The bottle neck for AI adoption - particularly in the type of companies the Economist is surveying - is the management structure, the "chain of command".

As a manager you have to strike a balance between absolute productivity/efficiency and accountability diffusion: lose enough people to spread accountability and it ends up on your plate when things go wrong. "AI f*cK it up" does not sound like as good an argument for why your chain of command messed things up as "joe doe made a mistake".

Also, AI agents don't make for great office politics partner.

As much as AI is being portrayed as a human replacement there are many social aspects of organizations that AI cannot replace. AI don't vote, AI don't get promotions for siding with different parties in office politics. AI can't get fired

To add to this point, AI is lagging in the debugging chain. It's great at generating code but who is going to inherit the maintenance of all this code and show confidence in its maintenance? Most developers already works on the maintenance side, writing a handful of line of code per week, spending most time in meeting, reading code, writing a small modification, doing all sort of manual operation, testing it, waiting for the test to complete, etc. More so, the whole Agile/Scrum movement encouraged to sort of micro-modification approach. You can't just drop an AI in place of those developers.