What happened to the money no longer spent on software by that one-third.
A global McKinsey survey published on August 25, 2026, reported one figure: 32% of companies surveyed said they had given up buying off-the-shelf software because they could build their own using agentic coding tools. The figure was close to half among high performers, and about 31% among other companies. The survey covered 1,719 respondents in 97 countries, with fieldwork conducted from May to June 2026.
The report doesn’t say how many of that 32% had done the math before taking action. The article offers six steps: first, write down a role that could be handed off, specifying when it runs, what it receives, what it delivers, and who makes the call when something goes wrong. Then count how many times it runs in a year, and how many of those runs involve the same action. Finally, put three costs on the same sheet: the one-time cost of building it, the ongoing cost of running it, and the cost of exiting when you stop using it or switch to something else.
Choose a path only after the third step. For low-frequency tasks that need a person to review each result, buy an off-the-shelf solution. For high-frequency tasks with clearly defined inputs and outputs, where the process is unique to your company, hire for a dedicated role. Build it yourself only when the task is high-frequency and the task itself is a source of differentiation. The final two steps are to draw the boundaries of its permissions and to walk through the exit process yourself before handing it over.
The same survey also found that about 20% of companies had already restricted some usage because of AI operating costs, including token consumption.
From our company’s official WeChat article, “What happened to the money no longer spent on software by that one-third”: all six steps and five self-check questions are in the original. Crypto assets are volatile; this is not investment advice.
A global McKinsey survey published on August 25, 2026, reported one figure: 32% of companies surveyed said they had given up buying off-the-shelf software because they could build their own using agentic coding tools. The figure was close to half among high performers, and about 31% among other companies. The survey covered 1,719 respondents in 97 countries, with fieldwork conducted from May to June 2026.
The report doesn’t say how many of that 32% had done the math before taking action. The article offers six steps: first, write down a role that could be handed off, specifying when it runs, what it receives, what it delivers, and who makes the call when something goes wrong. Then count how many times it runs in a year, and how many of those runs involve the same action. Finally, put three costs on the same sheet: the one-time cost of building it, the ongoing cost of running it, and the cost of exiting when you stop using it or switch to something else.
Choose a path only after the third step. For low-frequency tasks that need a person to review each result, buy an off-the-shelf solution. For high-frequency tasks with clearly defined inputs and outputs, where the process is unique to your company, hire for a dedicated role. Build it yourself only when the task is high-frequency and the task itself is a source of differentiation. The final two steps are to draw the boundaries of its permissions and to walk through the exit process yourself before handing it over.
The same survey also found that about 20% of companies had already restricted some usage because of AI operating costs, including token consumption.
From our company’s official WeChat article, “What happened to the money no longer spent on software by that one-third”: all six steps and five self-check questions are in the original. Crypto assets are volatile; this is not investment advice.