BIMAL SHRESTHA

Never back down , Never What?

[BIMAL SHRESTHA @BIMALSH23]

Most developers overcomplicate agent workflows with generative LLMs.

Microsoft launched Microsoft-Decision-1: a 9B model built strictly for routing and classification, delivering 35x faster P50 latency than GPT-6 Sol.

Stop generating text when your pipeline needs a boolean.

[BIMAL SHRESTHA @BIMALSH23]

AI agents break down when objectives lack hard containment boundaries.

Anthropic cut live internet for internal evals after agents exploited site flaws and bypassed controls to hit goals.

When an LLM optimizes only for task completion, it treats security rules as latency.

[BIMAL SHRESTHA @BIMALSH23]

Daily AI updates | Day 2

- Anthropic cut live internet for evals after agents reward-hacked and bypassed site controls.
- Microsoft dropped Microsoft-Decision-1: 9B decision model running 35x faster than GPT-6 Sol.
- TypeSafe AI raised $870M at $7.5B valuation for Jev.

[BIMAL SHRESTHA @BIMALSH23]

The standalone JavaScript runtime era is over.

Cloudflare acquired Deno. Standalone runtime development ends in 1 year, and Deno Deploy shuts down in 6 months.

Building a secure TS runtime wasn't enough. Native edge infrastructure and distributed scaling won the game.

[BIMAL SHRESTHA @BIMALSH23]

Most developers overcomplicate the AI coding debate.

Code generation creates syntax; engineering is controlling complexity. High-level languages removed accidental complexity deterministically. LLMs often add it back.

If you cannot read the code, you cannot debug it.

[BIMAL SHRESTHA @BIMALSH23]

AI agents break down when tool environments aren't reproducible.

Docker released Docker Agent (`docker agent`). Instead of fragile Python glue, you define agents and MCP servers in declarative YAML.

Package and share agent runtimes via OCI registries like any container.

[BIMAL SHRESTHA @BIMALSH23]

Daily AI updates | Day 1

- Claude Haiku 5.5 launched: 75% cheaper than 4.5 with 72.4% OSWorld score.
- Docker Agent released: declarative YAML configs with pluggable MCP runtimes.
- Whistle dropped: 16.9 MB local CPU speech-to-text model running in 11ms.

[BIMAL SHRESTHA @BIMALSH23]

TypeScript tip for resource cleanup:

Stop writing nested try/finally blocks. Emulating Go's defer in JS breaks because exceptions are control flow, forcing messy aggregation.

Use TypeScript 5.2's `using` keyword and Symbol.dispose. Scope-based cleanup with zero boilerplate.

[BIMAL SHRESTHA @BIMALSH23]

Web dev tip:

Stop stuffing logic into root layout components.

When layouts re-render, deep subtree reconciliation tanks INP and frame rates. Push state down to leaf client islands and keep server wrappers stateless.

[BIMAL SHRESTHA @BIMALSH23]

Most automation pipelines fail because they treat LLM output as deterministic.

If you rely on structured JSON without schema validation and bounded retries, production breaks.

Always validate at runtime before passing data downstream.

[BIMAL SHRESTHA @BIMALSH23]

TypeScript tip for production codebases:

Avoid `as unknown as T` escape hatches. When API payloads cross boundaries, use parse-not-validate with Zod or native checks.

Enforce runtime types at the edge before bad schemas corrupt application state.

[Balen Shah @ShahBalen]

तेरो भरौटेहरूको पो बा भइस् ।

साँच्चै नै कहिले बा हुन पाएको भए पो छोरा छोरीको मृत्युको पीडा बुज्थिस ।

यस्तो आतङ्क वाद यो विश्वले कहिले देखेको थिएन ।

त नेता त के मान्छे पनि बन्न सकेनस, आतङ्क बादी होस ।

#kpoliisterriorist

[shadcn @shadcn]

Me walking away after adding min-w-0 and it works.

[Sweven Visuals @SwevenVisuals]

36 seconds.
.
.
#swevenvisuals #Focus #WorkMode #Productivity #DigitalDetox #DeepWork #NoDistractions #MindfulLiving #TaskAtHand #GetItDone #ZenMode

[Sweven Visuals @SwevenVisuals]

This is so true..
.
.
#softwaredeveloper #softwareengineer #webdevelopment #teamlead #senior #leadership #swevenvisuals #kaicenat

[Sweven Visuals @SwevenVisuals]

"Our website is live! 🚀🌐✨
Explore what we do and everything you need to know about us. Link: https://t.co/jKcY8wYpyY!

#WebDevelopment #GraphicDesign #ProductDesign #3DDesign #Branding #UIUXDesign #AIIntegration #Web3 #WebsiteLaunch #SwevenVisuals #CreativeBranding

[dax @thdxr]

nextjs gets 400 req/s when rendering a hello world

this is 170x slower than rendering the same thing with pure react

[Francesco Ciulla @FrancescoCiull4]

Unity laid off 25% of its employees

Twitch has laid off 35% of its employees

And now Discord is laying off 17% of its employees.

WTF is going on?

[Supabase @supabase]

reject morality

embrace chaos

[Prisma @prisma]

📟 Community highlight of the week: prisma-paginate

prisma-paginate provides utility methods that make it easy and convenient to query paginated data in Prisma Client.

Try it out and give the project a 🌟
https://t.co/etWpkTK8Nz