Scott Watermasysk

Scott Watermasysk

@scottw.com

Husband, Father, & purveyor of fine Ruby on Rails apps. Always building something.

916 followers225 following760 posts

I wanted a really lightweight agent to-do list. Just something for me and an army (or two) of agents. Linear was too heavy. I'm never messing with something like Beads again. You can just build things.

Vendor lock-in is much easier to escape with AI. In the past, picking the wrong dependency could take months to undo (or kill the product/company). Today, a single developer can move an app from MySQL to PostgreSQL in a day (add a few for deep testing).

In an odd twist, I am even reconsidering my usage of the interview-me skill. The models are so good at being thorough that when pushed to go deep, I keep finding they go way too deep.

We have all forgotten the therapeutic nature of blogging. It helps you organize your own thoughts. Sharing knowledge with others is just a nice side effect.

I love how proactive agents have gotten in showing their work. Completely unprompted, Opus 5.5 shows up with a screenshot of what it implemented.

I can't remember the last time I stopped to write a support ticket for a bug. It is now far easier to just fix the bug. It typically takes the same amount of work, and fixing it is far better for your customers.

One year ago, this would have been a lousy Saturday for a small business owner. Today, I made a sandwich why all this happened. It is now monitoring the situation and I get to do some cleaning (that isn't a huge win 😀)

Set up T3Code for local Laravel and Rails work this week. My guess is 70% on orbs and the other 30% local. I love Herdr, but I don't see the benefit of being tied to the terminal.

Engineering is all about the setup and validation. The middle (coding) is solved. The more you can offload to robots, the more likely you will be successful.

An underappreciated Claude tool for validation is artifacts. Have it compare before and after results and generate a visualization. These are also easy to share and collect feedback from the team, client, etc.

I set up a webhook in @ampcode's Orbs to receive new Sentry errors, and now I wake up to PRs ready to review. Even better, it's smart enough to know when a fix isn't necessary.

I find Astra with the typical `implement`-type skills to be a disaster. It completely overcomplicates things. I was working on a streamlined bin/setup script for a project, and it created five sub-scripts and 32 tests. Simply prompting it has yielded much better results.

I temporarily set up a new Windows computer for the first time since Vista. So many up sells and offers to share your data. Super gross. MacOS has lost some of its luster, but it is still light years ahead.

The only negative about Chrome's sidebar tabs is that my tendency to have too many open tabs is now exacerbated. It is still better on the side, but now when I am lost, there is no way out except to close everything.

Laravel Cloud is such a clean and simple experience. It is almost criminal the way Salesfoce neglected Heroku and destroyed a 20 yr head start.

Swapped my Pi-Subagents over to @tintinweb’s implementation: github.com/tintinweb/p.... The “other” pi-subagents was very solid, but increasingly does far more than just run subagents. This one feels much more like a primitive than a review/workflow solution.

Like many of you, I use `npx skills` all the time to add skills. However, you can also use it to list your skills with `npx skills ls -g --json` and also use it to remove skills with `npx skills remove -g`.

Kicking the tires with bb (getbb .app). I am liking one app for local and remote dev (@ssh_exe_dev) - this setup misses the "throwaway" setup of Amp, but the option to have full control of the VM (ssh in/etc.) is quite nice.

Muv is still under review because apparently Apple has not been AI-pilled (and has no motivation to improve) yet. The app has no profile, no financial transactions, zero telemetry, and sends no data off the device. But we are now at 10 days since submission with zero feedback.

I could not figure out which open-source Rails app would be the easiest to showcase using Lewp with PG and Redis, so I gave the Codex App a simple goal to build an RSS reader and walked away. When I came home, I had - github.com/scottwater/..., which is far more useful than I intended.

github.comGitHub - scottwater/lewp-reader: Lewp reference web applicationLewp reference web application. Contribute to scottwater/lewp-reader development by creating an account on GitHub.

I wonder if “shared prompting” could help counter the deep code understanding that AI coding has eroded: two or more people working together with a single prompt/workflow.

My model breakdown: Fable - the only reason I use Anthropic models for anything. I typically use it for planning. Sol/Luna - the workhorses. Luna is underrated, especially on cost. Various one-offs with OpenCode Go (which is likely the best $10 you can spend).