Domain expertise still wanted: the latest trends in AI-assisted knowledge for developers (Category Edu)
Interesting. report Small pulse StackOverflow audience survey (900 man)It was designed with OpenAI and showed that AI is becoming a major entry point into engineering training, but it is not a substitute for expertise. The authors compared the results with the normalized results of the annual StackOverflow Developer Survey. 2024 and 2025 years (See my posts about them. 2024 and 2025). They had to be normalized because StackOverflow reports about 50k of engineers.
The main results and figures are similar
- AI has already built into the outline of engineering training: 64Percentage of developers use AI to learn 44percent 2025 and 37percent 2024
- Two main drivers: "starting from scratch" (28,2%) efficiency (26,3%)
- In parallel. 58% already use AI tools at work every day; among early-careers 68% among experienced - 56%
- 69Percentages spent the last six months learning something new
The authors noted that not only AI penetration, but also the approach to learning is changing - the number of sources that engineers use for training is shrinking: the share of those who use AI is shrinking. 8+ sources for training, fell approximately from 49percent 2024 before 9percent 2025 and 7% in this poll pulse. AI seems to be the new start page, especially for Juns and Midles. (36percent 39%, respectively, go to AI first step) Experienced developers are still holding on - they start with technical documentation in 30% of cases and only 29% start with AI:) ||I think I’m too late to post the website.|| system-design.space:) It’s encouraging to see that AI almost never works alone. Only 1They use it as their only source. The majority of AI comes with documentation. (58%)Searches/Forums/Communities (54%) Stack Overflow (50%).
The biggest stop factor is trust. 38% cite lack of trust in AI results as the main barrier to learning through AI. In the annual report 2025 The accuracy of the AI was 33Percentage, and untrusting 46%. At the same time, daily users have significantly higher trust than weekly users: 49percentage 30%. In other words, AI speeds up entry to the topic, but adds a tax to verify results.
Interestingly, those that AI wins by reducing the barrier to learning new things and reducing the cost of the “where to start” moment. You can see this from the fact that 35The percentage of those who do not use AI say that they do not have enough time to learn. At the same time, such a reason as an impediment to learning is indicated only by 7The percentage of those using AI:)
In the end, we are waiting for a future where the AI tool will be the starting page and the knowledge base. (Stack Overflow) It will provide a layer of trust on top of which AI will build its responses. The value of expertise at the same time will only grow, and the main skill will be not just good prompting, but also the ability to quickly and efficiently verify AI answers.
#Survey #Engineering #Software #AI #Culture #RnD #Thinking #Brain