Adventures of the Week
Hello again! ☺️
Yesterday, I took the kids to Legoland, and it was truly breathtaking to see all the cities constructed entirely from LEGO. I can’t help but wonder—who is responsible for building those amazing creations?
Apologies for missing out last week; I’ve been swamped with preparing my talk and finalizing some pages for a new course or manual—it’s a bit of a hybrid format, but I’ll reveal more details soon!
I’m off to San Francisco this Sunday, and I’m looking forward to a long flight with no internet access. Seriously, why do so many people crave WiFi on planes? I relish the forced disconnection.
Upcoming Events
I’ll be attending the following events:
-
OpenAI Dev Day
-
OmFest
-
Supabase Select
If you see me, don’t hesitate to say hi!
This Week’s Projects
Recently, I created a fun website showcasing popular devices from over the years, and I finished it in just a morning. I started using Codex and completed the project with Factory.
Stats:
-
87 devices, from 1976 to 2026, with images generated by Astra
-
40 messages from me
-
7 subagents
-
1,455 votes so far (thank you!)
Two weeks ago, I envisioned a website dedicated to forgotten devices from my childhood and sent it to Astra to explore its capabilities (I did waste a few attempts with builds like this 😅).
I want to create a site featuring nostalgic devices that look realistic and function as intended.
I need a grid with 50 devices.
With Astra’s help, a grid called “still works” was created, displaying images it generated and hosted on here.now. As you can see below, it has a very Astra-like aesthetic.
Then, I came across an impressive timeline scrubbing demo by Cole that showcased how devices could change under a cursor while dragging through the years. Inspired, I decided to merge my idea with Cole’s!
I often send tweets to my agents for inspiration and remixing. So, I asked Codex:
We created a website featuring images of forgotten devices on here.now. Can you determine where those images were rendered and created? I also want to develop a simple timeline scrubber that lets users see which device changes as they scrub through the years.
Within a minute, it identified the images and prompts responsible for their creation, and it generated three design options. I often prefer this approach since I usually don’t know what I want until I see it.
For option C, I want the timeline to be interactive, enabling users to drag across it, updating the device image based on their scrub position. We also need to accurately track the release dates for each device since many currently are just categorized in the 1990s.
The initial collection stopped at the 2000s, so I decided to extend the list to include devices up to the present day.
We also need devices from 2009 to today (as of now, it covers the iPhone duo). Utilize subagents for this task.
A subagent operates as another agent that processes information during its own session, which is useful for managing parallel tasks without bogging down the main thread.
For each device, it worked to identify the specific model, found a reference image, and generated a new image of that exact device.
The image prompt crafted for this new image was quite specific:
Subject: silver Casio Data Bank CD-401 (1984). It should match the silver metal bracelet model shown in the reference image. It has a square silver case, a black calculator face with four memory buttons above the rectangular gray LCD, and twelve calculator buttons below—exactly as laid out in the reference.
Then I thought it might be neat to let people share whether they owned or desired any of these devices, while providing a real count of votes. If they vote, they should receive a personalized page detailing their selection.
I moved over to Factory’s Droid, trusting it to leverage the models effectively.
When a user votes “had/wanted,” we should give them a shareable page showcasing their selected devices. Mock something up with image generation.
Three mockups were created again.
Next, I focused on the smaller details—buttons, keyboard shortcuts, and layout adjustments.
I envisioned this project on my own domain.
Taking steps to publish it on bentossell.com/devices and linking to it from my homepage (repos/bentossell on my machine).
My homepage and the devices site operate as distinct sites on here.now. The platform can “mount” a site at a specific path on your domain, allowing for seamless navigation without revealing their separate identities.
The device list and images are stored separately, while user votes are handled by here.now.
The “my devices” page does not retain any data (eliminating the need for authentication plus a database). User selections are embedded in the link; when shared, the URL carries them, producing an unrefined yet functional design.
Just before posting, I noticed some slowness. Upon loading the site, there’d be delays while dragging and waiting for images to appear. So, I launched a thread to address this.
The images should load instantly as users navigate the site; the delay is noticeable when scrolling.
In response, I received a lengthy, somewhat technical explanation, so I requested a visual representation.
Explain it to me as if I knew nothing about this topic, using an HTML artifact with large images and minimal text. Present the current situation and your proposed solution.
What was produced could have been improved, but since it assists me in completing a task, I’m okay with making further adjustments through skills or instructions. If it accomplishes the required task, that’s what matters.
It turned out that my testing had maxed out the limit for how many times a single user can read the votes within an hour. There is a ceiling at 25,000 votes.
Once the enhancements were made, the process now runs smoothly.
This journey began from a session on my phone—generating all those images on multiple occasions. It’s essential to remix others’ creations constantly (and give credit too) since so many individuals are sharing incredible demos daily. It’s hard not to get inspired by the community.
