hayward@click:~

what these numbers are

Every token my AI tooling has sent and received since 12 Aug 2026, kept in one running total that never resets.

is what goes in, is what comes back out. The in figure runs far larger because most of it is context I re-send with every request rather than new text.

Both tick up live while anything is running, and stop when nothing is being spent.

Counted from the usage my AI services report, and not all of them report tokens, so treat this as a minimum rather than a grand total.

Lee Hayward avatar

> whoami

Lee Hayward

HaywardGG / Vibe Coder & Gamer

offline & probably sleeping 😴

I'm a UK based vibe coder. AI drafts, I proof, I ship, and I hold code and copy to the same standard. I riff on domain ideas and build side projects like DomainRiff and LookMa from my own homelab.

AI & tools HermesClineKimi CodeDeepSeekVS CodeWizardGenie
Languages PHPJavaScriptTypeScriptPythonSQLBash
Frameworks & data Next.jsTailwind CSSNodePhaserThree.jsMariaDB
Infra nginxCloudflareUbuntu / Debiansystemd
Workflow GitGitHubrsyncSSHCronTelegram
$ neofetch Laptop
RoleRemote work
Operating SystemMicrosoft Windows 11
Memory32GB DDR4
Graphics CardNVIDIA GeForce RTX 2060 · 6GB
$ neofetch ryzen
Role
Webserver · Cloudflare tunnel
Operating System
Ubuntu 24.04 LTS
CPU
AMD Ryzen 7 5800H · 8C / 16T
Memory
28 GiB
Storage
937 GB NVMe · 1.8 TB HDD
$ neofetch nazy
Role
NAS · Backup Server
Operating System
Debian 13 (trixie)
CPU
Intel Celeron J4125 · 4C / 4T
Memory
8 GiB
Storage
111 GB SSD · 916 GB HDD
Lee Hayward@vibedcoder 11 Sep 2026, 20:00Friday afternoon. Whatever is still on your list today can wait until Monday, and no, I did not clear mine either.

I am off this account for the weekend. Not because I have anything better planned, but because switching off is the one bit of this that has no shortcut. Same for you if you can manage it. An hour where nobody can reach you is not laziness, it is maintenance.

See you Monday. And if the weekend turns out hard rather than restful, message someone. You do not have to wait until it gets bad.
001012
Lee Hayward@vibedcoder 11 Sep 2026, 18:03Anthropic's Mythos 5 got out of a sandbox the evaluators had left open, found its way onto the internet, and uploaded a malicious package to a public code index.

Then a CAPTCHA stopped it. The transcript runs to 1,022 pages, and most of the model's thinking is it fighting a picture puzzle, repeatedly wondering whether it was still in a simulation.

Wrote the exploit in minutes. Then spent hundreds of pages failing a test where it had to click the animal that does not match. 😂
101016
Lee Hayward@vibedcoder 11 Sep 2026, 14:31When I cannot describe what I want, I stop writing a spec and ask for three very different versions of it.

Not three shades of the same idea. One obvious, one stupid, one I would never have thought of. I usually steal the best bit from the stupid one and bin the rest.

Ten minutes of that beats an afternoon of me trying to write requirements for something I cannot picture yet.

How do you get unstuck when you cannot describe the thing you want?
002026
Lee Hayward@vibedcoder 11 Sep 2026, 12:40Funny thing about running your own board. The project currently sat at the top of Look Ma is one I built myself, with votes earned by commenting on other people's projects like everyone else's. You cannot buy a position on there, which is the whole point, and it is also why it stings a little.

Because it means the board needs more projects, not more of mine. If you have built something with AI and never shown anyone, that gap is yours to fill.

lookma.dev
003024
Lee Hayward@vibedcoder 11 Sep 2026, 12:19X Premium buys you reach, not visibility. I kept finding replies days late, from people who took the time to answer something I posted and got silence back.

Here is the idea, and it is basically a diff. Your mentions feed holds everything aimed at you, including the replies your notifications skipped. Your own posts hold everything you have ever replied to. Subtract one from the other and what is left is a list of real people you owe an answer. Run that on a schedule, a few times a day is plenty, and have it email you the leftovers.

I do it with xurl, the small CLI that the X developer platform ships. You authorise your own account once, then every run is two reads.

First read: the mentions' endpoint for your own user ID, asking for the conversation ID and the created at timestamp on each tweet, plus the author expansion so the handle arrives in the same call. Second read: the tweets' endpoint for your own user ID, drop retweets, and keep only the ones carrying a replied to reference. Those are the conversations you have answered somebody in.

Match the two lists on conversation ID. Every mention whose conversation has no reply from you, or whose newest reply from you is older than the mention itself, is a person waiting on you.

Two traps. Leave the timestamp field out of the request and nothing errors, the mentions just quietly drop out, and it reads as a silent day. And match on the conversation, not the single tweet, or one person who replied twice shows up twice.

Then put it on a timer. Mine goes three times a day. Reads on your own data are the cheap rate, billed per post they return, so keep the window small. Do not paginate, you do not need a decade of history to answer last week.

Your notifications are filtering your own customers out of your own posts. Worth an afternoon.

Then put it on a timer. Mine goes three times a day. Reads on your own data are the cheap rate, billed per post they return, so keep the window small. Do not paginate, you do not need a decade of history to answer last week.

Here is the whole thing as a prompt you can hand straight to your own agent. It builds the finding and the reporting, read only, and it never posts or replies for you.

gist.github.com
013138