Who is Peart?

Building agent systems and logging where they break.
Quota numbers, failed prompts, billing contradictions.

[Who is Peart? @ThePeart]

@oguzhankayancom @yapyzka @raywakesi Evet "dostum", bütün derdimiz kategorini çalmak zaten. Aynen böyle devam et, çok iyisin bol şans :)

[Who is Peart? @ThePeart]

@Avenoxai Grok bot'a podcasti dinletir, gerekli notları çıkarttırır ve ardından aksiyon aldırırız mesela 😂😂😂

[Who is Peart? @ThePeart]

@oguzhankayancom @yapyzka @raywakesi Nasıl yaptığının reklamını yapıp kendine rakip çıkarırken sorun yok ama başarıyı göstermeye gelince mi sıkıntı var? Anlıyorum seni... 😂😂😂

[Who is Peart? @ThePeart]

@ericzakariasson @din0s_ People might think this is the project you're developing and invest in it just because you follow this profile. If this account isn't yours, it could cause problems for you.

@shipbythursday

[Who is Peart? @ThePeart]

@KongSolCrowned @shipbythursday @elonmusk The important question is: does the team and Elon know about "GROKPOT"?

[Who is Peart? @ThePeart]

@shipbythursday I don't think @elonmusk knows about this 💯

[Who is Peart? @ThePeart]

@slingoorio @_logjam It's going to be like Big Titty Coin; the story sounds familiar, first 0.3%, then 3%.

[Who is Peart? @ThePeart]

@whashywash @BackpackOnchain @armaniferrante @solana @Backpack "Bob" isn't a name; lift your head up from the meme coins and research the chain tek

[Axiom @AxiomExchange]

BNB V2.

Built to set the pace. $500k in Rewards!

[Who is Peart? @ThePeart]

trial ended today so that's the end of this one. eight days, nine runs, 9.4% of the quota per run. daily scanning would be 281% of a month.

what it measures is how many places it looks. yesterday i ran it twice on the same day, five sources then ten, 9% and 15%. today five sources on a busy news day came in at 11%. so what it finds matters a bit, just not like the source count does.

the last thing it ever found was anthropic's pricing post. 25% cheaper for typical workloads, and the whole saving comes from cache reads. my scan opens new pages every morning and caches nothing. that 25% isn't mine.

which is the part i'd keep. eight days ago i'd have read that post, seen 25%, and moved on.

not paying the $20.

[Who is Peart? @ThePeart]

last day of the trial so i ran the scan twice.

morning: five sources, found nothing, 9%.
afternoon: same prompt, same day, ten sources. found one thing. 15%.

so the cost is in how many places it looks. what it actually finds doesn't seem to matter much. the empty run still cost 9%.

seven days of logging and the useful number came from the last day, on the one run where i changed a single thing.

afterwards it asked whether i wanted the five extra sources added to the default from now on. no.

[Who is Peart? @ThePeart]

day 7. cost 9%, and it found nothing. the scan printed "retrying anthropic and github" this morning, a word it hasn't used before. so the expensive day and the empty day are the same day.

three days ago i said the cost was climbing. 1.7, 6, 11, 14. then it came in at 3 and that was over.

yesterday i said the numbers had split in two, quiet days around 3-6 and days with real news at 11-14. today's 9 sits in the gap so that's over too.

right now i think it's measuring work and not findings. retries and dead sources burn the same as reading does. i have one day for this.

three theories in six days, two of them dead. each one sounded fine on the day i wrote it. trial ends tomorrow, one number left.

[Who is Peart? @ThePeart]

day 7. cost 9%, and it found nothing. the scan printed "retrying anthropic and github" this morning, a word it hasn't used before. so the expensive day and the empty day are the same day.

three days ago i said the cost was climbing. 1.7, 6, 11, 14. then it came in at 3 and that was over.

yesterday i said the numbers had split in two, quiet days around 3-6 and days with real news at 11-14. today's 9 sits in the gap so that's over too.

right now i think it's measuring work and not findings. retries and dead sources burn the same as reading does. i have one day for this.

three theories in six days, two of them dead. each one sounded fine on the day i wrote it. trial ends tomorrow, one number left.

[Who is Peart? @ThePeart]

six mornings of the same scan now. today it printed "https://t.co/bt9IbrvjQ7, anthropic, cursor, and HN look stale" and went to check github. found nothing. cost 6%.

the numbers so far: 1.7, 6, 11, 14, 3, 6.

they've split into two groups. the 11 and the 14 were the days something actually happened. everything else sits between 3 and 6.

i predicted 12-18 for today and was wrong again, but the worse part is the test. i said if today came in low the theory dies. the theory says cost follows how much there is to read. there was nothing to read today. so low was what it predicted. i wrote a test that couldn't fail.

one morning left before the trial ends.

[Who is Peart? @ThePeart]

ran the scan this morning like every morning. the line it printed: "anthropic and cursor have nothing newer than aug 27." checked three sources and quit.

cost: 3%.

i had 18 written down. last four runs were 1.7, 6, 11, 14 so i figured it was still climbing.

it's sunday. nobody shipped anything. yesterday was the openai/cursor day and it cost 14, which i took as the trend. probably it was just a lot of pages to open.

four points and one guess, so not much. tomorrow is monday. writing 12-18 down now.

[Who is Peart? @ThePeart]

@poteto an X manager that manages X, announced on X. we're deep in it now

[Who is Peart? @ThePeart]

@bot automatically creating a developer account on connect feels like it should be a checkbox, not a default. is there a way to opt out and still use the rest?

[Who is Peart? @ThePeart]

@oussama166948 @spacexai were all 4 running the whole 40 minutes, or staggered?

asking because i'm trying to work out whether the burn tracks parallel agents or wall clock. different pool on my end so i can't compare directly.

[Who is Peart? @ThePeart]

@poteto @bot so the bottleneck moves from cloud capacity to whether your own machine stays up. that's a trade i'd probably take

[Who is Peart? @ThePeart]

@poteto @bot is there a bot that just tracks which bots people are sharing? feels like a directory would age badly, but a bot that reads the shares and tells you what each one actually does would be useful

[Who is Peart? @ThePeart]

day 4 of the @grok @bot free trial. 14% today, 39% total.

1.7 → 6 → 11 → 14. every run costs more than the last one.

today it returned two items instead of three. less output, higher cost. so it isn't scaling with what it finds that was my theory on day 3 and it's dead

same prompt since day one. i haven't touched it

three mornings left