Leadership in AI Assisted Engineering (AI column)
Interesting. report Justin Reock, Deputy CTO of DX (recently bought Atlassian for $1 billion). Reock is known for its work on measuring and optimizing developer productivity, making it one of the industry’s leading voices on AI in development. In this talk, he talked about the productivity of engineers and gave many references to the research that I mentioned earlier. Below are the main points of the report.
1AI Productivity Paradox: Huge Variability in Results Contrary to hype, the real impact of AI on productivity is extremely heterogeneous. For example, Sundar Pichai, CEO of Google, in June 2025 told forward 10Increase in engineering capacity. In the same summer, the METR study showed 19Percentage of slowing down with AI – even though developers felt 20% faster. I sorted through this research in posts. 1 and 2 even podcastwhere it is shown that the study itself is designed for a specific result + represents a very narrow case. Based on the DX platform, where 300+ Companies, the averages look modest: +7.5% quality of documentation, +3.4% code quality, +2.6% confidence in changes, -1% change failure rate. But if you look at the variability of the results, some see +20% improvement in metrics, other 20Percent deterioration of metrics. This seems to be driven by an approach to implementing AI and measuring effects.
2Psychological security - the foundation of AI transformation The author of the presentation remembered about Project Aristotle from Google, which almost 15 years. So the guys at Google decided to do it. teamwork What makes a team effective at Google? The answer is psychological safety. Details about the study I have already told. Psychological safety is important when implementing AI Fear of AI replacement leads to sabotage of implementation Without trust, people don’t experiment with tools. Transparent Communication: AI Complements, Not Replaces
3️⃣ W. Edwards Deming: 95% - system, 5% people Edward Deming said that 90-95Percentage of productivity is determined by the system (processes, tools, culture). The rest depends on the individual qualities of the employee. If this is transferred to the development process, Performance reviews, individual coaching - work 5%, waste of human talent
- The focus should be on optimizationInfrastructure, Workflows, Developer Experience If something goes wrong, look for problems in the system, not in people.
4DX AI Measurement Framework: Three Dimensions of Success The speaker spoke about the approach measurement of AI effects three-axis
- Utilization (Use of use)
- Impact (Impact)
- Cost (Cost) I told you about the approach of the guys earlier. parsing + my podcast along with Zhenya Sergeev. But if we talk about the thoughts of the speaker, he says that the baseline in the form of standard metric developer productivity. (DORA, SPACE) AI-specific metrics are more important than AI-specific metrics – they show a real impact on outcomes.
5AI integration across SDLC, not just coding For most organizations, writing has never been a bottleneck – often they have been elsewhere: searching for information and context, delaying code reviews, coordination difficulties, handling incidents, and legacy code. Further, the author gave a number of examples of AI implementation in different companies: Morgan Stanley, Zapier, Spotify, Booking.
6Top--5 priorities for leaders
- Reduce AI Fear Through Transparency
- Invest in employee education + allocate time for experiments
- Introduce measurements and start an open discussion about metrics
- Unblock usage – implement a creative approach to compliance
- Create feedback loops for continuous improvement
#Software #Engineering #Productivity #DevEx #AI #Management #RnD #Leadership #Economy #Metrics