AI agents write more code, but not more software

Ars Technica describes a study that looked at what coding agents really deliver. They do help produce more code, yet the amount of finished software does not grow in step.

The reason is that efficiency gains get ‘absorbed’ by human review: every piece of code still has to be read, judged and accepted by a person. That creates a bottleneck, because the machine produces faster than people can check. The same pattern shows up well beyond development, wherever an agent prepares a result and a human keeps the final call.

Look at where review queues build up in your company: invoices, contracts, reports. Approvals can be offloaded too: 1Cifer agents run them along a route with a decision history and do automatic checks first, so the person only has to decide.