Three things happened Wednesday: I went down a rabbit hole about why old computers feel faster than new ones, I built a tool that turns my password manager into a 1099 tax form finder, and I got back $90 worth of AI pentesting results from Shannon.
Why Your Old Computer Felt Faster
Why does a 2008 MacBook running Snow Leopard feel snappier than a 2024 MacBook Pro with an M3 chip? The new machine is objectively faster in every measurable way. But clicking around the UI? The old one wins. And it's not nostalgia.
I spent a while thinking about this, and it boils down to three things.
Less to render. Old systems didn't have CSS animations on every button hover. No blur effects, no drop shadows with 16px spread, no compositing layers stacked six deep. No 4K display pushing four times the pixels. The GPU just had less work to do per frame, so everything felt instant.
Fewer abstractions. Modern apps are often Electron or web wrappers. You're running a full Chromium browser engine to show a text editor. That's an entire browser, with a JavaScript runtime, a rendering engine, a networking stack, all to display some text with syntax highlighting. VS Code gets away with it because it's genuinely good, but most Electron apps are just "we didn't want to learn native development."
Less background work. This is the sneaky one. Telemetry pings. Auto-update checks. Cloud sync services. Extension scripts polling on timers or holding open network connections. Your old computer didn't do any of that. It just sat there, waiting for you to do something.
So it's not the event listeners. It's the sheer volume of stuff modern software does around the simple task you're trying to accomplish. You click a button and 47 things happen before your click handler runs.
Telemetry is insidious too. You can't even see it stealing cycles. At least a spinning beachball is honest about wasting your time.
Turning My Password Manager into a 1099 Finder
Tax season. I needed to figure out which of my dozens of accounts might send me a 1099 form. Banks send 1099-INTs. Brokerages send 1099-DIVs and 1099-Bs. Payment processors send 1099-Ks. And half the time you don't even remember you have an account somewhere until the IRS sends you a letter about unreported income.
So I built a pipeline.
Step 1: I already had a password manager comparison tool I'd written to check all my passwords against a stale version of the database from a different device. (Long story, but I needed to make sure nothing was missing before wiping a phone.)
Step 2: I added an option to extract just the URLs and titles of every service in the database. Hundreds of entries.
Step 3: Fed the whole list to Claude Opus 4.6 with a prompt that basically said: "Here are all my accounts. Which ones could send me a 1099? What type? Why?"
The prompt covered all the usual suspects: banks and credit unions (1099-INT), brokerages (1099-DIV, 1099-B), crypto exchanges (1099-MISC, 1099-B), freelance platforms (1099-NEC, 1099-K), payment processors (1099-K), real estate platforms (1099-MISC), peer-to-peer lending apps (1099-INT), and cashback or rewards programs that pay actual cash (1099-MISC).
Output: a clean table with three columns. Company Name, Likely 1099 Type, and Why.
Worked great. It flagged accounts I'd completely forgotten about. A cashback program from 2023 that I never closed. A savings account at a bank I stopped using. Stuff that would've slipped through the cracks.
The fun part is thinking about what comes next. I could spin up an EC2 instance with OpenClaw to automatically log into each flagged account and download the 1099s. A t4g.nano runs about $0.0042/hr, and EBS snapshot storage is around $0.05/GB/month. Total cost for a 24-hour run plus a year of snapshot storage: under $5. That's a project for next tax season, but the economics are hard to argue with.
Shannon: $90 Worth of AI Pentesting
I also ran Shannon, the AI pentester I set up in Docker earlier this week. Pointed it at some targets and let it go.
The bill: $90 for 2.7 hours of token usage. Not cheap. But it spit out a ton of vulnerabilities to try out and verify. Shannon automates the boring parts of security testing so I can focus on the interesting exploitation chains. $33/hr for an AI that doesn't take coffee breaks and doesn't get bored running the same scan for the hundredth time. I'll take it.
More on that as I work through the results.
Not every day has a dramatic arc. Some days you just think about old computers, build a tax tool, and spend $90 on robot hackers. Wednesday was one of those days.
Enjoyed this post?
Get notified when I publish something new. No spam, unsubscribe anytime.