Location: Remote (US)

Remote: Yes

Willing to relocate: No

Technologies: Python, Go, TypeScript, Terraform, Kubernetes, GCP, AWS, GitHub Actions, ArgoCD, Dagster, LLMs/AI tooling (Pydantic AI, Langfuse, Qdrant)

Résumé/CV: https://blog.jamesolds.me/james-olds-resume.pdf

Email: james + hn @ oldsmail . com

GitHub: https://github.com/oldsj

Blog: https://blog.jamesolds.me/

---

Full Stack / Platform Engineer, previously at Trail of Bits.

Recent focus: Designed entire K8s infrastructure with secure best practices and Tailscale for intranet connectivity to self hosted services - supporting the team that placed 2nd at AIxCC DEF CON 2025. Built out multiple full-stack apps with Svelte, Tailwind CSS, Typescript, Python.

Currently building https://github.com/oldsj/mainloop so that I can more quickly build https://understanding.news, a reasoning layer for political news and comments.

Looking for: Full-stack AI native product development roles or Platform Engineering roles. Full time or contract

I like the idea. As a solo dev I've experimented with creating Claude subagents for multiple perspectives for "team leads" and will run ideas through them (in parallel). The subagents are just simple markdown explaining the various perspectives that are usually in contention when designing stuff. And a 'decider' that gives me an executive summary.

  agents/
    |-- customer-expert.md - validates problem assumptions, customer reality
    |-- design-lead.md - shapes solution concepts, ensures UX quality
    |-- growth-expert.md - competitive landscape, positioning, distribution
    |-- technical-expert.md - assesses feasibility, identifies technical risks
    |-- decider-advisor.md - synthesizes perspectives, executive analysis

I've experimented with something similar - my flow is to have the subagents "initialize" a persona for the task at hand, and then have the main thread simulate a debate between the personas. Not sure if it's the best approach but it's helpful to get a diversity of perspectives on an issue

s/validates/guesses/

Ftfy. You might as well toss a coin.

I think we’re slightly better than random at this point

You really think asking AI to spaff out some bullshit validates anything?

I’ve been working on running agents (Claude agent sdk) on k8s this looks great to control their egress

You can certainly use the Squid ACLs to limit the egress for agents. One of the current shortcomings (I explicitly mentioned it near the end) is that there's no per-namespace granularity, so you wouldn't be able to determine it on a per-agent level -- but you would be able to generally establish that all agents would only have access to a global whitelist.

I'm just glad it can run async of my main thread now, so I can spawn other agents for more interesting things

This looks excellent and very timely. We just started hand rolling our own ETL process with Python and Github Actions to ingest from various SaaS sources to a central DB for context to LLM(s) and traditional BI tools.

Speaking of Plaid, would love to see an integration there for personal finance

We haven't gotten into finance integrations yet but would love to understand use case more and see what we can do. Happy to discuss more in our community Slack.

Also what kind of SAAS sources are currently ingesting?

> Imagine spending a lot of time and money picking out the best possible gear for a hobby before actually starting the hobby.

Are you kidding me that's my favorite part!

I think there's something to this and have been thinking about this for a while too. Just today tried to see if diff.me was available, sadly it's not.

I was thinking something like a git repo with a flat file in it. You list your beliefs / opinions / things you think are facts in a structured way (maybe a tree like structure) going all the way back to some "root". Then you can easily diff this file with someone else and see where you agree and differ. Changing someone's mind could just be a pull request :)

You can pass in the current user’s uid/gid as an environment variable check out linuxserver images for an example https://hub.docker.com/r/linuxserver/unifi-controller

It is usually better to use --user $(id -u):$(id -g) to do this. You don't really need to pass it as a variable for the most part.

The author needs to add this and then try if everything works as expected (like npm and pip and so on)

Same setup here. Love using z-wave as a standard and then bridging everything over to Homekit for siri / phone control center access

Edge is just Chromium now https://finance.yahoo.com/news/microsoft-edge-chrome-chromiu...

Which means Blink is more of a Google-Microsoft collaboration than it was before. Google’s share of Blink power went down.

Chromium is "just" a logical, security first base OSS project,

that (theoretically) any org can adopt and strap additional modules onto.

Edge adopted chromium because microsoft could not compete with their monopoly power to enforce web standards as “however it is implemented in chrome”. If chrome ships a feature, there 80-90% market share means that every other chromium browser has to ship it or they’re “broken” and people switch to chrome.

if the other chromium wrappers have a seat at the table, that isn’t the table that makes the decisions.