What it is
Split your AI work into two tracks:
- The automation track. Repeatable, digital tasks with a clear input, a clear output and a simple check. AI does the task. A person handles the exceptions.
- The expert-help track. Expert work that depends on judgement, responsibility or being on site. AI prepares an analysis, a draft or a set of options. The expert decides.
Not all AI work is the same kind of project. The two tracks have different goals, different risks and different measures. Run them side by side, and do not judge one by the other’s standard.
What the evidence shows
The problems this practice deals with have been studied:
- AI helps on some tasks and harms on others. In one experiment, consultants using AI did better on 18 tasks AI handles well and worse on one task AI does badly.
- The task type decides. Across 106 experiments, a person working with AI did worse on decision tasks and better on tasks that involve creating content.
- Few jobs can be automated whole. A study that scored 18,156 tasks found that most occupations hold some tasks a machine could learn, and few could be fully automated.
- Assisting is the common case. The International Labour Organization’s estimate predicts that generative AI will mostly assist people in their jobs. The US Census Bureau asked more than 117,000 firms: 66% of firms using AI used it only to help staff with tasks, and 2% of firms cut jobs because of AI.
- Workers most often want a partnership. In a Stanford study, 1,500 US workers rated 844 tasks from their own occupations. Equal partnership between the person and the AI was the level they wanted most often. The researchers then matched start-ups funded by the investor Y Combinator to the tasks those start-ups build AI for. 41.0% of the matches fell on tasks that workers did not want automated or rated a low priority. The study has not been peer-reviewed.
How to do it
- Map the tasks in the roles you want to change. See map tasks before deciding what to automate.
- Sort the tasks. Tasks to automate are digital, repeatable and easy to check. Tasks for expert help need a person’s judgement or presence.
- Start the automation track with two or three tasks. Take a baseline first.
- Start the expert-help track with one expert group. Ask the experts what slows them down.
- Give each track an owner and its own measures.
- Move a task from one track to the other when the facts change. A task where AI helps an expert today may be safe to automate once its checks are routine.
What to measure
- Automation track: time from start to finish, number of exceptions, corrections needed.
- Expert-help track: quality of the expert’s decision, rework, time to diagnose or to finish. Not how much the AI did alone.
Where it fails
- Measuring assistants by headcount saved. The gain from an assistant is a better or faster decision. Counting posts removed misses that gain and alarms the experts you need.
- Automating expert work because the technology can do part of it. The part AI cannot see is often the part that matters.
- Only the automation track. Automation is easier to measure, so it gets the budget. The larger gain may sit with the experts.
- Forgetting the learning path. If juniors learned on the tasks you automate, plan how they will learn now. See keep the entry-level tasks that newcomers learn on.
By company size
- Medium: two short lists and one owner for both are enough.
- Mid-market and enterprise: separate owners. The automation track sits close to operations and IT. The expert-help track sits with the heads of the expert functions.
See it applied
Sources
- Bonney, Breaux, Dinlersoz, Foster, Haltiwanger and Pande (2026). The Microstructure of AI Diffusion: Evidence from Firms, Business Functions, and Worker Tasks. US Census Bureau, Center for Economic Studies Working Paper CES-WP-26-25 (April 2026). What this study did and found
- Shao, Zope, Jiang, Pei, Nguyen, Brynjolfsson and Yang (2026). Future of Work with AI Agents: Auditing Automation and Augmentation Potential across the U.S. Workforce. arXiv 2506.06576. Not peer-reviewed. arxiv.org
Claims on this topic, checked against these studies:
- “66% of small businesses use AI” Needs correcting
Related practices
- For each kind of task, test whether a person with AI beats either one alone Do not assume that adding a person to AI, or AI to a person, gives the best result. Test it for each type of task before you settle the process. Do next · Effort: medium (weeks, a few people)
- Map tasks before deciding what to automate Break jobs into tasks and look at each one. Published scores rate how much of a job AI could do, its “exposure”. Treat a high score as a place to look, not as a decision to automate. Do first · Effort: low (days, one person)
- Start with one costly process, not a company-wide rollout Pick one process that visibly costs time or money. Apply AI there, with a person checking the output, and expand only when the numbers move. Do first · Effort: medium (weeks, a few people)
- Budget for the work AI creates AI removes some work and adds other work: checking its output, keeping records, handling exceptions and looking after the tools. Count that work and give it an owner before you count the saving. Do first · Effort: low (days, one person)