State of AI4SDLC on HighLoad++: materials for the report (Category AI4SDLC)

I spoke today. Saint HighLoad++ 2026 State of AI4SDLC: How AI is Changing Development Processes in Large Companies The main idea I had was that AI4SDLC is now about redesigning the engineering system: platform, processes, metrics, security, economics and the very role of an engineer.

1️ In the first part, I showed that the implementation of AI has already happened, but trust and team effect do not keep up with it. Coding accelerates faster than change delivery (delivery)So the old bottlenecks just get more visible: task setting, review, tests, CI/CD, security, releases, and context. 2️ In the second part, I analyzed what it looks like on the scale of a large fintech on the 10 000+ engineers. The logic of “let’s let everyone have a good AI editor” no longer works. We need a platform for agents. (agent-first): model gateway, instrumental MCP gateway, capability registry, trust levels read -> recommend -> act, quality checks (evals)Telemetry and threat model. 3Next, I showed what blocks are already appearing in a real engineering environment: single access to models, managed access of agents to internal tools, internal assistant with company context, agent mode. (agent mode) on the development platform and domain agents throughout the SDLC: testing, review, design, security, operation, migration, data/DS and infrastructure. 4The other one was about measurement. Considering the share of AI code is almost useless: this metric is easy to inflate, and it does not answer the question of whether the engineering system has become stronger. Look at the chain. внедрение -> throughput -> качество/риск -> экономика: DORA, SPACE, DevEx, alterations (rework)Review time, failed pipeline runs, recovery time (recovery time)The value of tokens and human time.

The bottom line is that it is not the team with the most fashionable tool that wins, but the team that can turn AI into a managed production system. With a clear baseline (baseline)Owners, context, constraints, checks, observability and honest talk about value.

Below I provide additional material to this report. (my articles)

• HighLoad slides + +
• AI4SDLC Research 2025
• From classic PDLC to AI-native
• IDP is Dead? No, the GUI monopoly is dying.
• Agent-first IDP playbook
• Tg post about spec-driven development
• Tg post with an analysis of Anna Gromova's report on AI4SDLC metrics

More about measurements (DORA / SPACE / DevEx):

• Dora ROI: how to calculate the effect of AI-assisted development without magic
• DORA 2025 - State of AI-assisted Software Development
• DORA AI Capabilities Model
• How quietly DORA became not four, but five metrics
• The SPACE of Developer Productivity
• DevEx: What Actually Drives Productivity

#AI #AI4SDLC #Engineering #PlatformEngineering #DevOps #Management #Conference