Robots can do plenty of tasks. Making them economical is another challenge
A new Anthropic study separates what robots can do from what makes financial sense. Its findings need careful reading, especially when discussing entire jobs.
The short answer
A robot can demonstrate a task without making it economical in everyday work. An Anthropic study published on September 30 draws that distinction. It estimates that robots can perform 74% of physical work in the US under particular conditions, while being cost-competitive for only 0.3% of the total work examined. These figures have different denominators. Neither is a count of jobs that have already disappeared.
A task description is a starting point
The research uses O*NET, the publicly available US database describing occupations, work requirements and individual tasks. Breaking a job into steps can make an evaluation easier to inspect. A generic task description still cannot capture every detail of your workplace. “Moving parcels”, for example, leaves important questions unanswered: are the boxes identical, where do they arrive, and how often does someone have to step in? Those details belong in a local assessment.
Look beyond the demonstrated action
Our view: the handoffs deserve attention. Who prepares the materials? What happens to a damaged box? Who handles an unusual request? A useful trial deliberately includes those situations. Otherwise, it risks measuring only the easiest part of a working day. The people who know the process should be able to record what they observe and question the result, rather than being asked to accept a single headline success rate.
A practical way to frame a small trial
Our suggestion is to describe one narrow workflow and record how frequently it occurs. In an approved test environment, observe both normal runs and exceptions defined in advance. Track completed tasks alongside interruptions, rework and time spent supervising the system. Discuss the findings with the staff who understand the workflow. This is an editorial evaluation approach, not a robotics test performed by Findbest, and it does not promise a profitable deployment.
Announcement date: 30 Sept 2026
What changed?
The announcement is a new robotics study separating capability from economic competitiveness. It offers a way to frame a discussion about automation, rather than a universal timetable for replacing complete occupations.
Who is this relevant to?
The distinction is useful for organisations considering a robotics pilot. For employees, it supports questions about specific tasks, oversight and changes to daily work. A headline percentage is not enough to describe what happens in a particular workplace.
What this means for your work
Our assessment: define the success criteria before a trial and consider the complete workflow. Preparation, downtime and rework should all be visible in the evaluation. A successful demonstration of one action cannot settle whether a project is useful in practice.
Limits of this report
This is US-focused research using AI-assisted assessments, not a field test in every workplace. We expanded the news window to seven days for this topic. Original sources were checked on October 4, 2026.
Read next
AI research: from answer to reliable source
A compact workflow for checking claims transparently.
Introduce AI to a small team: run a pilot before adding subscriptions
Use real tasks, a simple evaluation sheet and a clear decision to find out whether an AI tool helps your team.
Check AI answers: turn source links into supported claims
A source link is not enough. Check what it actually supports and whether the context matches.
Was this article helpful?
Your feedback helps us improve our articles.
Votes are counted per language version. Repeat voting is limited; totals do not necessarily represent distinct people. Privacy