Effects of AI tools on productivity
By Tuukka Tähtinen
November 12th, 2026
In the modern day of AI coding agents, AI-first approaches are becoming increasingly more and more common among web development agencies and software development companies. While you may still be solving problems when prompting an agent, you are no longer inherently required to come up with the specifics of the solution.
Just review the result and push.
But how does this actually affect overall productivity?
Effects on workers
AI agents doing the coding for you can reduce the mental workload and open you up to work on something else as the agent works. And while this does improve the speed of development, it has a less considered consequence: the work becomes less stimulating for the human mind. As a result, this less stimulating workload can at worst lead to a state very similar to burn out called bore out.
On the other hand, companies may instead excessively increase the workload as a result of increased productivity, which can lead to reduced code quality due to lack of time for proper review processes. For smaller, simpler tickets, this may not end up incurring problems, but for larger and more complex features, you’ll end up with worse compounding issues.
With increased work pace, workers will have less time to review AI code, leading to a worse understanding of the codebase by the workers and possibly buggy or bloated code. As these two problems begin to compound over time, you’ll eventually reach a breaking point: The code either breaks or can no longer be built on top of, or the expected level of productivity overworks your developers and they burn out.
The cliché answer here is that there needs to be a balance between under-stimulation and over-stimulation.
Now this is more of a project workflow problem, and Rome wasn’t built in a day, but steps should be taken to balance the workload.
To be a bit exaggeratory, someone might look at this and think
“Well, AI is making development accessible to a larger workforce, which gives us more opportunities for swapping out any burnt-out employees.”
I suppose you are right. You do have access to a larger pool of workers who are capable of operating an AI agent, but you are still going to need people who understand what’s going on under the hood, and you don’t want to operate a rotating door of employees.
These new employees might know what a jigsaw is, but someone needs to know how the pieces go into place.
Furthermore, the feeling of decreased job security brought on by AI exposure in the workplace has a link to reduced employee morale, leading to reduced performance due to worsened workplace wellbeing. A high level of emotional intelligence can help offset this feeling of lowered job security.
This study wasn’t specific to developers and encompasses multiple sectors, but you get the idea.
Effects on overall task productivity
Based on this study published by Harvard analyzing GitHub activity and Jira tickets from 718 firms, overall task productivity has seen an increase.
In this study, we can see that completed Jira tickets per month have seen a slight increase as AI agents have been adopted. Along with an increase in the amount of commits per employee, and size of pull requests as AI agents have been adopted for a longer period.
Based on anonymous accounts of engineers from the same study, the coding process can be, and often is, faster, but the saved work is often offloaded onto the review process instead. This is less of a problem when working with simpler changes, but as you end up with more complex, deeper, problems, the review work increases.
Based on this, there is a point when the task gets complex or large enough where using AI will push you back down to reduced productivity due to the lengthened code review step. But on the opposite end of the spectrum, simple code changes can be implemented quickly by using AI without taking up more time in code reviews.
There is a difference between using AI agents and AI assistants, however.
With AI agents, the study displays a steady increase in averages for time spent on code reviews and changes requested on each pull request.
AI assistant use instead remains about the same with a nonsignifacnt decrease in days from submission to merge, and comments per pull request.
Additionally, it has been reported that some sites are blocking AI agents from accessing portions of their content, which is going to negatively impact the capability of these AI agents to gather information. If this practice ends up spreading, the effect is going to increasingly prevent AI agents from gathering accurate, up-to-date information which will require more human intervention when reviewing the results.
The financial cost of AI tools
Token prices have recently gone down, but with the increased token use caused by quantity and complexity of tasks, the overall token costs would remain about the same or even increase in heavy use. Subscription plans on the other hand, have gone through restructuration and have had their prices increased. Add on top of that the increased quantity of available AI tool subscriptions across existing tools and services, and you’ll have increased your existing costs at a steady rate.
Accenture was recently reported to be limiting the monthly costs of AI use for their employees, we could infer that the cost of using AI at a large scale was not cost-effective relative to the productivity increase. The financial cost seemed to be quite vast, previously averaging from 500 to 1000 USD per employee each month, to now being limited to 29 USD per employee.
Considering that AI companies are currently running at a loss, these costs are likely to rise in the future.
We can already see this happening with GitHub Copilot moving to usage-based billing this past June, and Uber reportedly spending their entire year’s AI budget in 4 months.
With that said, whether AI use is efficient or not, is dynamic, difficult to universally gauge, and depends on use cases, pricing, worker skill and effects that were discussed earlier in this post. In the case of Uber, it was stated that the usage of AI tools accomplished exactly what they wanted, but at a hefty price tag.
How much are you willing to pay for the results?
Contact us
Get in touch and let's discuss your business case
Submitting this form will not sign you up for any marketing lists. Your information is strictly used and stored for contacting you. Privacy Policy
Related posts: