Plea to the Reprogrammer
Part 1: Do more for the same
Today I was at an airport, awkwardly early for my flight but conveniently enough to get food in advance. My flight was squarely atop lunch time. I opted for a quick sandwich to go from Pot Belly. This is a medium-large international airport, and I’m quite accustomed to throngs of passengers jamming the terminal when several flights arrive in short succession. So I expect vendors to be prepared to handle traffic.
What I didn’t expect was how. The queue was eight people deep, and even with that mild delay, they sent someone down the line with a tablet to take orders. Normal enough. But when I got to the register, one employee was running four cash registers. I say “cash registers” because that’s what they used to be – now they are point-of-sale terminals, and the majority of patrons were paying by app or card. So it makes a certain amount of sense that the time per-payee is small, and – like a multi-threaded application – the “cashier” can multitask, teeing up a transaction on each terminal while a customer completes the sale on the previous terminal.
This got me thinking about job expectations and compensation. Is it better that, through this employee, Pot Belly can process more orders faster? Optimized for speed and revenue capture, the answer is clearly yes. However, not everyone would be as successful at multitasking. And yet, I expect each cashier is paid the same (minimum) wage, regardless of their throughput.
And so we’re back to AI. It’s clearly good for the company if its employees produce more widgets per time than they used to – and even better if the compensation for the employee is unchanged. Is it better for the employee? Acolytes of the hallowed meritocracy will tell you that the more “productive” employees will be compensated for their superior performance. But I submit that they should be paid for each register they run. So if a single employee can now do what four used to, should they not be paid four times a single historical salary?
No, of course not. The means of performing that job have changed. It is now expected that a cashier runs more than one register at a time. You can almost justify this if you reduce the job to only throughput. Input equals button A, button B, output equals button C. If the conveyor belt brings an endless supply of input, then the job is literally throughput (minus any potential upsells that are possible at the cashier stand).
Perhaps this is the difference between a “skilled” job and an “unskilled” job. In a skilled job, it’s not reducible to just throughout (hence the term “skilled,” implying not only an ability, but also a discernment to apply to that ability). While it’s quite possible to quickly do something that requires a skill, unless the task is repetitive, there is a level of consideration and discernment that needs to be applied. So you can argue that a skilled job, if demanded that it increase its throughput, should be paid in equal terms of the multiples of jobs it has fulfilled.
Now let me give you a magic tool. This tool allows you to do four times as much as you did before. But it requires that you spend more time applying your expertise in the consideration and discernment stages than you previously did; you do more of everything but the production. With this tool, you can do in 15 minutes what used to take you 60. So, if you work for 15 minutes and stare at a wall for 45 minutes, your hourly productivity is the same as it was without the magic tool.
But to do so would be anathema, as you’re not compensated for your time alone, but your time and productivity. So where is the compensation for your time, in this formula? You could almost argue that the 45 minutes are a time savings for you. And that if you can do all the same work in one quarter of the time, your eight-hour workday should justifiably be two hours, and the six that were saved are yours.
That’s not what happens. What happens is your day remains the same length and the expectation is that you simply keep going. The bar has been raised, and it has been raised to a level that benefits the employer. Whether that employer will reward the employee more as part of this shift is entirely up to the employer. And the employer, who funds the magic tool factory, views this expense as overhead to being able to produce more and spend the same amount as they were on people. Perhaps you work for a rare employer who does not subsequently reduce their workforce; surely, then, they will reward employees from their increased profits. Surely?
No. What happens is that a cashier is expected to multiply their efforts for the same pay as they used to receive for single-tracking their efforts. This seems wrong. I’ve tried to convince you, and likely I have not changed anyone’s mind in the course of this post.
Part 2: Here are three emails you need to read
At my work, I’ve heard several of my peers talk about how they’ve used AI to create a daily starter: they have it scour Slack channels, read emails, look at bug reports, etc, and then give them a summary.
This helps them keep up with the pace of information. If not for this, they would need to, themselves, scour Slack channels, read emails, look at bug reports, etc. And something about this feels to me like an orthogonal reality, like what Philip K Dick described in his 1977 address at Metz (transcript): we have had our reality reprogrammed and somehow I have a memory of a different now. And this now doesn’t make sense, because it doesn’t have to be this way.
In that different now, too much information to keep up with is the problem to be solved, rather than how to keep up with an ever-increasing tide of information. Meta levels stacked on meta levels, abstractions of summaries, these are all ways to keep the stream flowing, to maximize the throughput. But we will still be held as accountable as we ever were, we are still expected to be responsible for the details – the very same details we are actively avoiding, because there are just too many to reasonably keep up with. We are no different than that Pot Belly cashier.
I was recently asked if I track my team’s velocity via story points. In that moment, I was reminded of this same optimization: what can we do to make things faster. Because we can do more, therefore we should do more. And my answer to both AI and the raising of expectations due to the magic tool derives from this experience – optimizing for speed is going to be paid for in some other way. You don’t get fast for free; and if it’s not free, then the costs need to be known and agreed upon. Otherwise, the inevitable final cost is going to be a surprise. It’s going to be what “going fast” has gotten my team: churn in definition, scope, and time.
We want faster, but we don’t want to pay for the work. We want faster, but we don’t want to pay for fair. Maybe the problem is that we’ve been optimizing for the wrong thing all along. Maybe the next now we’ll get that right.