News and analysis
Teams after AI: where faster preparation gets stuck
JetBrains discusses handoffs; Google Research distinguishes output quality from employee skill. Assess the whole workflow rather than generation speed alone.
A developer prepares changes faster, yet the owner still waits for release. The delay may lie in agreeing the task, reviewing it or handing work to the next person. Two publications dated 7 October consider AI’s value beyond an individual’s speed. They support reassessing a workflow without promising the same effect for every team.
Fast creation does not complete the handoff
JetBrains reports observations from conversations with engineering organizations: individual agent practices differ, shared rules and settings are not always maintained, and bottlenecks may move into planning and review. These are qualitative vendor observations, not representative productivity statistics.
For a small team, ask a concrete question: what should a reviewer receive with the proposed change? If they must reconstruct the task, consequences and verification method, time saved by the author may be lost at the next stage. Shared instructions are useful when they describe the project’s actual decisions and change alongside them.
Output quality and independent skill are different
Google Research discusses a three-month randomized study involving practising patent attorneys. It examines work quality with AI and participants’ independent skills, with different results for different experience levels. This concerns another profession; its effect cannot be transferred as a percentage to developers or your company.
The comparison suggests a cautious conclusion: assess not only the finished document or code, but also the team’s ability to explain the decision, identify an error and continue without a particular tool. This is a design recommendation, rather than a measured outcome for your staff.
Inspect one queue of work
Choose a recurring task and follow it from the agreed request to acceptance. Record who waits, which questions return to the author and what the next person needs. Compare similar task classes. Line counts and the size of an agent response say little about completion of the operating workflow.
For a pilot, agree a short handoff package: task, decision, affected parts and completed checks. Give the reviewer a way to reproduce the result. Then ask another employee to explain the decision and handle a typical exception. This identifies where training and documentation are needed.
If faster creation expands the review queue, first reduce change size or clarify acceptance conditions. Buying another tool should not replace an examination of that handoff. Assign someone to maintain the shared working instructions, so each improvement can be understood by the next person rather than remaining in a private conversation.
Our business rules guide helps make expectations explicit. The consulting partnership confirms sustained work across different tasks without an invented AI productivity figure. For your internal system, discuss where the team waits for its next step.