One in Four Programmers Already Works With AI Agents: T1 Survey of 190 Developers
A quarter of the 190 programmers surveyed already use AI agents or orchestrate several models, according to a September survey by IT holding T1. Among them, 82% hand agents code writing and refactoring, 80% planning and debugging, and 61% deployment and infrastructure. The sample is small, so these figures are a snapshot of sentiment, not market data.
AI-processed from Habr AI; edited by Hamidun News
A year ago, AI in development meant suggesting the next line. Now some programmers set a task, and an agent breaks it into steps, writes the code, runs the tests and prepares the documentation. To find out how widespread this is in practice, IT holding T1 surveyed programmers through nine Telegram channels at the end of September 2026. 190 people responded. The sample is small and non-random, so the numbers are best read as a benchmark rather than industry statistics.
What Developers Hand Off to Agents
Roughly a quarter of respondents, about 47 people, use agentic systems or orchestrate several AI models. The percentages below refer to this group, so they represent dozens of people, not thousands.
The most popular task is the expected one: 82% trust agents with writing and refactoring code. But nearly as many use them for work around coding:
- planning and task decomposition: 80%;
- bug fixing and debugging: 80%;
- test generation and execution: 76%;
- documentation: 74%;
- deployment and infrastructure setup: 61%.
Here the agent works less like autocomplete and more like an executor across several stages, from parsing the brief to checking the result. Among those who use agents for tasks that don't involve writing code directly, 78% do so daily and another 11% several times a week. Deployment lags behind the other items, which is understandable: mistakes there are the most expensive, so teams hand those tasks over more cautiously.
Who Uses What
Manual coding hasn't disappeared. 46% of participants rely on AI assistants such as autocomplete and chatbots, and 15% still write code entirely by hand. Among assistants, Claude Code leads with 60%. It is followed by ChatGPT (57%), DeepSeek (32%) and Gemini (31%).
The picture is more interesting among agent users. OpenClaw and Hermes were each chosen by 39%; these are platforms where developers build their own agents. ZeroClaw and NanoClaw were named by 14% each. Another 48% use OpenCode, Cline, Cursor and Claude Code, as well as in-house tools and self-configured tool chains, including orchestration inside Codex.
For product managers and team leads, the takeaway is this: many don't take a ready-made agent out of the box, they assemble their own chain around the team's processes. Value is therefore shifting from the model itself to how the agent is integrated with the repository, tests and pipeline.
Demand for Russian Solutions
A separate block of the survey addresses import substitution. 56% of participants would like Russian AI-native development tools that match foreign platforms but run in a closed environment. Among agent users, 55% are willing to switch to a Russian alternative with comparable functionality. For corporate teams, what matters is not only answer quality but also where the model runs, where the code goes and how the tool fits security requirements.
How to Read These Numbers
There are several caveats. First, the sample of 190 people was collected through Telegram channels, including T1's own channel. It likely contains more AI enthusiasts than the market average. Second, the survey shows stated habits, not measured results: it cannot tell us how much time agents actually save or how often they make mistakes. Third, the material was published by the product lead of the platform Sfera, part of the same holding, and its conclusion about integration into corporate environments partly promotes the company's own product. T1 has not yet released a full report covering more than 11,000 IT specialists, and that report deserves a separate look.
The practical conclusion is simple. If your team is trying agents, start with tasks where the result is easy to verify: tests, documentation, debugging. Add deployment and infrastructure later, and only with clear access restrictions. And if you work in a closed environment, decide in advance what will replace foreign tools: according to the survey, there is demand for such solutions.
- Nov 21, 2026FDE job postings grew 4.6x in a year: what a forward deployed engineer actually does at the client's site
- Nov 21, 2026A swimming team captain handed ChatGPT three scheduled tasks and cut her routine work by a multiple
- Nov 21, 2026OpenAI GPT-6 Cyber Found About 35 Flaws in iPhone, Mac and Other Apple Devices
- Nov 21, 202622 Claude Agents Spent Six Weeks on Black Hole Problems — and Solved Two
Key Questions Answered
What Developers Hand Off to Agents?
Roughly a quarter of respondents, about 47 people, use agentic systems or orchestrate several AI models. The percentages below refer to this group, so they represent dozens of people, not thousands. The most popular task is the expected one: 82% trust agents with writing and refactoring code. But nearly as many use them for work around coding: planning and task decomposition: 80%; bug fixing and debugging: 80%; test generation and execution: 76%; documentation: 74%; deployment and infrastructure setup: 61%.
Who Uses What?
Manual coding hasn't disappeared. 46% of participants rely on AI assistants such as autocomplete and chatbots, and 15% still write code entirely by hand. Among assistants, Claude Code leads with 60%. It is followed by ChatGPT (57%), DeepSeek (32%) and Gemini (31%). The picture is more interesting among agent users. OpenClaw and Hermes were each chosen by 39%; these are platforms where developers build their own agents. ZeroClaw and NanoClaw were named by 14% each.
What are the key details on Demand for Russian Solutions?
A separate block of the survey addresses import substitution. 56% of participants would like Russian AI-native development tools that match foreign platforms but run in a closed environment. Among agent users, 55% are willing to switch to a Russian alternative with comparable functionality. For corporate teams, what matters is not only answer quality but also where the model runs, where the code goes and how the tool fits security requirements.
Build & Ship Production AI Agents Without Boilerplate Code
Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.