Skip to content
#Management

[1/3] Integrating AI into the development processes of a large company: why allowing all Cursor is not an option (Category Management)

#Management #AI #PlatformEngineering #Engineering #Software #Processes #Productivity

Tonight, 17.00 I'm doing this. Report at CTO Conf XA professional conference for technical directors from Ontico. I will talk about the hype topic of integrating artificial intelligence into the life cycle of software development in large companies. Below I will describe the structure of the report and provide links to additional materials that I use for references in my talk. In my talk, I start by thinking about copilot tools: GitHub Copilot, Cursor, WindsurfThey set some standard for AI assistants for developers and let you do a lot of things automatically. February 2025 Andrej Karpathy beginning A new development trend called vibe coding

There's a new kind of coding I call "vibe coding", where you fully give in to the vibes, embrace exponentials, and forget that the code even exists.

This trend was picked up by the guys from the startup accelerator Y Combinator and in March began to discuss this topic in podcasts:

Separately, it can be added that the hype is added by the statements of Sam Altman, CEO of OpenAI, or Dario Amodei, CEO of Antrophic. For example, Dario three months ago at the talk "The Future of U.S. AI Leadership with CEO of Anthropic Dario Amodei," which I already talked about, gave a prediction about the future of development.

I think we will be there in three to six months, where AI is writing 90% of the code. And then, in 12 months, we may be in a world where AI is writing essentially all of the code

The question is, how can this be achieved? The answer to using agents:

  • Last year, the guys from Antrophic presented MCP (model context protocol) Provide LLM access to additional tools This year, Google introduced a protocol for the interaction of agents. A2A (Agent2Agent) protocol In general, the topic is now hype and for creating MVPs in startups or pet projects by developers, this approach to using copilots in vibe coding mode is great. But for large companies, it is not so simple and I will explain why.

Engineering processes in large companies have evolved as follows: At one time, development and operation were separated, and this gap prevented business results. As a result, from the mid-2000s to the end of the 2010s, the DevOps approach was evangelized, which from a scientific point of view was justified in the book "Accelerate", which I described earlier in three parts: 1, 2 and 3. This approach often led to a heterogeneous IT landscape with a large duplication of systems, which did not allow to get the effect of masstaa - large companies went towards separating stream-aligned teams and platform teams that were supposed to create platforms, such as the Internal developer platform, which would allow an engineer in a self-service format to use tools such as working with code, artifacts, CI / CD pipelines, random, observability and so on. Then the platforms became quite complex and the owners of the platforms decided to go towards the user experience of their users, which are developers. Developer experience includes flow state, cognitive load, feedback loops. (You can read about it in the whitepaper DevEx: What Actually Drives Productivity. told earlier). This is important, because the complexity of platforms can be overwhelming, and you can demonstrate this by looking at different platforms. CNCF landscapewhere the number of cards of products goes off the scale and deal with what and how an ordinary person is extremely difficult.

Continuation in next post.

#AI #PlatformEngineering #Engineering #Software #Processes #Productivity