Do the ways consultants split tasks with AI — or hand them over entirely — also describe how other kinds of work get done?

This explores whether the three ways consultants were found to work with AI also show up outside consulting. I'm reading the taxonomy as the BCG study's centaurs (people who split tasks between themselves and the AI), cyborgs (people who weave AI into every step) and self-automators (people who hand the work over). One caveat comes first: the note that defines the three modes wasn't in these results, so none of the material below tests the taxonomy directly in other fields. What the results do show is the conditions each mode depends on, and those conditions shift a lot from one kind of work to another.

Self-automation looks weakest in strategy work, the field closest to consulting. When six LLMs ran 15,000 strategy simulations, they recommended the same side of every strategic tension. Changing the industry context moved their answers less than simply reordering the options did Do LLMs consistently favor the same strategic choices regardless of context?. Newer frontier models also scored below older models and MBA students on a strategy simulation, because they kept choosing short-term profit over growth investments Do newer frontier LLMs actually make better strategic decisions?. So handing strategic judgment to the model means inheriting its built-in leanings. The cyborg mode has its own catch. In a 348-person experiment, working closely with an LLM led people to consider more factors but didn't make their forecasts more accurate. It also left them feeling more overloaded and less sense that the decisions were their own Does using LLMs actually improve strategic decision making?. Blending changes how people think, but that doesn't mean they think better.

In evaluation and review work, the picture changes. Single LLM rankings of business models were inconsistent, but aggregating many of them produced rankings close to expert judgment Can aggregated LLM rankings match human expert judgments?. A UK government tool that sorts public consultation responses into themes disagreed with expert reviewers about as much as two reviewers disagreed with each other Does AI theme-mapping perform as well as human reviewers?. A pipeline that breaks a research paper's novelty check into steps (extract the claims, find related work, compare) reached 86% alignment with human peer reviewers Can structured pipelines make LLM novelty assessment reliable?. These look less like one person picking a mode and more like a fourth arrangement: the work is redesigned so AI handles the high-volume steps and humans check the results.

Design work suggests a different extension again. Designers using Canvil, a prompt-authoring widget inside Figma, didn't split tasks with the model or merge with it. They shaped the model's behavior directly, treating it as a material to work with rather than a coworker Can designers shape LLM behavior without deep technical knowledge?.

Here's what you might not expect to learn. The three modes probably describe how a person relates to a single task. Whether a mode works depends on the work itself: whether the model's defaults happen to suit the problem, whether outputs can be checked against a standard or against other reviewers, and whether a person can reshape the model's behavior. The taxonomy may well travel to other fields, but these sources suggest it would need new modes once it does. To test that properly, find the source study on consultants and compare it against these other kinds of work.