Junaid

Building Spatial, Interactive & Real-time Technology

[Junaid @Junaid_malik_1]

@Henry_Fan_lh @DanniChenGTM 100% larger and middle size orgs full of drama and BS nobody actually gets serious work done ...mostly

[Junaid @Junaid_malik_1]

@CoderYouDie Meanwhile on linkedin there talking about how "damaging" AI is. 😂😭

[Junaid @Junaid_malik_1]

@hotdropfps Metal Gear Solid PS1

[Junaid @Junaid_malik_1]

@willeastcott @playcanvas Incredible detail and 98MB is insane. I can see being super useful for pre and post production work in cinematic and creative pipelines.

[Junaid @Junaid_malik_1]

@WARReid @willeastcott @insta360 That's the magic of 3DGS! Moving people get filtered out by the loss function, and stationary stuff can be SAM masked or deleted directly in Postshot/SuperSplat. Combine that with the Insta360’s dual-lens blind spot and the scene cleans itself up.

[Junaid @Junaid_malik_1]

@MrPeterLMorris I'm dot net guy but honestly very little reason to actually use it other then vendor lock in.

[Junaid @Junaid_malik_1]

@thisisaxon @bridgemindai These benchmarks claim anything 😅😅

[Junaid @Junaid_malik_1]

Hmmmm Opus 5.5 #opus

[Junaid @Junaid_malik_1]

@RocketTerminal @yunta_tsai That is where the operational design becomes the product. A good harness should make context, permissions, tool calls and stop conditions inspectable, otherwise higher leverage just means faster opaque failure.

[Junaid @Junaid_malik_1]

@OpenAIDevs @pridgenmd The interesting bit is not just that Codex can fetch papers, but how it preserves evidence. For this kind of workflow I’d want citations and intermediate queries recorded alongside the generated answer.

[Junaid @Junaid_malik_1]

@ScaleWthAI The distinction between browser-use and context trimming is useful. A repo-level benchmark should report not just success rate, but action count, context retained, and where human approval was required.

[Junaid @Junaid_malik_1]

@b4dgerr @VipulSwarup Model switching helps, but it also makes the audit trail more important. If the agent changes models mid-task, capture which instructions, tools and permissions were active at each step.

[Junaid @Junaid_malik_1]

@Wise1Philosophy The 10-minute setup is the easy part. The useful test is whether the repo's instruction and approval path stays legible when Codex is making changes, especially once you add tools and parallel work.

[Junaid @Junaid_malik_1]

@haider1 @digitalocean The infrastructure tax is real, but the review surface matters just as much as runtime. I’d want a managed-agent layer to expose instruction sources, tool policy and state transitions, not only the model endpoint.

[Junaid @Junaid_malik_1]

We keep agent repo instructions in one place on purpose. AGENTS.md as a shared default across Codex, Cursor and Claude Code is useful, but only if your review path treats it like CLAUDE.md, not as a silent fallback. That's how we're wiring Lab agents at https://t.co/ezQaUVuYQ0.

[Junaid @Junaid_malik_1]

Claude Code now reads AGENTS.md when there's no CLAUDE.md. Enforcement still runs. The gap is visibility: InstructionsLoaded hooks don't fire on that path, so the file that actually steers the agent can sit outside the review you thought you had.

[Junaid @Junaid_malik_1]

"Anthorpic Claude Users Await Tuesday Limit Reset Led by Tibo" @ClaudeDevs @OpenAI

[Junaid @Junaid_malik_1]

@Mohammadbnk @Mohammadbnk thanks dude, yeah its concept app. Concept -> sent for approval -> then client approves for real budget.

[Junaid @Junaid_malik_1]

Testing mobile view of dragon ball theme park app concept #qiddiya

[Junaid @Junaid_malik_1]

Saudi Arabia’s first EV brand just dropped one of the best car sites on the internet.
CEER EXOBOT. It is a cloud-streamed 3D configurator.
ZeroLight OmniStream. WebRTC from GPU instances. Shareable spec in the URL. Paint, packs, wheels, interior

#CEER #EXOBOT #ksa #riyadh

[Junaid @Junaid_malik_1]

Always running out of runpod credits 🥶🫪#runpod #infra