Eric J Ma's Website

In this blog post, I reflect on how AI tools have accelerated my coding workflow but also made it more exhausting by collapsing my attention span and pushing me to juggle too many tasks at once. I explore why our brains are now the bottleneck, and share strategies for calibrating our work pace to match our neural capacity for sustainable productivity. How can we harness AI's power without burning out or losing focus?

I vibe-coded canvas-chat into existence in 48 hours, then spent weeks untangling the mess into a clean plugin architecture. The AI could execute my architectural vision, but it couldn't design it. This is the story of how I recovered from AI-generated slop — and why the architecture stays human.

In this blog post, I share practical ways to grow as a mentor and leader even when budgets are tight. Drawing from my experiences at Novartis and Moderna, I discuss strategies like one-on-one coaching, presenting at internal events, organizing communities of practice, and hosting informal gatherings—all without extra spending. I also offer advice for managers on supporting these efforts. Want to discover how you can foster career growth and mentorship, no matter the economic climate?

In this blog post, I explore the concept of "air gaps"—those manual steps in business or scientific processes where humans bridge the gap between digital systems. I share real-world examples from labs and software workflows, discuss why these gaps matter, and offer practical advice on identifying and closing them with automation and coding agents. Curious how closing even small air gaps can transform your team's efficiency and free up mental bandwidth?

Workflow-specific agent skills don't just automate tasks — they encode how you work, down to your tools, your file structure, and your philosophy. I explore this through two examples: a daily sign-off skill that inherits my Obsidian setup and bullet journal structure, and a scientific EDA skill that goes further, encoding a whole epistemology of how analysis should proceed. I argue there are three layers of implicit assumptions in any workflow skill — tool dependencies, organizational preferences, and epistemic preferences — and that the last one is the hardest to see and the most important to document.

I spent a weekend trying to make PyMC installable in WebAssembly environments via Pyodide. The journey involved making Numba optional, setting up Pixi development environments, and documenting the WASM build process. While PyMC can now technically install in WASM, the lack of WASM support in MCMC sampling backends (JAX, nutpie) means NUTS sampling remains out of reach. This represents a fundamental infrastructure gap, not just a missing dependency. Note: the Pixi-based approach described here was my weekend exploration; the actual PR to PyTensor respected their existing mamba-based toolchain.

I share how I use Obsidian and AI coding agents to manage personal knowledge at work. By choosing plain text, building structured note types, and encoding workflows into agent skills, I reduced knowledge management overhead from 30-40% of my time to less than 10%. The system helps me manage twelve people across two teams without losing context. This is an invitation to experiment with your own PKM system.

In this blog post, I share how coding agents can supercharge data analysis, but only if we stay in control. By slowing down, asking the right questions, and structuring sessions with journals and artifact gating, we avoid chaos and keep our scientific thinking sharp. I explain the skills and patterns that help teams focus on insights rather than just generating code. Curious how you can harness agent speed without losing your scientific edge?

In this blog post, I share ten lessons I've learned from experimenting with agentic coding in data science, from setting clear goals and structuring projects to leveraging coding agents for faster iterations and better insights. I discuss practical tips like maintaining logs, generating diagnostic plots, and treating the agent as a partner in exploration. Curious how you can make AI your jazz partner in data science and boost your productivity?

In this blog post, I share my hands-on experience using AI coding models, focusing less on benchmarks and more on the day-to-day feel—how model style, personality, and the right testing harness impact productivity and flow. I discuss the trade-offs between long-horizon autonomy and short-horizon iteration, and why a constructive, enthusiastic AI assistant matters as much as raw performance. Curious how the right mix of model and harness can transform your coding workflow?