Farming, forestry and fishing make 18.2% of what India produces but employ 43% of its workers. Services make 54% of output with 31.8% of the jobs. The gap between those pairs is the story of Indian development: output has moved to services faster than people have.
Each worker produces about 42% of the economy-wide average.
Each worker produces about 11% more than the economy-wide average.
Each worker produces about 70% more than the economy-wide average.
Dividing a sector's share of output by its share of workers gives its output per worker relative to the economy-wide average. On that measure a worker in services produces about 4 times what a worker in agriculture does, and one in industry about 2.6 times. It is a rough measure: it ignores how many hours people work, and farm work often goes with a second job.
Economists have long described development as a shift of workers from low-productivity farming to higher-productivity industry and services. India has done part of this. Output moved to services early, skipping the long factory phase that East Asia went through, but jobs did not follow: services employ under a third of workers, and factory jobs have grown slowly.
That leaves about two in five workers in a sector that produces under a fifth of output, which is why average incomes stay low even when the economy grows fast, and why farm incomes dominate rural politics. The way out the models point to is moving workers into more productive work, which needs manufacturing and services that hire at scale.
Where the jobs are inside industry and services: Manufacturing 12.1%, construction 12.0%, trade-hotels-restaurants 12.9%, transport-storage-communication 5.8%, other services 13.1%, utilities 0.7%, mining 0.3%. See the atlas's unemployment and women in work pages for the other side of the labour market.
Two definitions matter. The output shares are of gross value added at current prices; the worker shares are people employed by their usual status, counting both principal and subsidiary work. The two come from different surveys and different years, so read the pairs as a picture, not an identity.