When consultants team up with AI, which choices do they keep for themselves — and which do they hand over?

This explores which choices consultants keep, and which they give away, when they work with generative AI, and why that split matters. A field study of 244 BCG consultants points to two decisions at the center: who picks the task and who picks the method for doing it. The tool doesn't define the working pattern, and neither does the task. Who holds those two choices does. Three patterns came out of it: Centaurs, Cyborgs and Self-Automators Do humans and AI divide labor differently in knowledge work?. The less obvious finding is that each choice works like a learning channel. Centaurs, who kept control of the method and steered how the AI approached the work, got better at consulting itself. Cyborgs, who worked closely with the tool, built new AI expertise instead. Self-Automators handed off both choices and gained neither skill Do different GenAI collaboration modes produce different skill gains?. So the choices you keep are also the skills you keep building.

A third choice sits under those two: deciding whether this task is one the AI is actually good at. In the same BCG setting, GPT-4 made consultants 12% faster and lifted quality by 32% on tasks inside the AI's capability range. On a task just outside that range, their accuracy fell by 19 percentage points. The AI's confident, persuasive answers hid its mistakes, and consultants went along without noticing Does AI help or harm consultant performance on unfamiliar tasks?. Telling which tasks the AI can handle is the decision a consultant can least afford to give up, and it's also the hardest one to see you've given up.

Other professions point the same way. Lawyers stay accountable for the facts in an AI summary. When they can't see where the summary got its claims, checking it can take longer than doing the work by hand Does GenAI actually save lawyers time on fact verification?. Keeping responsibility without visibility into the reasoning is expensive. One proposed design fix flips the setup: the AI points out which parts of a case deserve attention, and the human still makes the call. This reduced anchoring, the tendency to stick with whatever answer the AI offers first Can AI guidance reduce anchoring bias better than AI decisions?.

Keeping control has a psychological side too. Across four experiments with 3,562 people, collaborating with GenAI raised people's sense of control when they went back to working alone, but it lowered their motivation and made the work more boring. The AI had taken over the engaging parts and left the routine leftovers Does AI collaboration drain motivation when workers return to solo tasks?. Interviews with 38 knowledge workers found skills developing, holding steady, eroding or changing in value, depending on which tasks people handed off How does generative AI actually change worker skills?. There's also a social cost. In interviews with 19 knowledge workers, people hid signs of AI use to look like experts, which stopped colleagues from learning how others split the work Why do knowledge workers hide signs of using GenAI?. The overall lesson: what to hand off is itself a decision worth keeping, and organizations rarely make it visible enough for people to learn it from each other.