▎Product @Ledger ▎Art project or research study? ▎Former music rights aficionado - mp3 as stonks
Paris, France
Joined June 2018
- Tweets426
- Following411
- Followers895
- Likes1.5K
Took me three pricing pages to understand how much i spend on my own setups in tokens and with regular resets, plans to decrease and change it's a wild west out there.
What's the meta now, paying for 22 subs for Claude and Codex?
1
67
NICE! I'd want one more layer: re-reads.
When doing this myself and counted my week, 98% of the tokens were agents re-reading their own context. About 2/3 of the spend, even though re-reads are the cheap tokens.
Improved my chip fab, and this version runs on real data
It replays my last 30 days of Claude Code, with 2,656 sessions, 86,131 tool calls and 27B tokens. Every session is a wafer lot, every tool call a process step, every error a bad wafer
Built by Opus 5.5 with viz3d
41
one manic hour of my coding agents burned 284 million tokens. 96.66% of it was re-reading
put that essay and 3 others on hashnode this week
honest question: where do you read long stuff about building with agents? substack, hashnode, medium, dev.to, your own blog?
morkeeth.hashnode.dev/where-…
1
2
87
Finally found the equivalent of torrents, never would have thought it would be .md instructions but we're out here downloading - throwing stars on github.
called a friend who said its life before and after @mattpocockuk BBQ skill
skillwire.net find it, peek inside, download it, try it on your own task
1
43,132
Caveman mode: The downfall of token maxxing is finding out that the unused tokens were not the waste. Sometimes the waste was the evening I spent trying to use them.
1
108
UI changes are legendary difficult to implement. Remember people being devastated by snapchats logo, Spotify basically unable to tweak anything.
But I approve of the brutalism clean squares of my inbox thank you @sundarpichai
81
The question of being overwhelmed in the agentic era, I have the answer, I efficiently dropped 1 bpm in pulse during heavy sessions. Token spend is efficiently correlated with resting.
Paired my garmin/apple health data over my transcripts/agent history/token spend
tokens: 168 million
heart rate with agents running: 65
heart rate without: 66
I'm a mini @bryan_johnson, and eager to see the future results of peoples
Curious to see who picks up the next experiment pairing it with: challenge @ericzakariasson @coolastekatten @flowstated @0xheavydev
-glucose monitor
-blood panel
-night time erection
-weight and body fat
1
3
180
It's sloppy, almost forbidden but trying to get things done across parallel projects are for sure the most hooking, dopamine inducing and real mid outputting. But the illusion of productivity.
The Era of Middle management in Agentic Engineering
1
62
Clear signing, the statement piece of transaction security. Great progress coming a long on the Solana Clear signing standard - sRFC 39.
@ledger <3 @solana
Our task force is shipping and more announcement shortly at OP3N and Break point!
Hardware wallets finally sign messages you can read.
Jupiter is the first to support @anza_xyz's new offchain message signing standard, for limit orders and DCA in the Jupiter Wallet and on jup.ag.
Live now with @Ledger via Jupiter Wallet.
jup.ag/wallet
4
187
It’s an art project, the most recurring question in the Wild West of new models, personal assistants, AGENTIC ENGINEERING:
What did you get done? The output mystery continues: 4.4k monthly decay
111
The day we waited for: pro plans for 500, the steady increase arise.
Tracking my spending: 2026 h2
Antrophic: 200
OpenAI: 200
X: 40
Cursor: 20
Random API, open router: flexible spend: 100
Vercel, railway, supabase: ish 60
3
81
Why the hell did nobody tell me agent transcripts get erased after 30 days?
I went looking for lost work and found an accidental journal instead. Maybe this is the final form of writing: messages to machines, before we edit ourselves.
teapawt.net/Morkeeth/status/209616…
The new generation of journaling: your agent transcripts. Perhaps a solution to the slopification era.
1
1
72
The new generation of journaling: your agent transcripts. Perhaps a solution to the slopification era.
Another week another experiment, once again coming back to the theme of whether the coding agents actually did what they said they did.
Pointed my agent-checker at 78,618 messages from my own AI agents. It said 42% of their claims were false. Then it caught itself. The bug was mine. real number is 8%.
built on Gemini 3.5 + Gemma, Cloud Run and @googlecloud ADK for #AllThingsAgentic
#AllThingsAgenticHackathon
76