AI and Platform Engineering (AI column)
Interesting. speech Igor Maslov, VP of Coretech & Data in T-Bank, at the opening of the Platform Engineering Night conference. told. Igor opened the conference and talked about how AI affects the work of engineers and the development of engineering platforms. The main thoughts of the presentation are as follows:
1. AI as an amplifier of existing platforms Now it is necessary to “smear” the existing engineering scenarios of AI – from the creation of pipelines to the analysis of telemetry. In this section, it’s interesting to look at Google’s whitepaper “Measuring Developer Goals” to see which major scenarios Google highlights. (me told before)and ninth edition "Research Insights Made Simple," in which my colleague Kolya Bushkov and I analyzed the whitepaper "What Do Developers Want From AI?" from the guys at Google. 2. Cognitive load and loss of skills Igor notes that the active use of AI leads to the loss of low-level skills, but considers it a normal evolution - conventionally, few people write in assembler and most engineers do not understand it and this does not bother anyone. We’ll also be writing code with Java assistants, and some people without assistants won’t be able to read it anymore.) 3. Copylot mode as a basic level Responsibility for the final result remains with the person, which minimizes the risks of "hallucinations" of AI models. Anthropic implements this philosophy in Claude, where the model acts as an assistant rather than a replacement for the developer. This approach is particularly important in critical systems, such as finance or healthcare, where AI errors are unacceptable. 4. The Future: AI as Platform Engineering In the long run, AI could become the primary interface for managing engineering processes, eliminating the need for traditional tools. This trend is reminiscent of the evolution of Kubernetes, which became the standard for container orchestration, but for AI, a similar “moment” has not yet arrived. 5. Vibe Coding and Personalized Software Generating code for specific business cases or individual user needs is a key trend. However, as Igor warns, such systems require careful validation to avoid problems with code support. 6. Unloading developers from routine 80Percent of experimental work can be automated, freeing up time for more important tasks. OpenAI’s Code Interpreter allows you to iteratively solve complex programming and data analysis problems. 7. ML Platforms: Flexibility and Readiness for Change n Investing in ML platforms is warranted, but requires readiness for technological shifts. For example, the transition from traditional neural networks to transformers 2020He demanded a complete overhaul of the infrastructure of a large number of companies. The speaker emphasizes that platforms must remain modular to adapt to new algorithms and hardware solutions. n 8. Lack of Kubernetes Moment in AI There is no universal standard for AI platforms yet, and it’s worth waiting for it to arrive. Each company develops its own approaches: OpenAI with Agents SDK, Anthropic with a focus on security, Google with comprehensive AI solutions. 9. The High Risk Nature of Current AI R&D Modern AI solutions require large resources and are suitable mainly for large companies. This is confirmed by the massive investments of OpenAI, Google and Microsoft in platform solutions. 10. Gradual approach to implementation The recommendation is to start by improving tools, but refrain from complex low-code processes. Anthropic follows a similar philosophy, providing powerful models with a focus on controlled implementation.
In general, Igor had a great talk with the main idea that the AI revolution in platform engineering is inevitable, but we need a balanced approach with a focus on gradual improvement of existing processes, rather than radical replacement.
#Management #Leadership #Software #SoftwareDevelopment #Metrics #Devops #Processes #AI #ML #DevEx