Refocusing the lens of workforce intelligence with a more collaborative approach to performance management

Traditional performance reviews are fundamentally misaligned with agile workflows. Built on subjective memory and cadence-bound check-ins, they act as lagging indicators that assess past behavior rather than accelerating current momentum. Meanwhile, legacy management tools focus on administrative compliance and invasive activity tracking rather than capturing authentic, day-to-day contribution. The same tool that helps a manager can, in the wrong hands, become a surveillance product.

We initially targeted clinical healthcare environments to test whether dynamic reward structures could surface organizational momentum. Clinical practices balance high-stakes care delivery with measurable business drivers (e.g., patient readmission rates, preventive care adoption, and client retention). I participated in interviews with three healthcare practice owners across veterinary and hospital systems. These sessions revealed a critical operational breakdown: clinical performance data lived inside fragmented software and relied on stale, self-reported, manual file uploads. The absence of continuous, automated data streams meant the system could not generate reliable, dynamic feedback without introducing prohibitive administrative friction.

We shifted toward software engineering, exploring GitHub commits, pull requests, and review threads alongside Jira tickets and sprint histories. Together, these sources offered a way to understand both the work being done and the priorities behind it. I helped shape how this evidence, combined with context people chose to share, could support team review handoffs and 1:1 manager-employee coaching conversations.

Through story-mapping workshops, we deconstructed the traditional performance review cycle: how HR administrators, managers, and individual contributors set expectations, gather feedback, and turn evaluations into development plans. We mapped tasks, goals, data needs, and participants’ perspectives and emotions to examine where context gets lost and manual work accumulates. This helped us establish ethical boundaries and explore where AI could support employees and more useful conversations before deciding on models or integrations.

View full size ↗

Close ×

Mapping the end-to-end human process first to expose the ethical risks of performance management, establish our core evaluation guardrails, and isolate the highest-signal workflow for rapid validation.

During one sprint, we explored whether data from GitHub and Jira could help us understand how teams work together and where engineers need support. Three questions guided that exploration:

The contributor recognition exercise

We later narrowed the initial prototype to GitHub repository artifacts to reduce integration overhead and test the core hypothesis. To explore whether an LLM could extract authentic qualitative insights from that data, we ran a blind consensus exercise with active open-source contributors. We generated anonymized team narratives and individual developer profiles, challenging contributors to identify their own repositories and teammates based solely on the AI-generated descriptions.

Contributors readily recognized their team's specific working patterns, communication bottlenecks, and review habits. This confirmed our core hypothesis: ambient repository data holds genuine organizational signal.

Close ×

Team Output · Original contributor exercise report

Close ×

Guess the Dev · Original contributor exercise report

Crucially, this exercise also revealed the product's primary ethical boundary. While the narratives were accurate, seeing human labor synthesized into passive profiles highlighted how easily automated summaries can feel like surveillance. It proved that accuracy alone is insufficient: for AI evaluations to be trusted, the interface had to move away from static reports and give employees inspectable evidence, context annotations, and active agency in shaping the conversation.

Traditional workplace intelligence tools are built for the manager’s gaze—auditing velocity and aggregating data into top-down reports. Our earliest concepts followed this default with punitive mechanics: decimal scores, traffic-light gauges, and stack-ranking curves that stamped engineers with fixed “weaknesses”. I pushed hard against moving forward with these surveillance-heavy designs, advocating instead for an equitable, growth-oriented model.

I inverted the perspective to center the individual contributor. The resulting Self-Directed Growth Canvas gives engineers a private space to curate which high-leverage projects to surface in 1:1s or use in performance evaluations. Managers stay organically aware of cross-team impact without requiring tedious self-reporting, while artifact-grounded AI coaching helps engineers turn day-to-day work into ongoing career growth.

Revised directionEarly concept

Close ×

Revised direction: A shared starting point

Built for engineer agency, letting developers curate work evidence, add missing context, and use private AI coaching to shape their own career conversations.

By validating our core hypothesis against open-source repositories and deconstructing traditional evaluation workflows, we replaced early stack-ranking concepts with an inspectable, human-centered experience. My engagement concluded with creating high-level screens to anchor the overall product vision in how future enterprise teams can ground growth in transparent work evidence, with actionable insights between managers and engineers.