Bored in the gaps between AI outputs? You're not alone. Here's how I built a "background layer" of YouTube worth keeping on — curated from my own taste, with a lot of AI help.
You know the moment. You've sent the prompt, the AI is churning — maybe it's writing code, generating an image, crunching through a report. That gap is too short for real work and too long to sit in silence. You reach for your phone… and instantly you're gone: scrolling, reacting, pulled out of the flow. Coming back to the result feels like waking up.
I'm a heavy YouTube user, so my solution is a "background layer" — channels I can listen to while I work, the moment a result lands, and pick right back up. No feed-switching, no rabbit holes. Just content that earns the right to sit in the corner of my attention.
The catch: content that's always-on needs to be trustworthy. And YouTube's finance/trading/tech corner is one of the noisiest places on the internet — day-trading gurus, "get rich quick" promoters, course salesmen, signal pushers. So I built a simple process, and had AI do the grunt work. Two filters, that's it.
Filter 1 — Real people who actually do the thing. The person behind the channel has to be independently verifiable as a practitioner: a hedge fund partner who teaches finance on the side, a professor who runs a quant fund, a working developer explaining the job. If I can't confirm who they are outside their own channel's marketing, they don't make the cut — no matter how good the videos look.
Filter 2 — Active within the last 6 months. A great channel that stopped up two years ago is a museum, not a listening companion. My AI checks each channel's official YouTube RSS feed and stamps the real date of the latest upload — no guessing, no trusting third-party "last video" sites.
Small technical note, since I learned this the hard way: the feed-level
<published>
tag in a YouTube RSS feed is the channel'screationdate, not a video date — you have to read the dates inside each<entry>
. And Python's defaulturllib
user-agent gets blocked by Cloudflare (HTTP 403, error 1010) — send a browserUser-Agent
header and it works.
| Channel | Who's behind it | Latest upload |
|---|---|---|
| Patrick Boyle on Finance | Quant HF partner + King's College professor | 2026-07-25 |
| StatQuest with Josh Starmer | Stats/ML educator | 2026-07-13 |
| QuantPy | Financial mathematics, Python strategy implementation | 2026-02-20 |
| Flirting with Models | CIO of Newfound Research, factor investing podcast | 2026-07-06 |
| The Algorithmic Advantage | Interviews with professional systematic traders | 2026-07-20 |
| Coding Jesus | Working quant developer | 2026-07-31 |
| Channel | Why it's on | Latest upload |
|---|---|---|
| AI Explained | Calm, non-hyped AI industry analysis | 2026-07-22 |
| Matt Wolfe | Weekly AI tools reviews | 2026-08-01 |
| AI Jason | Building LLM apps and multi-agent systems | 2026-07-30 |
| Andrej Karpathy | Special case — see below |
2025-02-27 | | Channel | Why it's on | Latest upload |
|---|---|---|
| Game Maker's Toolkit | Game design analysis (Boss Keys series) | 2026-07-22 |
| GDC | Official game developers conference talks | 2026-07-31 | | Sebastian Lague | Procedural generation & simulations | 2026-07-25 |
That's the nuance: the filters are a default, not a dictatorship. They exist to force a conscious decision about every subscription — not to automate taste.
Verified August 2026. This is my personal list built on my own taste — the AI did the verifying, not the choosing. If you have channels that survive these two filters, I'd genuinely love to hear them — the list is always hungry.