Shorten the feedback loop
I use models and agents to accelerate exploration, implementation, testing, review, and documentation when they improve the feedback loop.
Engineering practice
AI accelerates the work. Engineering judgment owns the outcome.
I use AI deliberately to examine more possibilities and shorten feedback loops. It is not a substitute for understanding the system, making the hard decision, or being accountable for what reaches production.
My working model
Better models expand what an engineer can attempt. They do not remove the need for clear boundaries, domain knowledge, verification, or operational discipline.
Coding remains part of my routine. I write the code for my contributions to the Rust compiler, Deno, and rust-analyzer, and I design and maintain Bitarena. This ongoing practice keeps my judgment current through implementation, tests, performance measurement, and maintainer review.
Keep judgment accountable
Architecture, data boundaries, security, performance, and failure behavior still need decisions that can be explained and defended without appealing to the model.
Match complexity to the problem
A prompt, retrieval system, deterministic workflow, or agent are different tools. I prefer the simplest one that satisfies the actual product requirement.
Build verification into the loop
Tests, types, benchmarks, evaluations, reviewable diffs, permissions, and production telemetry turn generated output into evidence an engineer can trust.
Evidence in practice
The operating model is grounded in shipped systems
These principles come from work where abstractions meet operational constraints, unfamiliar codebases, and product consequences.
Foundational product engineering
At Inferal, I work across the core engine, ontology system, and Relay—the data-synchronization capability I built from the ground up.
Inferal experience →Systems that have to operate
I founded MonitorMe and carried the open-source observability framework into several client environments through installation and configuration.
MonitorMe case study →Hands-on systems depth
Financial and healthcare systems, ten merged pull requests to rust-lang/rust, contributions to Deno and rust-analyzer, and the Bitarena crate show the underlying engineering range.
Open-source evidence →Technical writing
How I reason about AI systems
Choosing a simpler, more predictable system when autonomy adds no real value.
Matching the architecture to the product problem, operating cost, and risk.
Treating tool execution as a production dependency with explicit failure behavior.
Where this combination matters
I am interested in teams where AI supports a real product or engineering workflow. That includes systems-heavy products, agent infrastructure, developer tools, data platforms, and consequential backend or platform work.
Engineering teams
Roles where systems depth, product judgment, and AI-assisted execution can reinforce one another over the life of a product.
Founders and CTOs
Focused work where the challenge is choosing the right architecture, establishing a foundation, or making an AI-enabled production path dependable.