Dev K

Building agent-first Business OS for solo founders. 👉

[Dev K @DesiRichDev]

What if your CRM worked like this? 🤔

Dashboard → see everything
Claude → do everything

Both connected to your own database.
No sync issues. No SaaS lock-in.

[Dev K @DesiRichDev]

@ClaudeDevs you can now ship a live data dashboard without leaving Claude.
no backend. no deploy. no repo.
software engineers are fine. totally fine.

[Dev K @DesiRichDev]

@SpaceXAI Claude Code. Devin. Codex. now Grok Build open source.
every serious player is racing to own the agentic CLI layer.
the terminal is the most contested real estate in dev tools right now.

what do you guys think?

[Dev K @DesiRichDev]

Claude leans toward rigor in Russian. warmth in Hindi and Arabic.
the same model. different values. different language.
your AI product behaves differently depending on who's using it. worth knowing.

https://t.co/7cQVDq33eO

[Anthropic @AnthropicAI]

The values Claude expresses also vary with the language of the conversation, most noticeably along the Warmth vs. Rigor axis.

Claude leans most toward warmth in Hindi and Arabic. In Russian, it leans toward rigor—often asking the user for supporting evidence.

[Dev K @DesiRichDev]

X

[Dev K @DesiRichDev]

@claudeai the making of Claude Code is worth reading if you build dev tools.

Claude Code started as an internal CLI nobody was supposed to see.
now it's rewriting how the entire industry ships software.

the best tools are always built to scratch your own itch first.

[Dev K @DesiRichDev]

@OpenAIDevs reasoning + tool use in a realtime voice model.
that's not a mini upgrade. that's an agent that can think and act while talking.
voice interfaces just got a lot more serious.

if you're not building voice workflows yet, that excuse is gone.

[Dev K @DesiRichDev]

Toggle on from Opus 4.8 to Fable 5

[Factory @FactoryAI]

Fable is back in Droid today, available for everyone.

[Dev K @DesiRichDev]

@alvinsng or it could be furious 💨

[Dev K @DesiRichDev]

@cursor_ai Cursor for iOS.

my agents can now wake me up at 3am
from my phone instead of my laptop.

progress.

[Dev K @DesiRichDev]

@claudeai OpenAI on AWS Bedrock.
Claude on Azure Foundry.

the model wars are now procurement wars.
whoever is in the enterprise contract wins.

[Dev K @DesiRichDev]

@ClaudeDevs adding a judge took PR success from 25% to 80%.

verification is not a nice to have.
it is the entire architecture.

[Dev K @DesiRichDev]

@AnthropicAI fewer than 10% think they will lose their own job.
over a third think a junior colleague will.

everyone thinks they are the exception. 🤣

[Dev K @DesiRichDev]

Sol. Terra. Luna.

OpenAI just shipped a built-in three tier model stack.
frontier. balanced. cost-efficient.

the routing layer argument just became the product.

the most powerful model released today is not fully public yet.
that line deserves more attention than the benchmarks.

[Dev K @DesiRichDev]

@cursor_ai Opus 4.8 scores drop significantly under a stricter eval harness.

this is why i test every model on real tasks immediately.
benchmarks tell you what the model practiced. not what it can do.

the best eval is always a real task.

Cursor just proved why.

[Dev K @DesiRichDev]

@OpenAI non-developer Codex users up 189x in 10 months.

the agent is no longer a developer tool.
it is a work tool.

every department. every role.

[Dev K @DesiRichDev]

@warpdotdev the era of picking one model is over.

Warp just shipped YAML-configured model routing.
frontend tasks to GLM 5.2. architecture to GPT 5.5.

the routing layer is the new competitive infrastructure.

[Dev K @DesiRichDev]

@OpenAIDevs OpenAI shipped 30+ models in 6 months.

your stack decisions now have a shelf life of weeks.

the founders who win are not picking the best tool.
they are building in a way that makes switching cheap.

[Dev K @DesiRichDev]

nobody's talking about what Claude Fable 5 can actually do

I ran it - 89% of its context window

in that one session it built:
→ Tauri 2 desktop app (Mac + Windows)
→ multi-workspace manager
→ PTY sessions for Claude / Droid
→ built-in job scheduler
→ Telegram bot triggers
→ file manager + inbox
→ .dmg installer

the context didn't degrade
it shipped harder at 200K than it did at 50K

something big coming 👀

[Dev K @DesiRichDev]

the problem nobody solved:

running 5 client businesses under one AI account means:
→ one Gmail = all clients share it (wrong)
→ one Shopify = wrong store gets hit (dangerous)
→ create 5 accounts = nightmare

we built project-level connectors

client A's agents only touch client A's tools
client B's agents only touch client B's tools
never crossed. never mixed up.

Claude can't do this
Conductor can't do this
Factory can't do this

name dropping soon 👀

[Dev K @DesiRichDev]

was using @claudeai Fable 5 on @droid CLI by @FactoryAI

89% context window
zero degradation
shipped harder at 200K than at 50K

this is what agentic development looks like in 2026

[Dev K @DesiRichDev]

nobody's talking about what Claude Fable 5 can actually do

I ran it - 89% of its context window

in that one session it built:
→ Tauri 2 desktop app (Mac + Windows)
→ multi-workspace manager
→ PTY sessions for Claude / Droid
→ built-in job scheduler
→ Telegram bot triggers
→ file manager + inbox
→ .dmg installer

the context didn't degrade
it shipped harder at 200K than it did at 50K

something big coming 👀