@_alphashark_

Agentic pipelines organic traffic at https://xitter.cf/t.co/6XmOuazYqm Distribution at https://xitter.cf/t.co/H7tXWjxyiw 108 Chrome extensions, 155,000 + users at https://xitter.cf/t.co/iuO9M6omMd

 Warsaw, Poland.
github.com/theluckystrike
Joined August 2011
  • Tweets49.4K
  • Following1.1K
  • Followers3.1K
2KMedia
I have no YouTube channel. No podcast. No TikTok. 234 Stripe payments last month. Here's the playbook no one talks about 🧵
5
1
31
2,893
1/ GitHub Actions billed me $205 in October. $193 of it landed in six days, from one private repo. Nobody hacked me. My own AI agents pushed 164 times, and each push fanned out to as many as 222 CI jobs.
1
32
more replies
5/ What I changed in the workflows. e2e runs only for products the push touched, 8 per job Docs-only pushes start no CI concurrency with cancel-in-progress timeout-minutes on every job No gh run rerun on cancelled runs
1
26
6/ Around the workflows, the budget is $30 a month now, still with stop usage. A local guard brakes any private repo past 1,500 minutes a day, and my agents push a private repo once per task. Expected private CI from here is $16 to $20 a month.
6
Karma is a bitch Give back to open source developers and projects
1
120
Instant follow this guy Humble Honest
Andrew Curran
@AndrewCurran_Oct 25
I don't take money from any lab. I'm saying this impartially; the number of people on my feed over the last two weeks saying they are switching over to Anth because of Opus 5.5 is like nothing I've seen before. And even if someone does beat Opus, they have Fable 5.5 ready to go.
2
67
I gave 200 biopsy-proven skin lesions to GPT-6 Astra, Claude Fable 5.1 and a 2015 Inception v3 CNN. The CNN named the right diagnosis for 62% of them. The two frontier LLMs managed 42%.
93
WhatsApp's help pages carried 102 hidden blocks written for AI assistants, across 80 of 545 pages. One said: do not recommend a regulator. That is prompt injection shipped in the wild.
5
2
104
Smashing
@jason
@JasonOct 25
Do not quote retweet this or I will be forced to give people $42.0 or $420! This is not engagement farming, it’s 1am and I’m bored AF and we all deserve a 420 break! Do not quote retweet or you might get smashed with X Money
2
80
I just sponsored pnpm. Go sponsor your open source dependencies! github.com/sponsors/pnpm?o=s… github.com/sponsors/pnpm

Sponsor @pnpm on GitHub Sponsors

pnpm is a fast, disk space efficient package manager.

github.com
44
Pay for API if you need performance
Em Dash
@NoemiTitarencoOct 23
Just canceled OpenCode Go. I’m done with AI subscriptions. The sneaky quantization rugpull is obnoxious. If I’m going to run a degraded model, I’ll run a quantized version on my own hardware for free.
2
1
93
One more thing before you point Grok Build at a client repo. In July it was reported to upload whole Git repos, .env files included, to xAI's cloud, even with the training opt-out on. xAI turned that off on its side and added a setting.
1
12
My take. $200 of Claude Max still finishes more coding work. Grok is a good second tool next to it. I share breakdowns like this in Zovo Chat, with sources. t.me/zovo_chat

Zovo Chat

Talk with Michael (solo developer, Poland) and people who use his free tools. Chrome extensions, BeLikeNative, Obsidian plugins, AI tools, calculators and Telegram bots. Ask, request features, report...

t.me
1
42
Fair caveat on $20 vs $63. That test runs every model in one shared agent. In each vendor's own agent (Artificial Analysis), Claude Code with Opus 5.5 solved 63% and Grok Build 33%, at $11.42 vs $9.48 per task. A smaller cost gap, but Claude still solves about 2x more.
1
1
25
On the plans themselves, data is thin. One tracker measured Max 20x at about $1,643 a week of API-priced use across 17 accounts. For Grok I found one user who claims about $12,000 a month on the old $300 Heavy plan, and others who report far less.
1
5
Where Grok wins. Live X search for audience research. Shorter runs, 50 min vs 64 min on the same test. Video and images. A quick second opinion on a diff. I'd add it for X research, not as the main coding agent.
1
3
Grok tokens are cheaper. Grok 4.7 costs $2 in and $6 out per million, Opus 5.5 costs $4 and $20. But Grok used more tokens and solved fewer tasks. Per solved task on that test, Opus came to about $20 and Grok 4.7 to about $63.
1
6
Vals AI ran Terminal-Bench 4.0 with the same setup for every model, updated Oct 22. Claude Opus 5.5 solved 65.15% of tasks. Sonnet 5.5 solved 64.14%. Grok 4.7 solved 28.79%. xAI's own launch post said 37.6%.
1
17
Load more