ARTIFICIAL INTELLIGENCE

“The “jagged edge of AI” refers to the inconsistent and uneven nature of current artificial intelligence, where models excel at some complex tasks (like writing code) but fail surprisingly at simpler ones, creating unpredictable performance gaps that require human oversight.” – Jagged Edge of AI

The “jagged edge” or “jagged frontier of AI” is the uneven boundary of current AI capability, where systems are superhuman at some tasks and surprisingly poor at others of seemingly similar difficulty, producing erratic performance that cannot yet replace human judgement and requires careful oversight.^4,7

At this jagged edge, AI models can:

• Excel at tasks like reading, coding, structured writing, or exam-style reasoning, often matching or exceeding expert-level performance.^1,2,7
• Fail unpredictably on tasks that appear simpler to humans, especially when they demand robust memory, context tracking, strict rule-following, or real-world common sense.^1,2,4

This mismatch has several defining characteristics:

• Jagged capability profile
  AI capability does not rise smoothly; instead, it forms a “wall with towers and recesses” – very strong in some directions (e.g. maths, classification, text generation), very weak in others (e.g. persistent memory, reliable adherence to constraints, nuanced social judgement).^2,3,4
  Researchers label this pattern the “jagged technological frontier”: some tasks are easily done by AI, while others, though seemingly similar in difficulty, lie outside its capability.^4,7

• Sensitivity to small changes
  Performance can swing dramatically with minor changes in task phrasing, constraints, or context.⁴
  A model that handles one prompt flawlessly may fail when the instructions are reordered or slightly reworded, which makes behaviour hard to predict without systematic testing.

• Bottlenecks and “reverse salients”
  The jagged shape creates bottlenecks: single weak spots (such as memory or long-horizon planning) that limit what AI can reliably automate, even when its raw intelligence looks impressive.²
  When labs solve one such bottleneck – a reverse salient – overall capability can suddenly lurch forward, reshaping the frontier while leaving new jagged edges elsewhere.²

• Implications for work and organisation design
  Because capability is jagged, AI tends not to uniformly improve or replace jobs; instead it supercharges some tasks and underperforms on others, even within the same role.^6,7
  Field experiments with consultants show large productivity and quality gains on tasks inside the frontier, but far less help – or even harm – on tasks outside it.⁷
  This means roles evolve towards managing and orchestrating AI across these edges: humans handle judgement, context, and exception cases, while AI accelerates pattern-heavy, structured work.^2,4,6

• Need for human oversight and “AI literacy”
  Because the frontier is jagged and shifting, users must continuously probe and map where AI is trustworthy and where it is brittle.^4,8
  Effective use therefore requires AI literacy: knowing when to delegate, when to double-check, and how to structure workflows so that human review covers the weak edges while AI handles its “sweet spot” tasks.^4,6,8

In strategic and governance terms, the jagged edge of AI is the moving boundary where:

• AI is powerful enough to transform tasks and workflows,
• but uneven and unpredictable enough that unqualified automation is risky,
• creating a premium on hybrid human–AI systems, robust guardrails, and continuous testing.^1,2,4

Strategy theorist: Ethan Mollick and the “Jagged Frontier”

The strategist most closely associated with the jagged edge/frontier of AI in practice and management thinking is Ethan Mollick, whose work has been pivotal in defining how organisations should navigate this uneven capability landscape.^2,3,4,7

Relationship to the concept

• The phrase “jagged technological frontier” originates in a field experiment by Dell’Acqua, Mollick, Ransbotham and colleagues, which analysed how generative AI affects the work of professional consultants.^4,7
• In that paper, they showed empirically that AI dramatically boosts performance on some realistic tasks while offering little benefit or even degrading performance on others, despite similar apparent difficulty – and they coined the term to capture that boundary.⁷
• Mollick then popularised and extended the idea in widely read essays such as “Centaurs and Cyborgs on the Jagged Frontier” and later pieces on the shape of AI, jaggedness, bottlenecks, and salients, bringing the concept into mainstream management and strategy discourse.^2,3,4

In his writing and teaching, Mollick uses the “jagged frontier” to:

• Argue that jobs are not simply automated away; instead, they are recomposed into tasks that AI does, tasks that humans retain, and tasks where human–AI collaboration is superior.^2,3
• Introduce the metaphors of “centaurs” (humans and AI dividing tasks) and “cyborgs” (tightly integrated human–AI workflows) as strategies for operating on this frontier.³
• Emphasise that the jagged shape creates both opportunities (rapid acceleration of some activities) and constraints (persistent need for human oversight and design), which leaders must explicitly map and manage.^2,3,4

In this sense, Mollick functions as a strategy theorist of the jagged edge: he connects the underlying technical phenomenon (uneven capability) with organisational design, skills, and competitive advantage, offering a practical framework for firms deciding where and how to deploy AI.

Biography and relevance to AI strategy

• Academic role
  Ethan Mollick is an Associate Professor of Management at the Wharton School of the University of Pennsylvania, specialising in entrepreneurship, innovation, and the impact of new technologies on work and organisations.⁷
  His early research focused on start-ups, crowdfunding and innovation processes, before shifting towards generative AI and its effects on knowledge work, where he now runs some of the most cited field experiments.

• Research on AI and work
  Mollick has co-authored multiple studies examining how generative AI changes productivity, quality and inequality in real jobs.
  In the “Navigating the Jagged Technological Frontier” experiment, his team placed consultants in realistic tasks with and without AI and showed that:

• For tasks inside AI’s frontier, consultants using AI were more productive (12.2% more tasks, 25.1% faster) and produced over 40% higher quality output.⁷

• For tasks outside the frontier, the benefits were weaker or absent, highlighting the risk of over-reliance where AI is brittle.⁷
  This empirical demonstration is central to the modern understanding of the jagged edge as a strategic boundary rather than a purely technical curiosity.

• Public intellectual and practitioner bridge
  Through his “One Useful Thing” publication and executive teaching, Mollick translates these findings into actionable guidance for leaders, including:

• How to design workflows that align with AI’s jagged profile,

• How to structure human–AI collaboration modes, and

• How to build organisational capabilities (training, policies, experimentation) to keep pace as the frontier moves.^2,3,4

• Strategic perspective
  Mollick frames the jagged frontier as a continuously shifting strategic landscape:

• Companies that map and exploit the protruding “towers” of AI strength can gain significant productivity and innovation advantages.

• Those that ignore or misread the “recesses” – the weak edges – risk compliance failures, reputational harm, or operational fragility when they automate tasks that still require human judgement.^2,4,7

For organisations grappling with the jagged edge of AI, Mollick’s work offers a coherent strategy lens: treat AI not as a monolithic capability but as a jagged, moving frontier; build hybrid systems that respect its limits; and invest in human skills and structures that can adapt as that edge advances and reshapes.