Atlassian Engineering Leader Implements AI-Driven ‘Pit Wall’ to Preserve Empathetic Communication at Scale

In the modern landscape of distributed software engineering, the transition from managing small, localized teams to overseeing sprawling, global organizations often results in a measurable decline in communication quality. Fernando Garcia Valenzuela, the Head of Cloud Storage Engineering at Atlassian, found himself at the center of this challenge while managing a department of more than 70 professionals spanning three distinct time zones. Recognizing that traditional leadership methods struggle to maintain personal connection at such a scale, Garcia Valenzuela developed a sophisticated AI-driven feedback system designed to audit and enhance the "warmth" and effectiveness of his digital interactions.

The initiative stems from a fundamental observation regarding the nature of professional empathy in a high-pressure environment. Garcia Valenzuela notes that while many leaders perceive themselves as competent communicators, the reality of "busyness" often acts as a corrosive force on interpersonal connections. He posits that warmth does not scale automatically and that empathy cannot survive the rigors of a heavy workload without deliberate, structured effort. Refusing to allow a high volume of responsibilities to serve as a justification for his team members feeling overlooked, he sought a technological solution to bridge the gap between efficiency and human connection.

The resulting system, which Garcia Valenzuela refers to as his "pit wall," draws inspiration from the high-stakes world of Formula 1 racing. In professional motor racing, the pit wall serves as the nerve center where strategists and engineers analyze real-time data to provide drivers with actionable insights without distracting them from the primary task of driving. Garcia Valenzuela applied this mental model to leadership communication, creating AI agents that surface insights into his own messaging habits—providing feedback that is timely, contextual, and actionable without disrupting his daily workflow.

The technical architecture of this system utilizes tools already embedded within the Atlassian ecosystem. Data is sourced from Slack, the primary communication hub for the team. This information is processed through Atlassian Rovo, an AI-powered organizational search and agent platform, and the resulting reports and insights are stored and tracked in Confluence. This portable stack allows for a continuous feedback loop that monitors both the immediate tone of messages and long-term behavioral patterns.

To manage this feedback, Garcia Valenzuela deployed two distinct AI agents, each named after iconic figures in Formula 1. The first agent, "Smedley"—named after Rob Smedley, the race engineer famous for his calm guidance of driver Felipe Massa—operates on a daily cadence. Smedley’s primary function is to scan Garcia Valenzuela’s direct messages with his immediate reports. The agent is programmed to identify missed opportunities for recognition and to surface suggested rewrites of sent or drafted messages. Crucially, these rewrites are not intended to replace Garcia Valenzuela’s unique voice, but rather to demonstrate what a more empathetic or "warmer" version of the same information might look like, providing a constant benchmark for improvement.

The second agent, "Brundle"—named after the analytical F1 commentator Martin Brundle—functions on a fortnightly basis. While Smedley focuses on the "micro" interactions of daily DMs, Brundle analyzes the "macro" trends across a broader canvas. This includes peer threads, public channels, and skip-level conversations. Brundle is designed to detect subtle behavioral patterns that are often invisible to the leader in the moment. For instance, the agent analyzes whether Garcia Valenzuela’s tone shifts based on the seniority of his audience, whether his level of warmth significantly drops toward the end of the workweek, or if his attempts at reassurance lack the concrete details necessary to make them effective.

The efficacy of these agents is rooted in a highly structured "signal taxonomy" contained within their prompts. Rather than providing the AI with vague instructions, Garcia Valenzuela developed a rigorous framework for what constitutes "good" communication. This taxonomy draws from established academic and professional theories, including Teresa Amabile’s "The Progress Principle," which emphasizes the importance of small wins in the workplace; Amy Edmondson’s research on psychological safety; and Kim Scott’s "Radical Candor" framework, which balances personal care with direct challenge. By embedding these world-class standards into the AI’s instructions, the system can provide pragmatic advice that aligns with high-performance leadership theories.

As the experiment progressed, several key lessons emerged regarding the impact of AI on leadership. One of the most significant findings was the realization that small, negative signals compound more rapidly than major leadership failures. Garcia Valenzuela initially expected the AI to catch major tone errors in high-stakes negotiations. Instead, the system highlighted a consistent "drift" in low-stakes interactions. This included single-word confirmations sent to individual contributors and a measurable decrease in warmth when communicating with more junior staff members. While none of these instances constituted a crisis in isolation, their cumulative effect over hundreds of messages threatened to make team members feel seen or overlooked based on their position in the hierarchy.

The project also underscored the difficulty of defining communication standards. In the early stages of the "pit wall" deployment, the AI occasionally flagged messages that were contextually appropriate but appeared harsh in isolation. For example, a professional escalation between peers was initially misidentified as "venting." This forced Garcia Valenzuela to further refine the AI’s instructions, requiring him to articulate specific definitions for recognition, the necessity of tone variation by seniority, and the distinction between a one-off message and a problematic pattern. This process of "defining what good looks like" became a valuable exercise in its own right, forcing a level of leadership clarity that is rarely documented in such detail.

Perhaps the most unexpected outcome of the experiment was the "observer effect." Garcia Valenzuela noted that the mere knowledge that his messages would be audited by his AI agents began to change his real-time behavior. He found himself pausing to reflect on the tone and content of his messages before hitting send, effectively internalizing the patterns that the AI had surfaced in previous reports. To further capitalize on this, he added a pre-send module to his "pit crew" that provides on-demand editorial assistance, offering a real-time "mirror" for his communication style across different time zones and contexts.

The broader implications of this experiment are significant for the enterprise sector. Data from Atlassian’s Teamwork Lab, published in the "State of Teams 2026" research, reveals a stark disparity in AI adoption. While 89% of executives report that AI has increased the speed of work, only 6% can identify a clear, organization-wide return on investment (ROI). This gap suggests that most organizations are using AI primarily to optimize for velocity rather than for the quality of team dynamics or leadership effectiveness.

Garcia Valenzuela’s work suggests that the highest-value application of AI in the workplace may not be the automation of tasks, but the enhancement of human judgment and empathy. By using AI to surface the "blind spots" of leadership—such as forgotten recognition or inconsistent tone—leaders can maintain the connective tissue of their organizations even as they scale. The "pit wall" system demonstrates that AI can serve as a powerful tool for self-awareness, providing leaders with the data they need to ensure that their teams feel seen and supported in an increasingly busy and distributed corporate environment.

Ultimately, the experiment remains ongoing. Garcia Valenzuela is currently developing additional layers for his AI agents, including the ability to factor in organizational changes, employee tenure, and real-time nudges. The project reinforces the idea that improving leadership is a continuous process rather than a destination. As Garcia Valenzuela concludes, the goal is not to achieve a perfect system, but to establish a mechanism that forces a leader to "start looking" at the impact of their own words. Through this AI-assisted reflection, the gap between being a "good communicator" in theory and being an empathetic leader in practice can begin to close.