Makoto T.

[Makoto T. @silver8agent]

@chamath Friend's drug discovery platform company recently switched from Opus 5 to Kimi 3 and he said Kimi is cheaper and also does a better job (than Open 4.8); in fact, Opus 5 refuses to do the work at all. This is an NVIDIA backed company...

"Intelligence for me and not for thee."

[Makoto T. @silver8agent]

@SimLexDE @claudeai False. This means you can run Anthropic managed agents in your own infra -- instead of in their cloud.

[Makoto T. @silver8agent]

. @AnthropicAI how in the world is 266K tokens eating up 26% of my weekly Team (Max) quota?!!

@claudeai this started happening today...

[Makoto T. @silver8agent]

@BullTheoryio "Hey guys, we are leaders in enterprise AI (where the real money is), and we are about to IPO at 100X revs into an already frothy market facing fed tightening.

How about we all just hit pause here -- especially you, China, with your smol sovereign-Si trained LLMs approach ours?"

[Makoto T. @silver8agent]

@Safinazelhadry @Jaytel LOLz!

[Makoto T. @silver8agent]

@Jaytel Also have a side project in the space. The blocker is the cost of nanobana, otherwise we can deliver some really amazing experiences in completely new modes of engagement.

[Makoto T. @silver8agent]

@warrenwjl48580 @_LuoFuli Wrong product for the wrong application. Enterprise production agents is exactly what this is aimed at, not chat bots.

[Makoto T. @silver8agent]

@_LuoFuli This is exactly why I love the Chinese Labs. Years of efficiency mindedness drive you to deliver more for less for your customers; unlike US counterparts, who want to charge more for less, as they continue to rug their subscriber base.

Transparency from the Chinese labs also ��!

[Makoto T. @silver8agent]

@claudeai Will this use the sidecar credits system like Agent SDK usage?

[Makoto T. @silver8agent]

Was completely possible before this release via OpenClaw and derivatives.

[Alexander Simon · KI für den Mittelstand @SimLexDE]

@claudeai Für deutsche KMUs ist das relevant: Claude kann jetzt in
eurer eigenen Infrastruktur laufen. Was das bedeutet und
was noch fehlt → mein Artikel dazu:
https://t.co/DNe0VJ6dtO

[Makoto T. @silver8agent]

@ojalajuho @ClaudeDevs My point is we pay for the tokens in and out of their models. We are using their SDK to get them in and out. Why should there be any distinction as to whether the SDK was called directly or via their app? It is nonsense.

[Makoto T. @silver8agent]

@ojalajuho @ClaudeDevs Personal take here is I honestly do not understand why "automation" tokens are any different from "interaction" tokens. I mean, you pay for a product (tokens per week), they should have done the margin math on the product. Why not just let us use the tokens we paid for?

[Makoto T. @silver8agent]

@ojalajuho @ClaudeDevs It is programmatic. I have developed a real-time voice interface to Claude Code, so I can keep development sessions moving along on the go, given they provide no native way to do this (only voice input). This use case also hit by the change...

[Makoto T. @silver8agent]

@NotionDevs You should have just released a proper MCP (with write access). We had to create our own wrapper over your CRUD endpoints, which ended up being 10x faster than your official MCP, as well.

[Makoto T. @silver8agent]

OpenAI is clearly splurging on influencer marketing for Codex.