hyperbola :: blog

An agent is a parameterized program over a set of capabilities.

Better models will not obsolete the need to align their priors with what good work looks like for you.

Skills get privileged instruction-following fidelity through post-training and can straightjacket reasoning agents.

My homelab combines bespoke static verifiers, durable operating docs, and thin weekly automations to keep agent work coherent.

Dashboards and SaaS products encode scarce expert leverage into shared artifacts.

Code reds are expensive adaptive episodes.

CEO token mandates aren’t about measuring AI usage.

Loki’s Tailscale upgrades used to threaten the access path they depended on.

Verification was always the problem.

Let technical non-engineers go for it with coding agents.

A one-line red-team prompt sent Codex through the full security workflow on `intaglio`: prove impact, land the fix, prep the release, and publish a RustSec advisory.

With agents, a lint that finds 600 violations can ship with fixes and exhaustive coverage in one PR.

Full Japanese support for an alpha app would lose a roadmap fight.

Symphony is distributed spec-first: a coding agent can rebuild it in another language from `SPEC.md`.

Reasoning models scale with tokens.

@frantic asked for RSS, so I harness engineered the blog build (again): Vite-native SSR and assets, MDX posts, typed React composition, and static output that still deploys to GitHub Pages.

After six years of building a Ruby implementation in Rust, it’s time to archive Artichoke.

I’d just like to interject for a moment.

Ruby’s Enumerable mixin set the gold standard for elegant, composable iteration nearly three decades ago.

Coding agents fail on tools they can’t see.

What is the purpose of a blog?

Service meshes are less about tech and more about enforcing org-wide defaults without slowing teams down.

My résumé, written in LaTeX, is excessively over-engineered and 100% the way I want it

To build iteratively and shed complexity, do the simplest thing that could possibly work.

As a Staff+ engineer, I’ve honed my focus by distinguishing projects where failure is acceptable, creating space for my teams to learn and grow, versus where it is not, requiring more direct involvement.

It is not the case that all proposals are open to all feedback from all stakeholders at every point of their lifecycle.

Pair programming with ChatGPT on Artichoke Ruby, writing a lot of Rust.

Pernosco is a hosted omniscient debugger that makes debugging complicated bugs tractable.

Interface compatibility does not require implementing an interface or a trait.

Help your stakeholders by doing the hard work of distilling why the information you're sharing is important.

🌵 CactusRef lets you build cyclic data structures using strong references and knows how to deallocate unreachable cycles.

The border between Baarle-Hertog 🇧🇪 and Baarle-Nassau 🇳🇱 is one of the most complicated international borders in the world.

What I did this sprint: Less raw code output, more broad impact.

To be an effective senior engineer, you need to convince others to help you execute.

To learn Rust, I implemented an audio game.

I contributed to open source more than I ever have in 2018; but there is a lot I can do to improve.

Terraform misconfiguration of SSM PrivateLink endpoint completely brings down hyperbo.la.

Moving blue-green deployments into terraform-managed autoscaling groups makes rolling out new AMIs easier.

A successful partnership between Engineering and Finance creates better business outcomes.

Your AWS infrastructure is designed by your org chart.