Imagine Einstein works at your company.
One week, he writes a research paper that gets quoted in every industry conference for the next decade. The following week, he forgets to send a contract to a client, and you lose the deal.
Would you fire him? Would you keep him? How would you manage him? How about your team? Do you still need them now that you have Einstein?
AI today is starting to look a lot like that Einstein.
There is a name for this. In July 2024, Andrej Karpathy, a founding member of OpenAI, coined the term “jagged intelligence” to describe the strange, uneven nature of modern AI. State-of-the-art models, he wrote, can solve complex math problems while simultaneously failing at tasks a child could do. At the World Economic Forum in Davos this past January, Demis Hassabis, the CEO of Google DeepMind, picked up the term. AI today, he said, has “jagged edges.” It is, in Karpathy’s words, “a genius polymath and a confused grade schooler” at the same time.
The evidence is striking. In a landmark study by Harvard Business School and Boston Consulting Group, published in Organization Science this March, 758 consultants were given 18 realistic business tasks, some with access to GPT-4 and some without. On tasks that fell inside AI’s capabilities, the AI-assisted consultants produced work rated significantly higher in quality and finished noticeably faster. But on tasks just outside that zone, the same consultants were 19 percentage points more likely to get the answer wrong. Here is the part that should make every manager pause. The AI sounded just as confident either way. And the humans could not tell the difference. The brilliance and the failure looked identical from the outside.
This is not the future. It is the present. PwC reports that daily generative AI use among Thai workers jumped from 17% to 24% in a single year. We are not waiting for AI to enter the workplace. It is already here. The question is no longer whether to use it, but how to work alongside something that is, on any given task, either brilliant or quietly wrong.
The honest answer is that no one is fully sure yet. But a pattern is emerging. The organizations getting the most out of AI are not the ones with access to the smartest model. They are the ones doing harder, less visible work. They are designing workflows that account for AI’s jagged edges. They are investing in the right infrastructure, the right data, the right talent, and the right governance to turn raw model capability into real outcomes. Managing jaggedness is part of the answer, but only part. Getting AI right is an organizational project, not a model selection problem.
At SCBX, this is the question we keep coming back to. As we move toward becoming an AI-first organization, we believe the advantage will not come from chasing the smartest model. It will come from designing workflows around AI’s jagged edges, and building the human judgment that catches the moments it confidently gets things wrong. This is one of the central themes of the SCBX AI Outlook 2026: The Age of Abundant Intelligence, which explores what we call the harness, meaning everything an organization builds around AI to turn raw capability into reliable outcomes. In a world where almost every organization will soon have access to the same frontier models, the harness is becoming the real competitive edge.
Back to Einstein. You wouldn’t fire your team because you hired him. You’d redesign the work around him. You’d build a company that could hold his brilliance without breaking under his blind spots.
AI is asking us to do the same. Not less work, but smarter work. Not fewer humans, but better-placed humans. Not a race to replace ourselves, but a careful job of designing systems strong enough to carry intelligence that, for now, cannot carry itself.
Read the SCBX AI Outlook 2026 via LINE OA: https://lin.ee/8dvXKVs
Writer:
