Evgeny Ogryzkov | Indie AI Builder

Some view Pi Agent token efficiency as overstated compared to OpenCode. I was skeptical and checked myself on current projects.

raymel 👋@pseudokid

Whole day with Pi agent just used 10% of my $20 Codex weekly quota All gpt 5.5, various thinking levels I'm now 3 hours away from weekly reset and still have 42% left! For my typical 4-6h/day session: Codex Desktop - 20-30% a day, 1-2 days/week Pi - 10% a day, 4-5 days/week? My switch to Pi is really looking good so far

English

130

Evgeny Ogryzkov | Indie AI Builder@ogrizkov·2h

@Cointelegraph Agent SDK and GitHub Actions credits will make real agent loops more sustainable. Haven’t been hitting Claude Code hard lately — the current limits are already generous.

Evgeny Ogryzkov | Indie AI Builder tweet media

English

105

Cointelegraph@Cointelegraph·11h

🚨 NEW: Anthropic announced that starting June 15, paid Claude plans will include monthly credits for programmatic usage across tools like the Claude Agent SDK and Claude Code GitHub Actions.

English

205

17.4K

@OpenAIDevs Codex long agent runs hold up well in practice. The 2 free months promo is a solid nudge for teams still on pilots.

English

3.4K

OpenAI Developers@OpenAIDevs·15h

Want to (officially) use Codex at work? Send this post to your CTO to bring your team to Codex. Eligible enterprise customers who switch in the next 30 days get 2 free months of Codex usage for new users.

English

198

232

3.4K

830.7K

@GoogleDeepMind Pointing at messy code and just saying “fix this” actually works now. No more typing a full novel for the prompt. Already imagining this in daily flow.

English

Google DeepMind@GoogleDeepMind·1d

In the real world, we don't tend to speak in long paragraphs; we point and say: "fix this" or "move that". 💬 By combining gestures with speech, it lets you use natural shorthand to complete tasks.

English

260

20.9K

Google DeepMind@GoogleDeepMind·1d

We’re reimagining a 50-year-old interface - the mouse pointer - with AI. 🖱️ These experimental demos show how people can intuitively direct Gemini on their screens using motion, speech, and natural shorthand to get things done 🧵

English

401

8.2K

1.5M

@bridgemindai That’s not like I don’t believe in benchmarks. It’s just that almost every time Kimi K2.6 feels so much faster than every model. It always feels like Haiku in terms of speed but much smarter.

English

255

BridgeMind@bridgemindai·14h

Kimi K2.6 takes 41 minutes per task. Claude Opus 4.7 takes 5.8 minutes. Same harness. 7x slower. Stop using Chinese models for production coding. The benchmarks look competitive until you check the speed. Fast and smart beats slow and comparable every time.

English

348

31.3K

@ClaudeDevs Fired /goal on a 3-hour refactor that normally dies at 40min. It actually finished this time. Holds up on long sessions.

English

789

ClaudeDevs@ClaudeDevs·1d

How do you keep Claude working until the job is done? Claude Code helps with this in a few ways, including one we shipped recently: /goal.

English

390

922

12.6K

1.3M

@ithmzeth @pseudokid no it was just read AGENTS.md request with boostraping Obsidian vault, same for both

English

tmz Thomas@ithmzeth·1d

@ogrizkov @pseudokid plan mode will likely cost more token is it the same mode for pi?

English

raymel 👋@pseudokid·2d

raymel 👋@pseudokid

Pi Agent vs OpenCode token usage A lot of people recommended Pi Agent so I decided to check Pi Agent took 1.1k tokens in first turn OpenCode took 11.5k Setup: 1) Trimmed OpenCode (from usual 30k first turn to 11.5k) - 0 MCPs - 2 lightweight plugins (opencode-env-protect and openslimedit) - 8k char AGENTS.md - 11585 deepseek-v4-flash input tokens - for $0.0016 2) Vanilla Pi - 0 MCPs - 0 system prompt - 8k char AGENTS.md - 1114 kimi-k2.6 input tokens - for $0.0008 I think using better models with capped tokens per turn can keep usage nearly the same as uncontrolled DeepSeek V4 Flash? The challenge now is finding the sweet spot. Can't cap tokens if quality drops. We'll see. Video below - OpenCode vs Pi side by side "say hi back" first turns test, with OpenCode Go usage for each

English

506

106.5K

@ithmzeth @pseudokid yeah, what’s the difference?

English

tmz Thomas@ithmzeth·1d

@ogrizkov @pseudokid u ran opencode on planmode?

Nederlands

Look what OpenCode does. It wastes tokens for two full minutes just reading the project context.

English

I am going to switch for a while to test longer sessions. Actually it amuses even from the installation with those animations.

English

Some view Pi Agent token efficiency as overstated compared to OpenCode. I was skeptical and checked myself on current projects.

raymel 👋@pseudokid

English

130

@claudeai Some view multiple Claude Code sessions as difficult to track. The new agent view still lists all sessions in one place. I will test dispatching and inline replies on current agent projects shortly.

English

233

Claude@claudeai·2d

New in Claude Code: agent view. One list of all your sessions, available today as a research preview.

English

982

2.2K

28.7K

5.6M

@EdenKollcinaku @GeminiApp @gmail @googlephotos @Google @YouTube Wow, thanks for a tip lol

English

Eden@EdenKollcinaku·2d

@GeminiApp @gmail @googlephotos @Google @YouTube best when used this way 😅

English

1.1K

Google Gemini@GeminiApp·2d

Personal Intelligence can help you build a travel itinerary that’s fully custom to you. It saves time by connecting dots across your @Gmail, @GooglePhotos, @Google Search, and @YouTube history. Choose which apps to connect and manage your personalization settings at any time 🧵

English

103

203

1.6K

91K

@GeminiApp @gmail @googlephotos @Google @YouTube Grok now connects to private GitHub repos. Gemini, with 20 years of access to my Gmail, is planning trips.

English

531

Some view Claude API on AWS as standard cloud availability. As AWS certified professional I will test how the API works shortly. The setup uses existing authentication and billing on current agent workflows.

Claude@claudeai

The Claude Platform on AWS is now generally available. AWS customers get the full set of Claude API features, with AWS authentication, billing, and commitment retirement.

English

@0xyunss The 12 rule CLAUDE.md still holds up on multi-step agents. Fail loud rule especially saves hours when agent claims success but misses something.

English

539

yunus@0xyunss·2d

artikel ini worth banget buat yang lagi build agent atau pake Claude buat coding. singkatnya: Karpathy complain soal cara Claude nulis code. ada yang baca complain-nya, packaging jadi 4 rule di satu file CLAUDE.md, upload ke GitHub. 5.828 stars di hari pertama. 120.000 stars sekarang. 4 rule originalnya solid. tapi orang yang nulis artikel ini ngetes di 30 codebase selama 6 minggu dan nambah 8 rule lagi buat nutup gap yang ga dicover rule original, khususnya buat multi-step agent, token budget, dan silent failure. yang paling relate buat gua pribadi: rule soal "fail loud". berapa kali Claude bilang "migration completed" atau "tests pass" padahal ada yang kelewat diam-diam. ini yang paling sering bikin debug lama karena keliatan sukses padahal engga. sama rule token budget. tanpa ini agent lu bisa jalan 90 menit di loop yang sama tanpa nyadar dia udah nyaranin hal yang sama yang udah lu reject sejam lalu. kalau lu build apapun di atas Claude, paste 12 rule ini ke CLAUDE.md lu malam ini.

Mnimiy@Mnilax

x.com/i/article/2053…

Indonesia

613

29.9K