Started in 2011 as my outlet to share my experiences on personal life and work topics. My handle is khanmjk. My friends & colleagues call me "Mo". This blog is an experiment, work-in-progress.
Life: Self-awareness, personal development, growth hacking (quantified self).
Work: Software and Systems Engineering, Leadership, Management, New Ideas and OpEds.
So I had to test drive the new Firebase Studio that's powered by the latest Gemini 2.5 model, that has got the whole world talking, and leading the webdev llm arena.
For my test, I simply prompted it to build a game like the classic battleship, where the player plays against the computer. The game didn't work on first try, what industry refers to as "one-shot" builds - but it had a lot of potential, so I guided the assistant through addressing bugs, improving user experience of the game layout, and also tweaking the AI rules. Gemini was smart enough to use GenAI itself as the computer opponent - which I think was pretty neat. Historically, before AI, we would have needed to code the computer's moves, strategy and keep track of moves, etc. Instead, we initiate an AI agent, task it to play as the opponent in the game, and provide just enough context and guidelines for moves (like don't place a move on a previous move).
A walkthrough of an app I decided to build to help me with my work as a software engineering leader. Core thesis: I wanted a planning management tool that is centered around core system architecture, working backwards from there to model the organizational / team structures that feeds the planning process. I found most project management tools lacking support of including software architecture inputs - so I decided to build something to close this gap.
My journey started in late 2024 (October), when LLM models started exposing coding capabilities. I wasn't too impressed by the likes of ChatGPT and Claude (at the time). But with Gemini 2.5 entering the scene, I picked this project up again in April 2025 - and slowly built out the MVP, first as a single page application (html/javascript) with basic old skool UI (this video), then refreshed the UI, refactored the code (next video "Building apps with AI (2) - MVP...". Since I stopped coding professionally in 2010, this entire journey was quite the learning experience. I'm going back to school again, but I've learnt so much about how to engage with the LLM, that I'm increasingly confident I will find a place in the new world of AI. AI-assisted coding is definitely the future, but it is not that straightforward yet, even with the latest craze in vibe coding.
Just take a look at my document sharing my conversation with Gemini here.
50 pages and counting!!! Essentially you need to be very clear with articulating the requirements and interacting with the model as a true assistant, like you would with any human engineer, business analyst, or UX designer.
Listen to the podcast generated by NotebookLM here You can interact with my Google NotebookLM here
The rest of this blog post was written by partner, Gemini >>>>
From Vision to MVP: Building a Software Management & Planning Powerhouse (with an AI Co-Pilot!)
For any software engineering leader, manager, or even a senior individual contributor, the landscape of modern software development is a complex beast. We juggle intricate system architectures, dynamic team structures, ever-shifting priorities, and the constant challenge of aligning finite resources with ambitious goals. What if there was a tool that not only helped model this complexity but also empowered smarter, data-driven planning?
That's the question that sparked the creation of the Software Management & Planning Tool, an MVP we've been passionately building. This isn't just about another planning application; it's about a foundational belief: effective planning starts by working backwards from your systems and organizational architecture.
This blog post shares the journey of bringing this vision to life – from a core idea to a feature-rich Minimum Viable Product – and the unique experience of co-developing it with a generative AI coding partner, Gemini.
The Core Idea: Architecture-First Planning
The tool is built upon a hierarchical understanding of a software organization:
Systems: The overarching products or platforms.
Services & APIs: The functional components within those systems, each with defined interfaces.
Teams (2-Pizza Teams): Agile squads owning services, each with software development managers (SDMs) and engineers.
Dependencies: Critical platform components and inter-service relationships.
This rich data model allows the application to provide a holistic view, forming the bedrock for all subsequent planning and visualization.
Our MVP Journey: An Iterative Evolution with an AI Partner
What started as a concept rapidly took shape, built incrementally feature by feature. The initial scaffolding was a single HTML/JavaScript file – a testament to getting ideas off the ground quickly. But the vision was always much larger.
Key Milestones in Our Development:
System & Org Modeling: The first step was enabling the definition of systems, services, APIs, teams, and their intricate relationships. This involved creating UI elements for adding, editing, and connecting these entities.
Visualizing Complexity: Static data isn't enough. We introduced dynamic visualizations:
System Architecture Diagrams: Showing services, APIs, and platform dependencies.
Team Relationship Graphs: Illustrating how teams connect based on service ownership.
Service Dependency Views: Drilling down into specific service connections.
Interactive Org Charts & Team Breakdowns: Providing clear views of the organizational structure and team compositions.
Initially, these were separate views, but based on user experience feedback, we evolved the system overview into a slick carousel navigation allowing users to seamlessly glide through these critical perspectives.
The Game Changer: Yearly Planning: This became a cornerstone. The goal was an interactive table to manage initiatives against team capacity, showing what's "Above The Line" (ATL) and "Below The Line" (BTL). We implemented:
Dynamic team columns and SDE Year estimations.
Drag-and-drop prioritization.
"Protected" initiatives that lock to the top.
The crucial concepts of "Funded Headcount" vs. "Builders In Seats" (BIS).
Introducing "Away Teams": Recognizing that real-world capacity isn't static, we enhanced the model to include "Away Team Members" – borrowed resources. This led to the "Effective BIS" metric, providing a more realistic picture of available capacity.
Deep Dive into Capacity Tuning: This was a pivotal enhancement. We built a dedicated "Tune Capacity Constraints" module where users can define:
Global defaults (working days, holidays, org-wide events).
Team-specific adjustments (leave uptake, variable leave, team activities, recurring overhead).
The output? A Calculated Net Project Capacity for each team, which can then feed back into the Yearly Planning view for a much more accurate ATL/BTL assessment. This module also includes a detailed narrative explanation and a waterfall chart to visualize capacity erosion.
SDM Resource Forecasting: Leveraging the detailed capacity tuning, we integrated a forecasting model. SDMs can now select their team, input hiring and ramp-up estimates, and see a 52-week projection of their headcount and effective SDE availability, all adjusted by the previously configured capacity constraints.
UI/UX Modernization: More recently, we've focused on enhancing the user experience, refactoring the codebase into modular files, and adopting a cleaner, card-based UI reminiscent of modern web applications.
The Gemini Co-Pilot Experience:
This journey wasn't a solo flight. As the human visionary and lead developer, I partnered extensively with Gemini, my AI coding assistant. This collaboration was an experiment in itself:
Acceleration: Gemini was instrumental in rapidly generating boilerplate code, implementing UI components based on descriptions, and refactoring complex functions. This significantly sped up the development of the MVP.
Brainstorming & Ideation: We bounced ideas for UI/UX improvements, data model structures, and feature refinements (like the Team Load Summary table in the planning view).
The Learning Curve (for both of us!): It wasn't always smooth sailing. There were moments where context was lost between sessions, leading to regressions or suggestions that didn't quite fit the current state. Clarifying requirements meticulously, sometimes repeatedly, became key. Debugging involved a human-AI partnership – I'd often describe the unexpected behavior, and Gemini would help pinpoint potential issues in the code it (or we) had written. It was a process of learning how to "prompt" effectively and how to guide the AI when it drifted.
This experience has been a powerful demonstration of how GenAI can be a force multiplier for developers, turning ambitious ideas into working software at an accelerated pace, even if it requires a new kind of collaborative patience and precision.
Key Features That Shine
While the tool is packed with interconnected features, some of the "wow" factors that have emerged include:
Holistic System Modeling: The ability to define not just services, but also the teams that own them, the engineers within those teams (including AI SWEs and their attributes like skills and experience!), and the management structure, all in one place.
Insightful Visualizations: The carousel of system, team, and service dependency graphs provides powerful, at-a-glance insights into complex relationships.
Realistic Capacity Tuning: Moving beyond simple headcount, the "Tune Capacity Constraints" feature allows for granular adjustments, factoring in everything from public holidays and company events to team-specific leave uptake and meeting overhead. The resulting "Net Project Capacity" and its narrative explanation offer a truer picture of what teams can achieve.
Interactive Yearly Planning: The drag-and-drop planning table, combined with dynamic ATL/BTL calculations based on selectable scenarios (Funded HC, Team BIS, Effective BIS) and the option to apply the "Net Capacity Constraints," makes for a powerful "what-if" analysis tool. The Team Load Summary provides immediate feedback on team-level impacts.
Integrated SDM Resource Forecasting: This module closes the loop by allowing managers to see how their hiring plans and attrition realities, combined with the tuned capacity constraints, will affect their team's SDE availability throughout the year.
An Evolving, Modern UI: The recent shift towards a cleaner, card-based design with a persistent top navigation bar significantly improves usability.
The Journey Continues: The Power of Iteration
This MVP is a significant milestone, but it's also a foundation. The backlog, as detailed in our project's README.md, is rich with possibilities: enhanced yearly planning with ROI tracking and versioning, a dedicated roadmap and backlog management module, deeper detailed planning with work packages, and, most excitingly, integrating more AI-powered enhancements for analysis and interaction.
The development of this tool underscores the power of iterative development and the transformative potential of collaborating with AI coding assistants. It shows how a complex vision can be broken down and built, piece by piece, learning and refining along the way. While the path of AI-assisted development has its unique challenges (like ensuring context retention and guiding the AI through ambiguities), the ability to rapidly prototype, generate code, and explore solutions is undeniable.
For any engineering leader wrestling with the complexities of software planning, or any developer curious about the potential of AI partnerships, we hope this journey provides some inspiration. With the right vision and the right tools (both human and AI!), ambitious ideas can indeed be brought to life.
