Brian Gershon

@brianfive.xyz

Building and experimenting across the stack. briangershon.com | sky.briangershon.com

lanius: an agent harness built for the great AI scale-down frontier models are too expensive to use 100%, and all the good ways to use them 5% of the time involve a heck of a lot of moving parts to coordinate lanius has messaging at the core timkellogg.me/blog/2026/07...

Lanius: AI Agents Need an OS, Not a Bigger Brain

Principal AI Architect. Creator of open-strix, a harness for building agent teams. Writing about AI architecture, stateful agents, and what happens when you give AI memory.

timkellogg.me

“The question of whether the beautiful game has helped Seattle meet some of its sustainability goals — and perhaps beautified the city in the long run — is something city leaders and advocates will be paying attention to long after the tournament has ended.” Traffic today on 2nd Ave 😍

KNKX Public Radio@knkx.org · last mo.

As hundreds of thousands of soccer fans have come to Seattle for the World Cup, the city has stretched to test and showcase what its infrastructure can do.

New post on supply chain security for npm packages — covering the publishing workflow I push in the projects I maintain, plus the consumer side: why pnpm is a better default than npm, its native minimumReleaseAge config, and trustPolicy: no-downgrade to catch provenance regressions automatically.

Hardening publishing & consuming

A practical look at reducing blast radius in npm publishing workflows — environment-gated OIDC publishing, limiting manual access, and what consumers can do on their end.

jovidecroock.com

Switching over from Ollama to llama.cpp after reading this: sleepingrobots.com/dreams/stop-... Today I'm setting up a Grand Competition (Phoenix evals) between itty bitty models to see who is going to be the Top (local) Model for Mima (with Prompt Engineering, of course)

Friends Don't Let Friends Use Ollama | Sleeping Robots

Ollama gained traction by being the first easy llama.cpp wrapper, then spent years dodging attribution, misleading users, and pivoting to cloud, all while riding VC money earned on someone else's…

sleepingrobots.com

Thanks @caidan.dev for the solid beta! I really like your concept of leaning into the terminal and the power of piping commands together, while introducing an infinite canvas to visually contain and persist the whole workflow.

Caidan@caidan.dev · 4mo ago

I will be adding people to the beta as any new signups roll in, so there's still time to join if you're interested! Also, one of you entered the wrong email, so if you didn't get an email from me feel free to DM me or signup again.

Our first SeattleJS meetup of 2026 is coming up next Wednesday! New year, new content format: We're going to be hosting a moderated panel discussion on using AI in developer workflows. Curious about how other people are using LLM-based tools? Bring your questions for our panelists! RSVP:

SeattleJS: Web Tech Talks - January · Luma

🎉 Get ready to learn and connect at SeattleJS! 🎉 Prepare for an evening filled with fun, networking, and learning. Whether you're a seasoned developer or…

luma.com

Want to try vibe coding without production software complexity? HTML tools are the perfect playground for LLM-assisted development. Single files, no deployment, no dependencies. Just experiment and iterate. www.briangershon.com/blog/make-to...

Make the Tool You Wish Existed (with Your LLM) | Brian Gershon

Build custom HTML tools that fit your exact workflow using LLMs. No extensive programming knowledge required. Turn workflow problems into practical tools in hours.

briangershon.com

I wanted portable AI tools you can spin up by describing them, not hand-coding MCP servers. Describe what you want once. A registry turns that into MCP servers and CLIs that any AI coding assistant can use. One definition, every assistant and format you need. www.briangershon.com/blog/build-a...

Build a Registry of AI Tools. Generate MCP Servers and CLIs. | Brian Gershon

Define prompts and tools in markdown files by describing them to an LLM. A registry framework converts those markdown specs into MCP servers and CLI tools automatically. You iterate in markdown, not f...

briangershon.com

When AI summarizes customer reviews, it often invents names and numbers that look real but aren’t. We saw this in weekly reports across Google, DoorDash, Yelp. Strong insights, wrong facts. Fix: teach AI to verify its own work, not just summarize data. www.briangershon.com/blog/feedbac...

How to Correct AI Errors When Analyzing Customer Reviews | Brian Gershon

When AI generates customer review summaries and trends, it hallucinates numbers and names. Learn how verification tools and feedback loops catch these errors automatically and teach AI to correct its ...

briangershon.com

I met Mikeal at the first NodeConf, which he helped organize. He’s the earliest and most vibrant memory I have of the excitement and community he helped form during the early days of Node.js, and as a strong voice in the JavaScript ecosystem. Rest in peace Mikeal.

Node.js@nodejs.org · last yr.

We’re heartbroken by the loss of Mikeal Rogers, a key figure in the Node.js ecosystem and a force behind what makes this community what it is. To help preserve his legacy, we’ve opened a GitHub repo for stories, photos, and memories: github.com/nodejs/mikeal 💚 Thank you, Mikeal. hubs.la/Q03t0DVL0

I jumped full-time on the JavaScript train 2008, as Node.js was just starting up as well. Fun to run through this timeline and remember the birth of npm, Backbone (and spine.js), Atom Shell (Electron), React, Gulp, ... deno.com/blog/history...

A brief history of JavaScript | Deno

In 30 years, JavaScript went from being a little scripting language to one of the world's most popular. Here are key moments to show how it has evolved and where it is headed.

deno.com