Does the frontier smooth as models improve?

Unknown, and it matters a great deal. The experiment used GPT-4 in 2023, so tasks outside the frontier then may sit inside it now. Whether the boundary becomes smoother and more predictable, or simply moves while staying jagged, has not been studied. This research lists it as an open question rather than guessing.

What is the jagged frontier?

Q: What is the jagged frontier?

The irregular boundary between tasks an AI system performs well and tasks it performs badly. The point is that it is jagged rather than smooth: tasks that look equally difficult to a person sit on opposite sides of it, competence on one task tells you little about an adjacent one, and nothing in the output signals which side you are on. The term was introduced by Fabrizio Dell'Acqua and colleagues in 2023.

Q: What did the jagged frontier experiment find?

A pre-registered field experiment with 758 BCG consultants. On 18 realistic tasks inside the frontier, consultants using GPT-4 completed 12.2 per cent more tasks, 25.1 per cent faster, at more than 40 per cent higher quality. On one task deliberately chosen to sit outside it, consultants using AI were 19 percentage points less likely to produce a correct solution than consultants with no AI at all. Same people, same tool, opposite results.

Q: What are centaurs and cyborgs?

Two patterns among successful users identified in the same study. Centaurs divide the work, delegating whole sub-tasks to the machine or keeping them. Cyborgs integrate continuously, moving back and forth within a single task. Both terms have travelled widely, sometimes detached from the evidence that produced them, and neither has been established as superior.

Where the term comes from, the 758-consultant experiment that produced it, why the frontier finding matters more than the productivity figure, and what remains unknown about whether it smooths.

Questions this page answers

All 780 questions this research covers

The jagged frontier is the boundary of what an AI system can do well, and the point is that it is jagged rather than smooth. Tasks that look equally difficult to a person sit on opposite sides of it. Competence on one task tells you very little about competence on an adjacent one, and nothing in the system's output tells you which side you are on.

Definition#

The jagged technological frontier: the irregular boundary between tasks an AI system performs well and tasks it performs badly, where the two categories can be almost indistinguishable in apparent difficulty, and where the system gives no signal of having crossed from one to the other.

Where the term comes from#

It was introduced in 2023 by Fabrizio Dell'Acqua and colleagues at Harvard Business School, Wharton and MIT, working with Boston Consulting Group, in a pre-registered field experiment with 758 consultants, roughly 7 per cent of BCG's individual-contributor consultants.

On 18 realistic consulting tasks inside the frontier, consultants using GPT-4 completed 12.2 per cent more tasks, 25.1 per cent faster, at more than 40 per cent higher quality. On one task deliberately selected to sit outside it, consultants using AI were 19 percentage points less likely to produce a correct solution than consultants with no AI at all.

Same people. Same tool. Same week. Opposite results, decided by which side of an invisible line the task happened to fall on.

Why it matters more than the productivity figure#

The productivity numbers from that study are quoted constantly. The frontier finding is quoted far less. It is the more useful half, because it explains why organisational results from AI are so inconsistent.

A smooth frontier would be manageable. You would learn roughly where competence ends and degrade your confidence gradually as tasks got harder. A jagged one is not manageable that way. You have to learn the shape empirically, task by task, in your own domain, and the map does not transfer to anyone else's.

This is also why fluency is such a poor guide. The system does not become hesitant when it crosses the line. It produces the same confident prose on both sides, so the consultants outside the frontier were not careless: they were reading output that gave them no signal.

Centaurs and cyborgs#

The same study identified two patterns among successful users. Centaurs divide the work, delegating whole sub-tasks to the machine or keeping them. Cyborgs integrate continuously, moving back and forth within a single task. Both terms have travelled, sometimes detached from the evidence that produced them, and neither is established as superior.

GPT-4 in 2023, and the frontier since#

The experiment used GPT-4 in 2023, so the specific tasks outside the frontier then may sit comfortably inside it now. Whether the boundary smooths as models improve, or simply moves while staying jagged, is unstudied. It matters a great deal, and this research lists it as unknown rather than guessing. The finding also comes from a working paper rather than a peer-reviewed journal, so it is graded Tier B.

Why "is AI good at this?" misses#

The frontier is the reason "is AI good at this?" is the wrong question. The useful question is "in what circumstances is this system likely to be wrong for the kind of work I do?", and answering it requires a map you build yourself over months. That map is one of the few genuinely non-commoditised assets available, because it is made from your own work and cannot be bought.

It also has an uncomfortable corollary. Building a frontier map requires enough expertise to recognise a wrong answer in the first place, which makes it available to experienced practitioners and largely unavailable to anyone early in a career. That is synthetic seniority stated as a measurement problem.

The practical version is how do I know when AI is wrong. On the tendency that makes the frontier dangerous, automation bias. On what it means for pairing humans and machines, human-AI collaboration. On what leaders should take from it, what AI literacy means for leaders.

Key sources

Dell'Acqua, F., McFowland III, E., Mollick, E., Lifshitz-Assaf, H., Kellogg, K., Rajendran, S., Krayer, L., Candelon, F. and Lakhani, K. (2023). Navigating the Jagged Technological Frontier. Harvard Business School Technology and Operations Management Unit Working Paper 24-013.
Yu, F. et al. (2024). Heterogeneity and predictors of the effects of AI assistance on radiologists. Nature Medicine, 30(3).

About this definition#

The jagged technological frontier belongs to Dell'Acqua and colleagues, not to SuperSkills. It appears here because the concept is referenced throughout this research and deserves an accurate account of its origin and its limits. Rahim Hirji is the author of SuperSkills (Kogan Page, 2026). On a 90-day review cycle, because model capability moves.

About this research#

Written by Rahim Hirji, author of SuperSkills (Kogan Page, 2026), keynote speaker on AI and human capability, and founder of The SuperSkills Intelligence Company.

How this research works · Reviewed quarterly · Found an error? Tell me and it is corrected on the page.

Cite this

Hirji, R. (2026). What is the jagged frontier? The SuperSkills Intelligence Company. Last reviewed 26 August 2026. thesuperskills.com/research/what-is-the-jagged-frontier

Box of Amazing

Rahim’s free weekly letter on AI and human capability

If this was useful, the weekly letter is where the thinking happens first. Most of what ends up on this site starts there. Weekly essays on AI, capability and the future of work. Read by 25,000 people, every week since 2017. Free, and one click to stop.

Opens Substack to confirm. No pitch in it, unsubscribe in one click, and nobody follows up because you read something.