Brian Lovin

Absolutely insane how much progress you can make on a small side project with a well-made codebase and rigorous verification systems (tests, logging, tracing).

Yes here is my 10 minute breathless rant about why I'm so excited about Notion Workers + Custom Agents... Context: I spent this afternoon building a custom agent to help me manage Shiori (a side project I shipped last weekend). I gave the custom agent everything it needs to understand what's happening in my product (email, log drain, sentry alerts, stripe payments, etc) and to do work on my behalf (access to coding agents). In an afternoon of tinkering, this agent can: - Diagnose bug reports proactively by looking through past email conversations, system logs, and database records - Draft replies to user questions with the correct answer based on past email threads, or help me proactively reach out to churning paid users - Self-construct a database of feature requests with an understanding of who is requesting the feature and how they're using the product today - Answer any question I have about how people use the app and what I should be thinking about next - Initiate Claude Code workflows to open PRs proactively in the background when someone sends a bug report or feature request This custom agent is now my "Side Project Chief of Staff" (I don't really know what a chief of staff does but this sounds right). I didn't write a single line of the worker code because I didn't need to: models are so good that I can link to the Workers readme, yap my desired outcome into a microphone, and I get a super-personal and highly-capable AI agent out the other side. So fucking cool. The future is now! I'm excited to see what everyone makes.

I've had "Write a post about Prototype Playground" on my todo list for 3 months now...fortunately, @clairevo invited me onto her podcast to talk about how it works and dig in to how we're prototype with AI @ Notion! Watch the video: https://youtube.com/watch?v=s4HGbIhUgVo Some thoughts: Prototype Playground has evolved quite a bit over the last year, and even since recording this in December, the app looks and feels quite different. But at it's core the Playground is simple: a single repository where anyone can explore ideas with AI and real code, unconstrained by the production codebase and without fear of breaking things. Prototype Playground is AI-first, naturally. We have skills that make developing and deploying easy, and they can help teach non-technical team members what's actually happening under the hood. Some of my favorites: /brainstorm — interviews you about what you're trying to make, think through edge cases, key flows, and finds components that can be reused from existing prototypes /deploy — walks through the entire branch → commit → push → PR → CI workflow and teaches people how git works along the way /find-icon — a script that helps find the right icon instance using synonyms, so "search icon" can resolve to "MagnifyingGlassIcon" in code We also have a few subagents that are particularly helpful: ci-monitor — watches CI on a pull request and self-heals broken tests, formatting issues, type errors, etc figma-verifier — uses the Figma MCP to loop back and forth between a Figma mock and what the agent built in the browser until the output is correct One of the most valuable parts of Prototype Playground is that everyone's work is in one place. It's easy to yoink great ideas from other people's prototypes or find that one random idea that would have otherwise been lost to a Slack thread...every prototype rolls up into a unified feed so we can see at a glance what people are exploring. It's worth mentioning: models have gotten so good in the last 3 months that we're seeing more designers at Notion prototype directly in production with feature flags. In the long-run I wonder how useful this type of isolated code-first prototyping environment will actually be... Also: in the video I mentioned that I was spending ~60% of my time in Figma. That is no longer true. It's now closer to 5-10%. The last 3 months of model improvements have changed what's possible to design and prototype in code. I can now multi-task different versions of an idea with parallel agents, kick off work to a background agent from my phone during a meeting, and agents can work for longer periods of time on more complex prototypes to build the core idea at a much higher level of fidelity. It's an incredibly exciting time to be a designer. Designers should embrace AI, learn how it works (there's no magic here!), and start developing an intuition for what models can do so that we're always building product experiences on the bleeding edge of capabilities. Let me know if you have any questions! I should still write that blog post...

I tried the Figma MCP Code → Canvas update and it works pretty well! I'm not sure where this fits into my workflow, but I'll keep playing...

We’ve reached Peak Sidebar. No more please. What’s next?

Windows is crazy: Bluetooth doesn''t work so you spend an hour troubleshooting and eventually find a missing driver but installing the drivers doesn''t work so then you find a Reddit comment from 2022 saying to open the drivers with WinRAR and then Bluetooth magically works.

Pretty cool: - tried https://detail.dev - pointed it at Shiori repo - found some bugs (and a few false positives) - converted them all to github issues - pointed @conductor_build at all the issues - all bugs fixed/patched in minutes in parallel worktrees

Claude Code throwing 500s this morning. Tried Codex. Codex is good!

The Sentry MCP is great! I fixed some issues with Shiori this morning and turned the process into a skill here: https://github.com/brianlovin/claude-con… Now I can use `/fix-sentry-issues` and go get coffee.

It’s so silly that this works, but alas: after plan mode, ask “what are the edge cases you didn’t consider?”

My first impressions with Omarchy (and generally, my first time running Linux at home): brianlovin.com Omarchy first impressions From brianlovin.com

Claude skills/commands/subagents composition is important: - create a generic skill like "nano-banana" with a script to use the Gemini API for image gen - create a generic skill like "invert-image" that can create dark mode variants of any image (for example) - create a generic skill like "upload-image-to-notion" that has access to the API - create a generic skill like "write-guide-from-codebase" that can analyze a part of your codebase and write a tutorial for how it works — can include visualizations, diagrams, code snippets, etc. (bonus points: connect it to Notion MCP or wherever you host your docs so it can publish the guide automatically) Then compose all of these into a subagent like: `guide-writer` — the subagent knows how to analyze the codebase, write the guide, come up with the best prompt for an image, gen the image, make the dark mode variant, create a Notion page for the guide, and upload the images to the guide. The subagent works and keeps the main context thread clean. You only supply the initial prompt for the guide you want to write, and the main agent will respond when it's done with a link to the guide. The important thing is because it's all well-composed, you can use the skills individually at any point, or easily create new subagents that use the same skills for entirely different workflows.

The thing that finally made Shiori work for me was an iOS Shortcut to quick-capture on mobile. The installation flow sucks a bit, though, so I wrote this guide: https://shiori.sh/docs/ios-shortcut I have Claude working nights and weekends to get a mobile app going in the background, which of course is the better long term solution 😅 One thing at a time...

Damn, 17,000 bookmarks added to Shiori in the last 48 hours... Let's just say uh, I did not rate limit API access or AI summarization. So yes, this is why I am charging $5 (and losing money) 😆

My latest flights-n-weekends project — I made the first version of this last spring and it's the first time a read-it-later app has stuck for me long-term. I hope you find it useful 🔖 https://shiori.sh

If you work on computers all day, the least you can do for yourself is learn how to navigate between things quickly. Use the ✦ Hyper key! https://brianlovin.com/writing/use-the-h…

Agent teams seem like the bigger announcement of the day. Testing now.