# Redesigning How Software Gets Built With AI Agents — Sonar & McKinsey Panel

## Executive summary

Scaling AI agents in the Software Development Life Cycle (SDLC) requires a systemic overhaul that extends beyond tooling. While most companies are experimenting with AI agents, fewer than a third are achieving measurable business impact. Successful scaling hinges on redesigning the entire workflow (process), implementing robust tooling, and, most critically, remodeling organizational roles and fostering explicit knowledge transfer. The transition involves moving from human-guided agents (Horizon 2) toward fully automated, end-to-end orchestrated pipelines (Horizon 3), building trust incrementally through stages like AI code review, blocking, autofixing, auto-approval, and auto-merging.

## Key takeaways

- Scaling requires three factors: Process, Tooling, and People: Organizations achieving high productivity from AI must redesign their entire workflow end-to-end, not just optimize traditional SDLC stages. Furthermore, the boundaries between traditional roles (SWE, PM, Designer) are blurring, necessitating a remodel of the operating model.
- Role modeling is insufficient for adoption: Simply establishing 'lighthouse teams' that use agents effectively is not enough to drive company-wide change. The critical missing link is helping the broader organization think differently and understand the 'why' behind agent usage.
- Trust builds through incremental automation: The journey to full automation is gradual. Trust is built by moving through stages: AI reviews $\rightarrow$ blocking on issues $\rightarrow$ automated fixes $\rightarrow$ auto-approval $\rightarrow$ auto-merge. This precision is key to unlocking higher levels of automation.
- The future demands generalists and critical thinkers: As the SDLC becomes more automated, the most valuable skills are becoming explicit thinking (making implicit knowledge visible), critical thinking (questioning AI's high-conviction output), and generalist expertise, rather than deep specialization in one area.

## Technical details

- SDLC Maturity Model: The maturity scale progresses from Horizon 2 (human-guided and validated agents executing tasks) toward Horizon 3 (fully end-to-end automated workflow with human oversight).
- Automation Workflow Stages: The automation pipeline progresses from using AI for code reviews and CI failure analysis to setting rules to block PRs based on agent findings, then implementing automated fixes, and finally achieving autonomous approval and merging for specific code sections.
- Developer Productivity Metrics: Large organizations must establish baseline metrics (e.g., code throughput, quality, error rate, developer sentiment) to measure the impact of AI agents and guide iterative improvements across the PDLC.

## Practical implications

- Shift focus from merely implementing AI tools to redesigning the entire organizational workflow and operating model.
- Treat the adoption of AI agents as a process improvement challenge, not just a tooling upgrade.
- Prioritize building developer metrics and baselines before deploying agents to accurately measure ROI and guide iteration.
- Encourage cross-functional generalist roles to break down historical departmental silos.

## Topics

AI Agents, Software Development Life Cycle (SDLC), DevOps, Process Engineering, Organizational Design, Sonar, Gitar by Sonar

Source: https://www.youtube.com/watch?v=XqF-IFHBCkM