Discover Your Professional AI Work Style:
Cyborg, Centaur, or Self-Automator?

What Actually Drives AI Performance? 

Executive Summary 

The new Harvard-led research suggests that professionals tend to fall into three working styles with AI.

Cyborgs collaborate deeply with AI and develop AI-related skills.

Centaurs use AI selectively while keeping strong human control and produce the most accurate work.

Self-Automators hand over much of the task to AI and gain speed, but they lag in innovative solutions and professional growth (Randazzo et al., 2025; Walsh, 2026).

Although there are three work styles, each of us has a dominant AI work style. But your dominant style is not necessarily the best style for all types of AI-augmented work. 

Most of us change styles depending on the type of work, the workflow and the business situation. 

Being able to consciously change work styles to fit the work at hand can be a powerful insight to building your professional skills.

The Risk of Not Adopting AI in Professional Services 

Key Findings From the Research

Which Are You? Cyborg, Centaur, or Self-Automator 

Why Centaurs Often Produce the Best Decisions 

Centaurs used AI in a controlled way. They asked targeted questions, used AI to gather examples, clarify methods, or refine language, but they kept ownership of the analytical path.  

In the study, the Centaur AI work style produced the highest accuracy in the final recommendation. For professional services, that is a critical point because the most valuable work is usually not just polished writing. It also gets the judgment right (Randazzo et al., 2025; Walsh, 2026). 

Centaurs may be best for professional services 

In practical terms, Centaurs behave like disciplined senior professionals. They use AI to accelerate learning and remove friction, but they do not surrender the logic of the work.  

This makes their workflow especially useful for high-stakes activities such as strategic recommendations, client diagnosis, positioning decisions, proposal strategy, and investment or growth recommendations. 

Why Cyborgs Are Powerful But Need Guardrails

Cyborgs collaborated fluidly with AI across multiple stages of work. They probed, challenged, refined, and re-ran the system repeatedly. This made them strong on persuasive output and helped them develop AI-specific skills such as prompting, testing, and pushing back. The upside is clear: Cyborg behavior can dramatically increase output and build new capability in how to work with AI (Randazzo et al., 2025). 

Cyborgs can increase output and build new capabilities. 

But Cyborgs were not automatically the most accurate. The research suggests that close collaboration with AI can still pull people toward AI's confident but wrong logic. For firms, this means Cyborg workflows are powerful, but they need checkpoints, validation steps, and a human decision owner for high-value work (Walsh, 2026; Candelon et al., 2026). 

Recent research, as of March/April 2026, is showing that AI can be extremely persuasive in pulling even professionals toward it's recommendations, even when they are wrong.

Cyborg’s need to be aware that the AI can “persuade” them into accepting a wrong or sub-optimal answer. 

Why Self-Automators Can Be A Hidden Risk

Self-Automators produced fast and polished output by offloading most or all of the work to AI. That can look efficient on the surface. In reality, it creates two dangers.  

First, quality can fall because the user is no longer testing the reasoning deeply.  

Second, learning slows because the professional is no longer building either domain expertise or AI expertise. The paper describes this as 'no skilling' (Randazzo et al., 2025). 

Automators produce fast results, but it is more often
of low quality and they don’t develop their own expertise. 

The self-automation work style can be especially dangerous in consulting, finance, accounting, strategy, and other advisory work where expertise is the product. If professionals become dependent on AI to think for them, the firm may gain short-term speed while weakening the judgment, credibility, and training pipeline it depends on for long-term advantage (Candelon et al., 2026; Walsh, 2026). 

Who is Your 'Human in the Loop'? 

The difference between AI work styles is important for more than just individual work. Many firms assume they can have any proficient 'human in the loop' at critical junctures in a workflow. Research shows the type of human in the loop is critical to success.

A human can be in the loop of AI or agentic automation - but their style may not be producing the best results or validating the system.

A person can be technically in the loop while quietly giving away the most important parts of thinking, judgment, and learning. That is why workflow structure and role clarity matter so much in professional services (Candelon et al., 2026; Randazzo et al., 2025).

Action Items: Move to the Best AI Work Style 

You don’t have to jump overnight from basic AI user to expert AI user.

The best move is to climb one level at a time. Research suggests that higher-value AI use is built through deliberate practice, not passive dependence.

Learn AI in workshops using “real-world” workflows
that match your real-world work.

Avoid AI keystroke and feature training.
This type of training has low retention and lacks expertise.

For over 100 years, research into learning has shown that workshops using real-work situations have twice the retention and impact of generic keystroke/feature training.

Take AI workshops where you learn using
real-world work examples and context.

Below are practical moves that help consultants and professionals improve their AI-collaboration style while protecting judgment, accuracy, and building your professional skills.

• Audit one important recurring task and take note of where human judgment is required. Keep those points under human control.
• Break large tasks into workflow stages such as framing, research, analysis, recommendation, review, and packaging. Identify ‘human in the loop’ points.
• Use AI selectively for expansion, summarization, example-finding, refinement, and formatting before trusting it with final conclusions.
• Add a validation step at the end of any AI-assisted process. Require that you or your team challenge at least one major AI recommendation before accepting it. Learn how to create prompts and agents that cross-check AI's assumptions and validate results.
• Track both your productivity and learning. Measure the time you saved in the workflow, but also ask yourself what professional skill AI is helping you build.

References