Explain why AI can do one task very well and a similar-looking task badly.
Farah is a marketing executive at a bubble tea chain in Singapore with eight outlets. For months she has used an AI assistant to draft social media captions, turn rough notes into briefs, and come up with twenty campaign names in a minute.
Then her manager asks which outlet should get the next promotion. Farah uploads the sales spreadsheet and a page of comments from the outlet managers, and asks the assistant to decide. The reply is confident and well reasoned, but it misses one line in the comments: the Bugis outlet's sales fell during three weeks of renovation. Because the assistant did not account for that, it picks the wrong outlet.
Farah would have rated the captions and the outlet decision as about equally hard. For the model they were very different jobs. The captions were inside what researchers call the jagged frontier, and the outlet decision was outside it.
In 2023, researchers from Harvard Business School and several other universities ran a field experiment with Boston Consulting Group (BCG) to test how AI affects professional work. The results were published under the title "Navigating the Jagged Technological Frontier".
Hundreds of BCG consultants worked on realistic consulting tasks. One group could use the most capable AI model of that time, and the rest worked without AI.
The first type of task was everyday creative and analytical consulting work: coming up with ideas for a new product, planning how to take it to market, and writing persuasively about it. On these tasks, consultants with AI completed more tasks, finished faster, and judges rated their output higher in quality.
The researchers chose the second type of task deliberately, because they expected it to sit just beyond what the AI does well. Consultants solved a business problem using spreadsheet data and interview notes, and the right answer depended on spotting details that the data alone did not make obvious. On this task, consultants using AI got the correct answer less often than those without it. The model gave plausible, polished recommendations that were wrong, and many consultants accepted them.
The same tool, used by similarly skilled people, improved work on one task and made it worse on another.
Much of this course has been about where the line falls and why. The jagged frontier is the researchers' name for the line between what AI does well and what it does badly. It is jagged because AI ability does not rise steadily as tasks get harder, so "good at easy work, poor at hard work" is too simple a description. Two tasks you would rate equally demanding can fall on opposite sides of the line. In the BCG study, writing a persuasive product pitch was well inside the line and reasoning through a messy business case was just outside it, yet both feel like skilled consulting work.
Work inside the frontier is work the AI handles well. Lessons 3.2 and 4.1 showed that drafting, summarising and rewording suit what a model learned from huge amounts of text, so these tasks tend to fall inside.
Work outside the frontier is just past the AI's ability, where its output is polished but wrong. Module 6 covered the tasks that tend to land there: ones that hinge on one specific detail, a precise figure, fresh facts, or careful reasoning over data. These are where invented answers and quiet misses happen. Farah's outlet decision hinged on one line about a renovation in a page of comments, which put it outside.
The AI's reply gives no signal that a task is past the edge. As Lesson 6.1 noted, it sounds just as confident when it is wrong as when it is right. In the BCG experiment the wrong answers were well written and persuasive, and consultants who had done well with AI on other tasks had little reason to question them.
The bigger danger is trust that carries over. Success on one kind of task builds trust, and you then apply that trust to a neighbouring task where the tool has not earned it. Farah's months of good captions and briefs are an example: they made it easy to hand the outlet decision to the assistant without checking its reasoning against the comments. The more useful an assistant has been to you, the easier this mistake is to make.
General claims cannot show you where the edge is in your own job. Vendor demos, colleagues' success stories and headlines about AI all describe other people's tasks, so set them aside as a guide to yours. The BCG study shows that the edge exists and roughly what sits on each side of it, but it cannot show where your own tasks fall.
The reliable method is to test the AI on your own work, using tasks where you already know the right answer or can check it. Lessons 5.4, 6.4 and 7.5 each gave you a test for doing this, and Module 8 builds on those three tests. Lesson 8.2 gives you questions for scoring any task, and Lesson 8.4 turns those scores into a map of your own job.
Start by looking for pairs. Somewhere in your week there are probably two tasks that look much alike, as Farah's did, where you suspect the assistant would do one well and stumble on the other. Spotting those pairs is the first step to seeing your own edge.
Write down two tasks from your job that look alike, where you suspect AI would do one well and the other badly. Say why you think so.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).