I design the architecture behind AI agent systems in React and TypeScript: how much autonomy an agent gets versus how much it's constrained, where execution boundaries sits, how tool access is scoped, and what guardrails hold when an agent tries something it shouldn't. The code is rarely the hard part. The hard part is the tradeoffs: reliability versus latency, isolation versus performance, flexibility versus a surface you can actually secure. Most of what I write comes directly out of those decisions, not from tutorials.

Specular Audits a shipped product against its own design token spec by reading what the browser actually rendered, not what the code claims it should look like. Core engine is done; adapters for public design systems (GitHub, IBM Carbon) are in progress.

Openbundle Openbundle is a real, working JavaScript bundler, plus an interactive website that visualizes its own real internal analysis.

the-human-touch A Claude skill that catalogs the recurring habits that make writing read as AI-generated, with a working checker script for fourteen of them. Built from editing enough of my own drafts to notice the patterns.

and more to come ...

150+ articles on frontend systems, performance, and the parts of the job that don't show up in tutorials, published through Level Up Coding, gitconnected, and JavaScript in Plain English.

→ medium.com/@akshatmtiwari