AI chatbots displace tasks, not jobs: the task-based model of automation

Most companies that implement AI technology to automate customer service do not entirely replace the entire workforce. Instead, AI chatbots automate the repetitive tasks that customer service agents are expected to do so that they can spend more time assisting with more complicated tasks and problems that require human intervention and intuition. For example, one Fortune 500 company implemented a Generative AI assistant to support their agents and reported a 14% increase in productivity of their customer support team, and a 34% increase in productivity of their most inexperienced customer support representatives.

Automation focuses on tasks, not entire occupations

The economic theory of Task-Based Automation was initially proposed by economists Daron Acemoglu and Pascual Restrepo. Task-Based Automation considers that every job is made up of bundles of tasks, where machines or robots can take over some of the tasks, but new tasks will still be needed, which is called the reinstatement effect. A customer service role is an excellent example of this model. A customer service role includes answering questions such as "Where's my order?" which AI technology is designed to handle. However, calming an irate customer, interpreting an unclear complaint or persuading a customer to buy something are tasks that require the use of human empathy and common sense.

AI chatbots will allow companies to eliminate many of the repetitive customer service tasks currently performed by humans and will therefore shift the responsibilities of customer service workers from answering repetitive questions to handling more complex customer concerns. If jobs were measured by the number of tasks performed instead of the number of people doing the job, the question of "Will AI take my job?" would change to "What tasks will AI take over and what tasks remain for the customer service representative to perform?" Whether this leads to structural unemployment depends on which effect (displacement or reinstatement) dominates. The displacement effect results in decreased wages for those workers whose jobs are mostly routine. According to Acemoglu and Restrepo, the employment-to-population ratio decreased by 0.2 percentage points for every additional industrial robot per 1000 workers in a given area between 1990-2007; similarly, the average hourly wage decreased by approximately 0.42%. If we extend this same logic to conversational AI, then once a bot is able to perform a variety of tasks (including password resets and delivery queries) at very low marginal cost, companies will be able to have fewer employees perform those same functions. Thus, the type of unemployment that Keynes originally discussed will become concentrated in the most "scripted" positions. Additionally, the impact of the automation will not necessarily be felt uniformly: automation will displace rules-based jobs first, while jobs requiring more judgement/analysis (e.g., complex, managerial tasks) will be less affected. It's also important to remember that in terms of the overall impact of automation on wages, the overall wage impact was not substantial, which is why there are continuing discussions on what jobs should be created to accommodate the displaced workers.

There will be many new tasks created due to the reinstatement effect of automation

Displacement is only half of the overall model. The second half of the task-based model is that reinstatement creates new tasks in which humans can continue to work, and because labour demand is a derived demand, it will shift toward these new tasks as opposed to disappearing entirely. Deploying a bot also creates new roles for the design, flow mapping, escalation handling and model management of chatbots, for example. Companies such as Innowise are developing this type of system, and for each deployment, a number of roles are created that were not available a decade ago. This process is an example of creative destruction occurring within a single job description, but it is also creating opportunities for employees to climb the human capital ladder, starting from answering tickets and moving toward training the bot that will ultimately answer the tickets. This pattern has similar characteristics to previous waves of office automation, where spreadsheets provided a means of performing the manual work of tabulation but also created an increased demand for employees with the analytical capacity to utilise spreadsheets effectively.

Chatbots create value for agents more often than they take away value from agents

The current evidence supports the premise that chatbots create greater value for agents by helping them perform their job functions better. For example, using data from Brynjolfsson, Li and Raymond's (2025) Quarterly Journal of Economics publication, we find that after deploying a Generative AI assistant, support agents were able to resolve 14% (and as high as 34% for novice agents) more issues per hour than they did prior to using chatbots. Further, according to the Economic Index published by Anthropic, by the beginning of 2026, roughly 52% of all Claude.ai conversations were being used for augmentation, while only 45% were directly attributable to full automation. Most importantly, agents remained responsible for each conversation — the chatbot model makes suggestions, while it is up to the agent on whether or not to utilise those suggestions. In essence, the tools are increasing productivity through augmentation but have yet to substantially replace agents, thereby supporting the reinstatement effect of continuing to use chatbots as tools that augment human judgement.

On balance, chatbots are still generating supervisory and design work. If this pattern continues, it will shape the future of customer service agents and build upon the basic premise of the task-based model. The customer service agent continues to be rebuilt a task at a time rather than written off completely.