[3/3] Integrating AI into the development processes of a large company: why allowing all Cursor is not an option (Category Management)
Examples of large companies in previous post You can see that they come from human empowerment scenarios, and some of them have been successfully agented. And if you look at our approach at T-Bank, we're going about the same way. In fact, this path can be learned in detail from the speeches of my colleagues at Platform Engineering Night.
- AI and Platform Engineeringfrom Igor Maslov
- Developing your own AI assistant for code: sprint or marathon? from Denis Artyushin In a nutshell, we've gone through the first level of hygiene, now we've closed many of the second level scenarios, and we've come very close to delegating the work of engineers to agents. It's interesting that we have a centralized team that makes our copilot NestorBut at the same time it is an umbrella brand for other initiatives that are driven by our working groups on planning, analysis, design, development, testing, SRE. For example, here. speech About the introduction of SRE AI-assistant. In general, we, with our observability platform Sage, are going in about the same direction on AI-fiction as Datadog, mentioned earlier.
Intensively, this general trend about the development of platforms in the genAI world is clearly visible from the RedHat report from November. 2024 State of Platform Engineering in the Age of AI (review)where the main conclusions are
- Platform Engineering Becomes Strategic 62% of companies already have dedicated platform engineering teams. Platform engineering is not just infrastructure automation, but a strategic tool to accelerate innovation and AI adoption.
- The impact of artificial intelligence 76% of organizations already use generative AI for development tasks (documentation, code generation, tips). 45Percentages see AI as a central element of their platform strategy.
In the end, if you answer the question posed in the title “why to allow all Cursor is not an option?”, then you can say so.
For small businesses, this is an option, because These tools allow you to quickly issue code for landing pages, prototypes, MVP They are not very expensive, and there are usually no special risks with intellectual property, data, sanctions.
- You still need engineering skills - so that it does not turn out like in the story of how the agent demolished all the code, and you can not recover from git, since the vibe coder did not know what git was and what it should use.
For large companies, the approach is more complex, as I described above, because Out of the box, Copilot, Cursor, Windsurf do not know how to work with platform tools. This can be taught if implemented for MCP server platform products, but... For large companies, the risks around information security, intellectual property, and personal data are often too great to use external SaaS solutions. In the process of working on LLM within the SDLC, such companies want to practice on cats and learn how to AI-fiction products that are safe and understandable. And then with this expertise go towards customer products of the company.
- Well, big risks arise due to sanctions, since external SaaS solutions may be officially unavailable, and having access to another, it can easily be lost and left at a broken trough. As a result, large companies often build their LLM solutions and learn to integrate them into their SDLC and products. This can often be justified economically on their scale. Interestingly, this is facilitated by the publication of scales for models such as LLama, DeepSeek, Mistral, which allows you not to lag too far behind mainstream.
But medium-sized companies have a difficult time - external decisions can be too risky, and to create their own may not have enough money. But now some Russian companies are trying to make their products around SDLCs, in which AI becomes a first-class resident.)
#AI #PlatformEngineering #Engineering #Software #Processes #Productivity