dooopSoftware · Organization · 5 min

How to Discuss Productivity with AI in Leadership

Lead a productivity conversation with AI by distinguishing perception, measurement, and outcome. Define comparable tasks before setting goals.

Published on September 6, 2026

MAIN THESIS

Different areas may use productivity to support incompatible decisions.

A goal must specify the unit of work, the quality preserved, and the scope of comparison.

A leadership conversation about productivity with AI needs to start by defining the work to be compared. Producing more code suggestions, completing tasks earlier, and delivering more value to the customer are different measures. Before setting a goal, agree on the unit of work, the completion criteria, quality signals, and comparison conditions. Without this agreement, different areas may use the same word to defend incompatible decisions.

Separate perception, measurement, and product outcome

Team perception is a useful input. People may report that they started faster, found an explanation easily, or spent more time reviewing. These reports help locate what to investigate. They should not be directly converted into a productivity percentage.

A measurement needs to specify what was counted and in what context. Implementation time does not necessarily include waiting, review, or later correction. The number of changes alone does not indicate the size or usefulness of each.

Product outcome is another layer: was the user’s task easier to complete? Was there a reduction in a relevant failure? A team may improve its technical flow without a single change immediately affecting a business metric. The conversation should preserve this distinction.

Use available evidence with its limits

In the February 2026 update, METR treated its new data as an unreliable signal of AI’s current effect on productivity. Among the difficulties noted were participant and task selection and measuring time with competing agents.

This reference does not allow concluding that AI always increases or decreases productivity. It shows why measurement design matters. When presenting an internal result, explain who participated, which tasks were included, and what was excluded.

The DORA 2025 presentation describes AI as amplifying existing organizational strengths and weaknesses. Use this perspective to examine the work system. A review bottleneck or an imprecise task definition may remain relevant after adopting the tool.

Prepare an agenda that ends with a decision

A leadership meeting can organize the discussion into four blocks. First, the observed problem. Then, the available evidence. Next, possible explanations. Finally, the decision the organization can sustain now.

Ask each proposal to answer:

What type of task is being evaluated?
What counts as completed?
Which review and correction efforts are included in the observation?
What quality signal must not worsen?
What comparison would be reasonable under current conditions?
What result would lead to expanding, adjusting, or stopping the practice?

This agenda prevents the discussion from ending only in a preference for a tool. It also allows a limited decision: maintain use in a specific task while the team collects better evidence, without turning a pilot into general authorization.

Fictional example: faster implementation, longer review

Consider a team using AI to propose changes in internal forms. Developers report that the first version was ready earlier. The person responsible for review reports more changes outside the request and more questions about validation rules.

Leadership does not need to choose which report is “correct.” Both may describe different parts of the flow. A useful investigation would separate preparation, implementation, waiting, review, and rework into comparable tasks.

One possible intervention would be to reduce the scope of requests and require the person to check the difference produced before requesting review. The test would observe whether the change reduces returns without harming behavior. The example illustrates an organizational hypothesis; it does not claim the intervention produced a gain.

Avoid turning the metric into an incentive for wrong behavior

If the goal rewards only the number of proposals opened, the team may prioritize fragmentation and volume. If it considers only approval speed, it may pressure review to close doubts too early. Treat these possibilities as risks to observe when designing the metric.

Combine a flow indicator with quality criteria and a reading of context. Urgent changes, exploratory work, and repetitive corrections should not be compared as if equivalent. Record these differences before interpreting variation.

The Microsoft ExP describes the link between hypothesis, measurement, and iteration in development. For a team practice, the proposed application is to make explicit the expected effect and the subsequent decision. It is not necessary to call every change a controlled experiment if the design does not support that classification.

End with a verifiable scope

The final decision should record a practice, a task type, a person responsible for observation, and a review moment. It may be to expand a limited use, correct the flow, repeat the comparison, or stop an application that did not present sufficient evidence.

Also record what cannot yet be affirmed. An improvement observed in one team or task does not prove an effect for the entire organization. An inconclusive result should not be presented as success nor as definitive failure of the technology.

In the next conversation about productivity with AI, replace the generic goal with a verifiable question: which part of the work do we want to improve, preserving which quality conditions? The answer should guide observation and limit the conclusion leadership can present.

If you want to discuss this decision in the context of your company, talk with dooop.

Sources

NEXT DECISION

Discuss application in the company

Conversation about the software company context

Content by dooop. Registration allows linking this agenda to the reader’s journey and tracking interest in the topic.