The narrative surrounding AI in software engineering often oscillates between two extremes: that human developers are becoming obsolete or that AI is merely a fancy autocomplete tool. In a recent webinar featuring leadership from Platform9, Monday.com, and PlayStation, a more nuanced reality emerged. Teams are no longer just "trying" AI; they are restructuring their entire Software Development Life Cycle (SDLC) around it.
I had the opportunity to attend this webinar, and the notes below are a summary of the key ideas, practices, and observations that stood out to me during the discussion.
The Tooling Landscape: Cursor and Claude Take the Lead
While marketing campaigns feature a wide array of tools, among the panelists, Claude Code and Cursor appeared to be the most commonly adopted AI coding tools.
- Platform9 utilizes a tiered approach: UI engineers prefer Cursor, backend engineers lean toward Windsurf, and terminal-heavy users utilize Claude Code.
- Monday.com has converged on Cursor as the primary local agent for developers, while Claude Code powers the automated phases of their SDLC.
- PlayStation primarily uses GitHub Copilot due to its seamless integration with their existing git repositories and ecosystem.
- Notably, Gemini was highlighted as the least favourite for coding tasks, though it remains useful for non-technical needs like image generation.
The Rise of the "Orchestration Engineer"
A significant shift is occurring in the definition of an engineer's role. Michael from Monday.com described the transition to "agentic engineering," where developers act as orchestration engineers.
- Monday.com shared that it is experimenting with having engineers rely primarily on AI-generated code for a month-long internal initiative.
- One recurring theme was that code itself is becoming less of the differentiator; architecture, product understanding, and system design are becoming increasingly valuable. The human's value now lies in defining architecture, articulating product goals, and establishing feedback loops. For example, one team successfully built a high-performance product in Rust without prior experience with the language by letting AI handle the syntax while they managed the architectural constraints.
AI in Infrastructure and Operations
The application of AI has expanded far beyond the IDE into complex systems management:
- Root Cause Analysis: Platform9 uses agents to interrogate Kubernetes APIs. By providing LLMs with context about their specific deployments (a process they call "context engineering") they were reportedly able to identify performance and configuration issues in roughly 50–60% of investigated cases, drastically reducing resolution speed.
- Legacy Code and Documentation: Retrieval-Augmented Generation (RAG) is being used to make sense of years of technical debt and documentation. Tools like Windsurf’s "deep wiki" can generate live documentation for repositories, which is invaluable for onboarding new engineers.
Security, Compliance, and Guardrails
For large enterprises like PlayStation, security is the primary constraint.
- Data Privacy: To address privacy concerns, enterprises often rely on managed AI platforms such as AWS Bedrock or Google Cloud's AI offerings, which provide stronger controls around data handling and governance than consumer-facing services.
- Human-in-the-Loop: At major firms, AI-generated code never hits production without rigorous manual review.
- Automating Bureaucracy: AI has proven highly effective at handling non-coding "drudgery," such as summarizing international meetings for leadership or automating 300-question security questionnaires for sales teams, tasks that previously cost engineers and leadership hours of manual work.
Emerging Challenges: Noise and Bias
Despite the productivity gains, several hurdles remain:
- The Review Bottleneck: AI can generate code faster than humans can review it, leading to a signal-to-noise ratio problem in GitHub. The volume of bot comments and mentions makes it difficult for reviewers to find critical changes.
- Creative Biasing: There is a concern that using AI to approach a problem can "limit your thinking." If an engineer relies on an AI's suggestion too early, they may become biased toward that specific path and fail to explore more creative, independent solutions.
- Model Diversity: Some leaders advocate for using one model (like Claude) to generate code and a different model to review it, ensuring a "diversity of opinion" that catches more errors.
Conclusion
The future of engineering is moving toward autonomous agents that can navigate the entire SDLC. While startups are already "living on the edge" by deploying AI-generated variants directly to customers to see what works, larger enterprises are focused on building sophisticated internal guardrail systems. Regardless of the scale, the core remains the same: the most successful engineers will be those who can ask the right questions and define the right environments for their AI agents to operate within.
My biggest takeaway wasn't that AI is replacing engineers. It was that the role of the engineer is shifting. The discussion repeatedly returned to architecture, context, and decision-making rather than raw code production. Whether that vision becomes reality or not, it's clear that leading teams are already experimenting with workflows that would have sounded unrealistic just a year ago.
Top comments (0)