Jagged Frontier: How AI Shapes the Future of Work (Centaur vs. Cyborg)
Jun 20, 2025
Discover how GPT-4 reshapes consulting work, when AI helps most, and whether your team should be a Centaur or a Cyborg.
As the world of work shifts under the weight of AI, we need to be future-focused; not frightened. I recently read an excellent paper, Navigating the Jagged Technological Frontier (co-authored by researchers from Harvard, Wharton, Warwick, MIT and BCG). It’s a crisp, evidence-driven look at how AI changes what we do well and where humans still matter most. It sheds light on how AI can revolutionize our approach to tasks, highlighting its potential and limitations.
What the study found (quick TL;DR)
When BCG consultants used GPT-4 for tasks requiring creativity, writing, and analysis, their outputs were often higher quality than those who didn’t use AI.
For complex, high-context problems; e.g., synthesizing insights from executives or tasks demanding deep domain intuition; humans without GPT-4 sometimes performed better.
GPT-4 produced the largest improvements for consultants who scored lower on a baseline assessment: AI acted as an equalizer.
The paper identifies two human–AI collaboration styles: Centaur and Cyborg ; both valid, but different in how they mix responsibility and agency between human and machine.
Unpacking the insights
AI as a boost - especially for the bottom half
It was observed that GPT-4 delivered a bigger boost in quality and efficiency to those consultants who scored lower on the initial assessment task. This suggests that AI serves as an equalizer, enhancing our skills in areas where we may lack strength. The study shows that the biggest relative improvement came from those who began with lower baseline performance. 𝐓𝐡𝐢𝐧𝐤 𝐨𝐟 GPT-4 𝐥𝐢𝐤𝐞 𝐚 𝐜𝐚𝐥𝐜𝐮𝐥𝐚𝐭𝐨𝐫: 𝐢𝐭 𝐡𝐞𝐥𝐩𝐬 𝐭𝐡𝐨𝐬𝐞 𝐧𝐨𝐭 𝐠𝐫𝐞𝐚𝐭 𝐚𝐭 𝐦𝐞𝐧𝐭𝐚𝐥 𝐦𝐚𝐭𝐡 𝐦𝐨𝐫𝐞 𝐭𝐡𝐚𝐧 𝐦𝐚𝐭𝐡 𝐰𝐡𝐢𝐳𝐳𝐞𝐬. That’s important for teams: AI doesn’t only augment the top performers , it helps raise the baseline.
Where humans still win
But it’s not an across-the-board win. Tasks that require deep domain knowledge, judgment, and tacit intuition ; for example synthesizing conversations with senior executives or navigating ambiguous tradeoffs still favor humans. AI can help, but it’s not a substitute for experience and contextual judgement.
Two collaboration archetypes: Centaur vs. Cyborg
𝐓𝐡𝐞 𝐬𝐭𝐮𝐝𝐲 𝐢𝐝𝐞𝐧𝐭𝐢𝐟𝐢𝐞𝐝 𝐭𝐰𝐨 𝐝𝐢𝐬𝐭𝐢𝐧𝐜𝐭 𝐚𝐩𝐩𝐫𝐨𝐚𝐜𝐡𝐞𝐬 𝐭𝐨 𝐡𝐮𝐦𝐚𝐧-𝐀𝐈 𝐜𝐨𝐥𝐥𝐚𝐛𝐨𝐫𝐚𝐭𝐢𝐨𝐧: 𝐭𝐡𝐞 𝐂𝐞𝐧𝐭𝐚𝐮𝐫 𝐚𝐧𝐝 𝐭𝐡𝐞 𝐂𝐲𝐛𝐨𝐫𝐠. The Centaur, like the mythical creature with human and horse combined clearly divided tasks between themselves and the AI, leveraging AI for it's strengths. On the other hand, other consultants operated like the Cyborg: a human seamlessly integrated with technology, these consultants blended AI into every step. They guided it with prompts, even giving it their own "consultant" personality to refine its output.
Centaur : a clear division of labor. The human identifies the problem and decides what matters; AI is used as a tool for specific strengths (drafting, ideation, formatting). Think human-led, AI-assisted.
Cyborg : a seamless integration where humans and AI are blended into a single workflow. The human continuously prompts, tunes, and shapes the AI; the AI becomes part of the person’s process and voice. Think integrated, co-created output.
Both approaches can be effective. Centaur workflows are easier to audit and control; Cyborg workflows can be more fluid and creative but require more discipline and guardrails.
Practical implications (for teams and leaders)
Use AI to raise the floor: prioritize tools that help people below the top tier get better faster.
Reserve pure human judgment for problems that demand deep domain expertise or political/social nuance.
Decide deliberately whether your team will operate as Centaur (clear handoffs) or Cyborg (tightly integrated). Train people in the style you choose.
Focus on governance: data quality, prompt design, and review processes matter more as AI is woven into workflows.
Conclusion
AI is not a binary replacement, it’s a reshaper. The Jagged Frontier paper shows AI can substantially boost output, especially for those starting from a lower baseline, but human judgement still wins where domain intuition and synthesis matter. The smarter question isn’t “Will AI replace us?” but “How will we design our teams - as Centaurs or Cyborgs - to get the most out of it?”
Which would you pick for your team: a Centaur approach with clear roles, or a Cyborg approach that blends AI into the workflow? Tell me which and why.