In 2023, Harvard Business School and BCG ran one of the most rigorous field experiments on generative AI at work. Dell'Acqua and colleagues gave 758 consultants realistic tasks and random access to GPT-4.
On tasks inside the model's capability, consultants with AI completed 12.2% more tasks. They worked over 25% faster and produced work rated more than 40% higher in quality.
On one task just outside its capability, the picture reversed. Consultants using AI were 19 percentage points less likely to reach the correct answer.
The frontier is real and hard to see
The researchers called this boundary the jagged technological frontier. Some tasks fall inside it. Others, often similar on the surface, fall outside it.
You cannot tell from the task which side it sits on. The people harmed by AI tended to accept its output with too little scrutiny.
The frontier keeps moving
Stanford's AI Index 2026 reports that AI agents' success on a benchmark of real-world tasks rose from 20% to 77.3% in a year. A task outside the frontier in January may be inside it by autumn.
The frontier is still jagged. In Ju and Aral's 2026 experiment, human-AI teams produced better ad text but worse images than human pairs.
Experience helps people work the frontier. Anthropic's Economic Index report of March 2026 found experienced users attempted higher-value tasks and succeeded more often.
Build this capability in your organisation
Growth Performance designs bespoke programmes that turn insight into measurable behaviour change.
Levelling up, and sameness
In the 2023 study, below-average consultants improved by 43% on tasks inside the frontier. Above-average consultants improved by 17%. AI narrowed the gap between weaker and stronger performers.
There was a cost. AI-assisted ideas were less varied. A 2026 review in Trends in Cognitive Sciences warns that large language models can homogenise how people write and think.
Ju and Aral found the same pattern in teams. Human-AI teams produced more output, but it was more alike.
What this means for leaders
Working the frontier well is part of Adaptation: directing, combining and checking AI in real work. Three habits help.
Map your own frontier. List your team's recurring tasks. Mark where AI does well and where it does badly. Review the map every quarter.
Check hardest near the edge. Tasks near the boundary need named checkers and more time. Treat every output there as a draft.
Protect variety. Ask for one human idea before anyone opens the AI. Keep some work unaided so judgement stays sharp.
Try this
Take three recurring tasks. Complete each once with AI and once without. Ask a colleague to rate both versions without knowing which is which.
Use the Experiment Planner to set a measure and a decision date. You will learn where your frontier sits.