Data specialists market survey 2026 (Исследование рынка data-специалистов, 2026)
B · Useful, limited DevCrowd's third annual survey of Russian data specialists, run with Avito, gives useful self-reported figures on how AI changed data work, but it is an opt-in survey of 481 people with no sampling or weighting described.
Summary based on the full text·Original: about 8 minutes to read
In short
DevCrowd's 2026 survey of the Russian data job market, run with Avito, covers 481 data scientists, ML, data and AI/LLM engineers, analysts and MLOps engineers, surveyed 7 September to 6 October 2026. A year ago 3.7% used no AI tools; now only one respondent does, so the survey asked how deeply AI is embedded. The headline: AI took over code production and handed back review. Most people spend more time checking AI output, few have a formal process for it, and most say speed outruns checking.
Key numbers
- 74.4% / 67.3%code writers whose coding time fell / whose time checking and fixing AI output rose
- 63%respondents with no formal process for checking AI-generated output
- 74%who say output speed now outruns the quality of checking (48% fully agree)
- 58%using Claude or Claude Code; Claude rose from 22.4% to 36.3% in a year, ChatGPT at 75.2%
- 350,000 roublesmedian take-home monthly pay of a senior; junior, mid and senior medians unchanged in a year
As reported by the source.
Russian·Web page·opens devcrowd.ru
What it reports
- 41% are senior or above; 8.5% chose roles new to the survey (28 AI/LLM engineers, 13 MLOps engineers). 76% of role changes lead toward models.
- Only 49% of data scientists deploy models themselves; 38% of those working with models hold neither a data scientist nor an ML engineer title.
- Tools: Python 76.9%; Greenplum rose from 2.7% to 18%; Claude API in model-building rose from 3.7% to 12.3%, OpenAI API 23.9%, GigaChat and YandexGPT APIs 1.6%.
- AI modes: 46% of juniors use AI only in chat; on average 34.1% mainly use an IDE assistant and 23.6% work with agents.
- Checking: 63% have no formal review of AI output (66% in companies over 10,000 people, 60.3% under 50); 74% say speed outruns checking.
- Trust: 34% trust AI in code but not statistics; 14.2% fully trust it in code, 4.6% in statistics.
- 63% think juniors now find it harder to learn, yet only 24.8% fear their role will be unnecessary in three years.
- 44% learn from LLM answers, a source twice as common as courses. 32% of interviewees met AI bans in job interviews.
- Pay medians flat year on year; 15.5% actively job-hunting.
How they got it
An online questionnaire by DevCrowd with Avito; respondents self-select. The web report gives percentages and some group splits; it does not describe recruitment, weighting or margins of error.
Limits
- Self-reported and opt-in; the sample likely tilts toward people in DevCrowd's and Avito's networks, and Avito is also named a most-wanted employer.
- Time changes are perceptions, not measured hours.
- Some subgroups are small (28 AI/LLM engineers).
Why it matters for companies
The figures put a number on a common pattern: AI moves the workload from producing to verifying, and most teams have not set up a verification process. Data leaders can treat review of AI output as a defined step with owners and criteria, especially for statistics, where trust is lowest. The junior-learning concern and the rise of learning from LLM answers argue for deliberate training paths rather than leaving juniors to the chat window.
Key passages in translation
Russian original
Время на написание кода сократилось у 74.4% тех, кто его пишет, а на проверку и доработку сгенерированного выросло у 67.3%. AI забрал производство и вернул приёмку.
Coding time fell for 74.4% of those who write code, while time on checking and reworking generated output rose for 67.3%. AI took production and handed back acceptance.
Russian original
У 63% нет формального процесса проверки того, что сгенерировал AI.
63% have no formal process for checking what AI generated.
Russian original
63% считают, что джунам стало сложнее учиться, но за свою роль через три года боятся только 24.8%.
63% think it has become harder for juniors to learn, but only 24.8% fear for their own role in three years.
Summary based on the full text. Read the original for the full argument and data.Read the original on devcrowd.ru