Dark gains

Why you'll be forced to surface them soon

Everyone loves finishing a task faster than expected. You feel good, you feel smart. Your boss is happy. Dealing with gain is a much easier problem than coping with slip.

When I first started working, I used to proudly tell my boss when I made a gain. Then it occurred to me perhaps I could use some of the time I’d saved to make my life better? So I’d ask could I improve the build system? Create a debugging tool?

Often the answer was 'no' - the boss had other plans for the time. A few times they’d ask me to write a proposal. In theory this was a way to justify spending the time, but I soon realised it was an avoidance tactic. Pretend to engage, and then use the proposal as a reason to say no.

Now you might argue the boss is better placed to see the bigger picture. Can optimize across a wider set of people. But you’ve got to think of incentives. If someone gives you something and you consistently offer nothing in return, chances are the offers are going to dry up. And that’s what happened to me. I realised I could spend much of the gain making my life better and still have a sliver left over to offer to the boss. They’d still be happy (a gain is a gain, right?) and my life would be better. Everyone wins. Sort of.

Right now, we’re in a similar situation with AI. AI enables individuals to work much faster. Studies show different gains but it seems to be somewhere between 5-100% depending on task. But what incentives are there for employees to declare gains?

Not many. Your company might fall behind your competitors who are embracing AI. Your company might be slightly less profitable. But those are not strong incentives for individuals.

In reality it’s even worse as there are active disincentives. AI means fewer people can do the same work. Embracing AI might be making yourself (or your colleagues) redundant. This isn’t about hiding gains, it’s about self preservation. You are much more likely to hide a gain if it results in you losing your job.

Dark gains

And so we end up with dark gains. Gains that individuals make through using AI, but don’t declare to their boss. Or only partially declare. Gained time which ends up being spent on distraction - YouTube, reading, chatting. This lack of alignment in incentives means measuring AI gains in the corporate world is going to be hard.

And I know this myself. I’ve found myself building new tools just because I can. Or using saved time to explore AI theory. Others tell me similar stories - they’ve used some of the AI gains to kick back a little. And when the gains can be so big, it’s easy to hide a little bit. If I can complete a 5 day coding task in 1 day then taking a day to kick back is hard to spot.

But for the self-employed it’s reversed. AI gains have a direct impact on your bottom line. Frittering away the gains only hurts you. If you are working for yourself the incentives are aligned. Could AI provide a significant boost to the self-employed and promote the growth of self-employment?

A key challenge for organisations - and for society - is going to be working out how to correctly incentivize the gains. On the surface it seems this will be approximately impossible in large corporations. But, there’s a darker reality. One that makes it highly likely employees will be forced to surface dark gains.

Consider established, stable industries. The amount of work is fixed; head-count is stable. The companies hire a fixed number of entry-level new starters each year to replace retirees and leavers.

AI is often described as being akin to an hyper-enthusiastic intern or new graduate. So replacing entry-level roles with AI is an obvious first step. Staff cost money. Fewer staff implies increased profitability. So, pausing recruitment for a year or two to let this play out is an obvious first step. And that’s exactly what Dario Amodei (CEO of Anthropic) noted last week:

AI could wipe out half of all entry-level white-collar jobs — and spike unemployment to 10-20% in the next one to five years.

Existing teams will have to embrace AI since they won’t be given any new employees. Management will be told to make it work. Dark gains are forcibly surfaced.

In the short term this likely strengthens the position of existing employees - managers won’t want to lose the staff they have. But as AI improves it may not just be entry-level roles that are at risk. And given the potential cost savings, companies may look to more than natural attrition to reduce staffing costs.

Conversely, for growth industries, AI may fuel expansion. You can do more with the same headcount. In smaller, growing companies employees are more likely to be aligned with company success. And so more likely to surface AI gains.

But for stable industries with fixed workloads, the math is stark: if AI can do the work of multiple people, you simply need fewer people. Yet few are paying attention. As Amodei noted:

"Most of them are unaware that this is about to happen," Amodei told us. "It sounds crazy, and people just don't believe it."

And so?

In reality, the era of dark gains is likely to be brief. As AI reshapes society, established companies will come under massive pressure to remain competitive. They won’t have time to figure out incentive structures - dark gains will have to be surfaced if they are to survive. When your job depends on implementing AI-enhanced productivity, alignment becomes irrelevant. Survival has a way of clarifying priorities.

And if you’re one of the lucky few currently wondering whether to surface a dark gain, enjoy it. You’re unlikely to have the luxury of making this decision for much longer.