The Killable Hour

In yesterday’s Wall Street Journal, the global chair of Baker McKenzie, Sunny Mann, announced that the firm is moving away from the billable hour, towards value-based pricing as its default.¹ AI-assisted work, Mann argued, makes the hourly rate an increasingly meaningless metric. Coming from one of the world’s largest law firms, that’s a significant signal. It is likely that AI will kill the billable hour, but what would killing it actually kill?

»An hour of human attention is no longer a necessary ingredient in producing an hour’s worth of professional output«

The history of the billable hour is more interesting than its reputation suggests. In 1913, Reginald Heber Smith became counsel to the Boston Legal Aid Society, which handled around 2,000 cases a year on a charitable budget. Harvard accounting professor William Morse Cole helped Smith introduce cost accounting, which reduced the Society’s net cost per case from $3.93 to $1.63 by 1915. When Smith joined the Boston law firm Hale and Dorr in 1919, he brought the approach into private practice. Lawyers began recording their time in six-minute increments, making their work measurable and chargeable. The client was no longer paying just for an outcome, but for the professional time spent producing it.²

The arrangement bundled judgement and execution. Both required the same expensive resource, a trained person’s time. The hour concealed the distinction. But today’s AI makes that distinction impossible to ignore. Drafting, reviewing, searching, and comparing can increasingly be delegated to machines. Execution becomes cheap, while judgement about what should be done, and whether the result is any good, remains consequential. The two no longer share a cost structure.

In his 1974 book Labor and Monopoly Capital, political economist Harry Braverman described how industrialisation transformed skilled work.³ As factories grew, the planning and knowledge of production were increasingly taken over by managers, leaving workers to carry out tasks that had been designed by others. Braverman called this the separation of conception from execution. Artificial intelligence introduces a new twist: the same separation can now happen within a single professional’s job, as a human retains the judgement while a machine does the execution. Eventually, even the judgement may be automated.

Consider architecture. An architect might be paid based on the cost of the building, or the hours spent producing its drawings. Neither tells us much about the quality of the design. The first can reward designing a bigger and more expensive building; the second can reward a slower design process. Imagine AI doing in minutes what once took a thousand hours of drawing. The architect has saved time, but also lost a thousand billable hours. Petter Wallberg, co-founder of the AI startup AI-BOB, has pointed out that automating work on a project by Dutch architecture firm OMA in Eindhoven could »tear apart the economics« if the same project were carried out in Sweden.⁴ Why? Swedish consultants typically charge by the hour, while their Dutch counterparts charge by the square metre. The same productivity gain could result in a substantial loss of revenue in one market and a competitive advantage in another.

A similar model already exists in UK infrastructure contracting. NEC Option C agreements set a target cost, with savings and overruns shared between client and contractor.⁵ The same principle could apply to AI-assisted work: agree on the process, set a stopping rule, and share the savings.

Fixed fees, however, introduce another problem. The hour had one advantage: it was auditable. A client could inspect a timesheet and challenge it. A fixed price reveals what the client pays, but not how the work was done. In Capital, philosopher Karl Marx described how a finished commodity conceals the labour behind it. The buyer sees the product, not the process that produced it.⁶

And what happens when an AI agent becomes better than human professionals at directing other agents? The human contribution becomes harder to define, and harder to price.

So what does killing the billable hour actually kill? The idea that professional value can be measured by the time someone spends sitting at a desk. An hour of human attention is no longer a necessary ingredient in producing an hour’s worth of professional output.

The harder task is finding a new way to account for what remains: responsibility, judgement, and the value of an outcome. We’re learning how to make machines do the work. We still haven’t figured out how humans should get paid for it.

References