Better tools do not automatically lead to better outcomes for workers.
People often talk about productivity as if it settles the argument. If a tool helps a company move faster, reduce costs, or produce more output, then the assumption is that progress is happening, and everyone should simply adjust.
But that is not how power works.
A system can become more efficient while becoming less fair. For example, a workplace can get faster while workers lose discretion, privacy, or meaningful say in how their jobs are changing. A company can even call that “innovation” while the real result is tighter control, weaker bargaining power, or a quieter form of surveillance.
That is why productivity does not answer the labor question, and also why I do not think the real question is whether AI can increase productivity. The harder question is who really benefits when AI productivity rises.
And that is what today’s post is about.
Productivity is not the same thing as progress.
I think that is one of the biggest mistakes in how people talk about AI productivity right now. The conversation often narrows to whether a tool can save time, reduce labor, or speed up decision-making. Those things may matter, but they do not tell us enough. They do not tell us who benefits, who absorbs the risk, or who gets pushed further out of the decision-making process.
Have you ever noticed that once productivity becomes the main focus, other questions start to disappear? Questions about what those weird new workplace changes are, whether the “new system” is ACTUALLY helping people do better work, or why you suddenly inherit three more employees on top of your existing twelve, with or without a raise.
Those aren’t technical issues. They are governance issues. That is why the real question is not just whether a tool works, but who gets a say before those changes become standard practice.
A bigger pie does not settle the labor question.
Before we dive too deep, there is one part of this debate that I do think people should take seriously: the possibility of real AI productivity gains. In Machines of Mind: The Case for an AI-Powered Productivity Boom, Martin Neil Baily, Erik Brynjolfsson, and Anton Korinek argue that generative AI could improve productivity by making cognitive work more efficient and by accelerating innovation over time. They also argue that those gains may take time to appear because adoption, training, and business process changes all matter.
Of course, if there’s a possibility that AI can improve things by making tools that can truly help people, by all means. Plus, some workers may genuinely become more productive, especially when new systems support decision-making, reduce repetitive work, or make expertise easier to access. The Brookings piece makes a serious case for AI productivity gains. It also notes, however, that gains may be hard to measure, unevenly distributed, and accompanied by disruption, wage pressure, or deskilling.
Ultimately, a bigger pie does not tell us who gets more power, nor does it say who controls the “terms of adoption.” It also does not tell us who absorbs the risk when those choices start falling downward onto workers. And it definitely does not tell us whether those gains will make work better for the people (actually) doing it.
The real choice is not AI or no AI. It is how AI gets used.
That is the part of Choosing AI’s Impact on the Future of Work that I found most useful. Daron Acemoglu and Simon Johnson argue that AI is not heading toward one inevitable future. They describe two paths: one centered on automation, in which firms use AI to perform as many human tasks as possible, and a second centered on augmentation, in which technology helps workers make better decisions, solve more complex problems, and expand human capabilities.
That distinction matters because it disrupts the notion that technology simply arrives and everyone else has to adapt. It does not work that way unless we let it. These systems are built, deployed, funded, and normalized by (some of) us, through human choices. The article is very clear that the direction of technology depends on the vision and choices of those in power, not on some natural law of innovation.
That means the real issue is not whether AI exists. The issue is whether we are using it to help people do better work and live better lives, or whether we are mostly using it to strengthen institutions that already have too much power.
That is a much more honest question.
What gets optimized can quietly become the new normal.
This is also where I think a lot of people miss what creeping authoritarianism can look like today. It won’t arrive through some dramatic break from the past (it sort of has in some ways though), but it could show up through a process (made by humans). It could also be reflected in workflows, “optimization” language, and management practices that frame worker input as friction (and/or surveillance as common sense). Once that happens, decision-making can start disappearing into the system.
What do all those have in common? Each and every one is made by human choice.
Once that starts happening, a company no longer needs to say it is reducing worker autonomy. It can simply say it is improving efficiency, without admitting that a system is tightening control. It can also say it is increasing consistency, but does not need to explain why workers are being monitored more closely. You can call it whatever you would like: productivity, compliance, or performance improvement.
In any direction, that is part of how a new normal gets built. Over time, that is also how harm becomes routine inside a system and once a new normal hardens into policy or practice, it becomes much harder to challenge. Not because people suddenly agree with it, but because they begin to treat it as inevitable.
Worker voice matters before efficiency hardens into policy.
This is where Acemoglu and Johnson’s argument becomes especially important. They argue that labor needs a voice in how new technologies are used. In the paper, they mention that matters not only because workers can resist excessive labor-cost cutting, but also because they often know which parts of their jobs would actually benefit from automation and which would not. They also argue that worker buy-in can reduce the push toward intensified monitoring and surveillance, while making it more likely that AI productivity gains are shared more fairly.
That point should not be treated as a side note. It is central.
Worker voice is not just about fairness. It is also about knowledge. The people doing the work usually understand much better than senior leadership which processes are broken, which tasks are repetitive, which decisions require judgment, and where automation could genuinely help without stripping away dignity or discretion.
That is why I do not think worker choices matter in and of themselves. Collective choices matter more.
It takes workers, professionals, unions, communities, policymakers, and ordinary people being willing to question what is being optimized in the first place. The same is true online, where human counterspeech still depends on people being willing to speak up together. Otherwise, “better tools” can simply become better tools for control, and we forget that freedom has to be built through collective action, not just declared from above.
This is not anti-technology. It is about whether people still get a real say.
To be clear, none of this means AI should be rejected outright. That is not my point. The point is that too much of the public conversation gets trapped between panic and hype (looking at you, marketing-led AI-solutionism). AI will not save everything, nor will it destroy it. AI is not a panacea. Either way, both frames flatten the actual choices in front of us.
I am more interested in whether we are approaching new opportunities with enough foresight to ask what kind of society they (we) are building.
Are we using these tools to support human judgment, reduce repetitive work, and make important work more “sustainable?” Or are we mostly using them to concentrate power, cut labor, intensify surveillance, and make corporations stronger and richer?
It may be possible to have both innovation and broader public benefit, but better systems do not appear on their own.
Here’s the full circle back to human choice: it depends on governance, accountability, worker voice, and whether people are willing to push back before the terms of the future get locked in by default. Acemoglu and Johnson explicitly argue that a more pro-worker direction will not happen automatically and will require pressure from labor, civil society, and government.
That feels much more realistic to me than pretending better tools automatically create better outcomes.
Or confusing demo fluency with production competence, but I’ll save that for another post.
At the end of the day
Productivity, better tools, and innovation matter, but none of them alone answer questions that truly matter.
What matters more is who gets a say, who absorbs the risk, and whether the people most affected by “new systems” have any real power to shape how those systems get used. It is also about what gets enforced once a system moves from rollout to routine use. That is true in workplaces, in policy, and in the quieter ways in which institutions teach us to accept what is normal.
So no, I do not think the question is whether AI can make work more efficient.
I believe the harder question is who really benefits when AI productivity rises, and what workers, citizens, and communities are willing to accept as the new normal.
Because what gets optimized tells us a lot about what a system values.
Not the other way around.