AI is Making Employees More Productive: How Does the Cost-Benefit Stack Up?

Is AI making life easier or harder? Are people getting more accomplished, or just producing more slop? We’ve asked these questions before. Email, Slack, the MS Office Suite–each one promised to be a boon to productivity and the cure to employees’ stress and overwhelm, only to end up as the bane of most employees’ existence.

Now AI has arrived, and it’s time to consider the cost-benefit analysis.

Understanding the Human Costs of AI

From the human perspective, you can evaluate the cost in terms of the impact on people’s workload: which I’ll define as the effort required to produce the outputs and outcomes their job requires. But that’s not enough. You also need to assess the impact on thoughtload, a term I coined to describe the effort required to manage and metabolize all the cognitive demands and emotional burdens associated with work (and life in general). For most roles, thoughtload far exceeds workload, and importantly, workload can decline without the associated drop in thoughtload.

The impact of technology is a case in point. While technology can have a beneficial effect on workload by making tasks easier or more efficient, it can simultaneously increase thoughtload by introducing more distractions, increasing complexity, triggering anxiety, or depleting energy more rapidly.

Myriad studies have shown that AI contributes to thoughtload. First, it can increase cognitive demands, which are a function of the draws on one’s attention, the complexity of the ideas to be manipulated, and the fatigue associated with the number of decisions required. Using AI, particularly agentic AI, can cause frequent distractions, as the AI bombards the user with outputs they must verify and iterate on. While AI might alleviate some cognitive demands, research suggests that automating more superficial or tedious tasks may concentrate cognitive load into more complex information verification, response integration, and task stewardship. Thus, the benefits of AI for alleviating some cognitive demands must be evaluated in the context of other increases.

The impact of AI isn’t purely cognitive; there are emotional costs as well. Employees using generative AI to co-produce work face heightened levels of work alienation and studies that show “knowledge hiding,” where employees obscure what they know in hopes of protecting their roles, provide additional evidence that people are feeling under threat.

A survey of CEOs by Dataiku highlighted the combined cognitive and emotional impacts of current AI deployments. “AI is a daily operator, but not yet a decision-maker. Leaders trust AI enough to use it, but not enough to rely on it — and that gap is becoming operationally expensive.” Employees must still be involved, and the cost is both increased cognitive demands and also the emotional burdens and anxiety that AI is making mistakes and increasing risk.

Maximizing the Benefits of AI

Nothing I’ve said about AI’s impact on thoughtload suggests that using it is a poor choice. The puts and takes on employees’ thoughtload might be worth it if the benefits to using AI are significant. Those benefits depend on what measure of work you’re looking at: activities, outputs, or outcomes. Activities are the tasks you do: read a report, attend a meeting, monitor a Slack thread, write a line of code. Outputs are the finished products you create: a report, a presentation, a roadmap, a feature. Outcomes are the business results you’re chasing: growing user base, increased ARPU, lower turnover, reduced cyber penetration.

How are you thinking about the benefits of AI in your organization? Is your evaluation of the value of AI focused on activities, outputs, or outcomes?

If you focus on how many tasks agents are doing or how many tokens you’re using, that’s valuing activity. If you’re looking for improvements in features shipped or content pieces posted, that’s valuing output. If you’re measuring changes in customer adoption or increased inbound leads, that’s valuing outcomes. Where you put your attention will drive completely different behaviors and have distinct impacts on both business results and employees’ thoughtload.

Many organizations have tried to ramp up their use of AI by focusing on activity: asking people to document how they’re using chat, agents, or vibe coding, or even monitoring token use or call requests as a proxy for activity toward feature development. This might make sense if you’re trying to overcome fear and anxiety and encourage free play with the tools, but rewarding activity will likely increase busyness and trigger great fear of being displaced without adding much business value.

Another problem with focusing primarily on AI activity is that constant notifications and distractions, plus the relentless requirement to feed the tools with validation and iteration, can detract from employees’ ability to work in flow on their other tasks. Constant distraction is not only cognitively, but also emotionally taxing, and physically and mentally draining–a trifecta of terrible for increasing thoughtload.

If you need to focus on activity in the short-term to overcome anxiety and drive adoption, it’s probably alright. But monitor the impact on costs and thoughtload closely. As quickly as possible, shift away from the AI exposure therapy approach.

The next stage in your evolution is to focus on the extent to which AI is helping your people produce outputs more efficiently and effectively. This may be where AI contributes the most. How much more efficiently can marketing produce content? How many more features can engineering ship? How quickly can your engineers turn geosurvey data into a 3D map of a brownfield site?

Unfortunately, the one area where we can easily demonstrate AI is an asset might also be its greatest liability. The ability to generate more and more outputs with decreasing friction means we’re amping up production that isn’t always in service of better outcomes.

One area where this is negatively affecting people’s thoughtload is the proliferation of dense decks and weighty reports that people are expected to read, digest, and evaluate on the torrent of information they’re already trying to consume. This can be especially upsetting if one colleague uses AI to generate a lengthy report that oversteps into another person’s responsibility. While it’s easy to get carried away with questions about how your peers might do their jobs more effectively, it’s both intellectually taxing and emotionally triggering to have someone question your contribution and suggest that AI could do your job better than you.

Even production that seems valuable can be less helpful than you think. A software engineer I spoke with shared the story of his startup working twelve-hour days, six days a week to pump out new features, only to have a customer request that they only push security-related fixes. They were happy with the platform when they signed up, and their employees couldn’t afford the time or energy to keep up with constant changes in process or functionality. Countless hours of activity, impressive feature production, negative customer value.

Taken together, these ideas point to the importance of being deliberate about when and how to encourage employees to use AI. The tools are costly, both in hard dollars and in the cost to employees’ thoughtload. To increase the likelihood that AI use is both effective and sustainable, managers need to coach and support employees in how they use AI and discourage the production of outputs that fail to make a difference in the business outcomes that matter.

Taken together, these ideas point to the importance of being deliberate about when and how to encourage employees to use AI. The tools are costly, both in hard dollars and in the cost to employees’ thoughtload, so assessing the ROI is essential.

Remember those engineers pushing twelve-hour days to ship feature after feature? Their customer didn’t want more output. They wanted a working, secure platform that lowered their thoughtload, rather than raising it. Instead, it was countless hours of activity, impressive feature production, and negative customer value.

That’s the test worth applying before you greenlight the next AI initiative: not “how much are we using it” or even “how much are we producing with it,” but “will this move the number we actually care about”–clicks, ARR, engagement, bugs. If you can’t draw a line from the AI activity to that number, you’re paying the thoughtload cost without collecting the reward that justifies it.

To increase the likelihood that AI use is both effective and sustainable, managers need to coach and support employees on how they use AI, and discourage outputs that fail to move the business outcomes that matter.