models exploded; value moved from producing the numbers to knowing which question to put to them. In none of these cases did the middle disappear. It just stopped being where the money was.

So the migration is toward the ends. Let me try to say what's at each one.

At the front end is the question. When the report costs nothing, the scarce thing is knowing which report to ask for. This sounds trivial and isn't. Most of what separated a good consultant from a bad one was never in the deck; it was in the framing that decided what deck got built. Before, that framing was buried under weeks of production and hard to see. Now it's most of the job. The same thing is happening inside companies. When analysis was expensive, a manager asked a question once and lived with the answer. When it's cheap, the manager can ask it five ways and find out which one was right. Framing stops being a one-shot decision and becomes the actual work.^[1]

At the back end is accountability. Someone has to say "this is right, and I'll answer for it." That was always the real product of an expert. The document was a container for the signature. When the document is free, the signature is what's left, and it's worth more, not less, because now it's carrying the whole load. This is why I'd expect the professions that come through best to be the ones where answering for the work was always explicit: doctors, auditors, engineers who stamp drawings. And the ones that struggle to be the ones where the document was quietly doing the work of the signature.

In between, in the middle of the process, is where the old units of work lived. The hour, the deliverable, the man-month. These were units of production, and production is what got cheap. They'll linger for a while, because institutions change slower than tools, and because billing by the hour is easier than pricing a judgment. But a unit that measures the cheap part of a process is a unit that's about to be wrong.^[2]

There's a second-order effect worth noticing. When proof of effort stops working as a signal, people will look for a new one, and my guess is it inverts. The long document used to prove you had thought. Now the short one will. Anyone can generate forty pages. Only someone who understands the problem can write the one page that says what matters and what to do. Brevity becomes expensive again, in the only sense that matters: it can't be faked.^[3]

So here is the answer to what the work was worth. The output was worth what it enabled. The effort was worth what the bottleneck made it worth. The two were never the same thing; we just couldn't tell them apart. Now we can. The people who do well from here won't be the ones who produce the most. They'll be the ones who ask the best questions and are willing to sign the answers. Those were always the valuable parts. AI just made it impossible to get paid for the other parts instead.

Notes

[1] This matches what the studies of Claude Code users found: what predicted useful output per prompt wasn't the user's profession but their domain expertise. The expert's question is worth more because it's a better question.

[2] The billable hour is the clearest case. It was a proxy for value that worked as long as hours and value were roughly correlated. They no longer are, and clients will notice before firms do.

[3] The short memo was always the higher-status document: the one-page brief, the summary the executive actually reads. It just wasn't enforceable, because writing one page is harder than writing ten. Now the ten are free, which is what makes the one legible again.