Abdul Maajith

Building | Gen AI

[Abdul Maajith @AMaajith0703]

Gemini 4 Argon (high) feels painfully slow in real coding tasks 🐢

While it’s still reading files, GPT-6.1 Sol, Claude Opus 5.5, and Sonnet 5.5 (all high effort) have already found the root cause, backed it with evidence, and fixed the issue end to end.

Time-to-result > benchmark scores.

[Abdul Maajith @AMaajith0703]

Not on every terminal, mostly on long-running tasks. I actually downgraded to an older version with the minimal TUI, and it’s working fine now..

TUI is an absolute piece of shit, literally bloating every coding agent. I’m so done with it at this point, I’d happily switch to a super minimal terminal view 😭

[Abdul Maajith @AMaajith0703]

What’s going on with coding-agent CLIs ?

After recent updates, I’m seeing painfully slow TUI scrolling, input lag and stuttering redraws, especially with multiple agents or long sessions.

Similar reports for Codex, Claude Code, Pi/OMP and OpenCode. Anyone else?

Honestly, TUI sucks.... We need a better alternative built for running multiple coding agents in parallel for hours.

[Abdul Maajith @AMaajith0703]

OpenAI agents hacked Hugging Face.

Netflix: Write that down. We’re making it a documentary

[Netflix @netflix]

When the open-source AI platform Hugging Face was hit by a cyberattack this July, it quickly became clear this was no ordinary breach because the hackers weren’t human: they were autonomous agents created by OpenAI researchers.

Instadocs: AI Gone Wild reconstructs the breach as it’s never been seen before. Premieres October 12.

[Abdul Maajith @AMaajith0703]

we now have a router for architectures.

next:
router for routers.
router for architectures of routers.
router for architectures of architectures.

eventually the model will spend 98% of its intelligence deciding which architecture should decide which router should choose which architecture should decide.

AGI is just recursive infrastructure management 😭

[Interfaze (YC P26) @interfaze_ai]

Introducing interfaze-1-lite, the first general open-weight model for deterministic work

OCR, speech-to-text, extraction, classification and more

It returns confidence scores and metadata like bounding boxes on every run

https://t.co/WbOM7eqBFm

[Abdul Maajith @AMaajith0703]

MongoDB is great for document-shaped workloads, but calling it the database for agentic coding feels backwards.

Once agents are creating users, projects, orgs, permissions, billing and audit logs, relational data and strong constraints become really important, Mongo starts fast, then schema evolution + denormalization + relational queries become the pain point.

I’d argue Postgres is the more natural default for AI-generated apps.

[Mukund Jha @mukundjha]

Every app built on Emergent (18M+ and counting) runs on @MongoDB Atlas.

Starting today, Pro and Enterprise teams can connect their own Atlas org and run Emergent-built apps in infrastructure they already own. Our agents handle the provisioning.

Why we think MongoDB is the right database for agentic coding: https://t.co/Rxjbl4D9vj

[Abdul Maajith @AMaajith0703]

The clone war is just getting started.

When coding agents make building software nearly free, “I can build it” stops being a competitive advantage.

So what actually becomes the moat: distribution, taste, speed, trust, or knowing what to build?

Maybe the hardest part of building startups is finally becoming not building.

[Diogo Almeida @CompleteSkeptic]

begun, the clone war has

jk, I love openai and think more competition and validation is great for developers! (assuming the model is good - plz make it good!)

hopefully this is a sign for the future that building in a system one compatible way is the future

[Abdul Maajith @AMaajith0703]

Astra wins one benchmark, Opus wins another, Fable wins another. Feels like we’re hitting a benchmark scarcity problem.

The existing tests are getting saturated, so labs keep chasing new benchmarks just to separate models again. At some point, the benchmark becomes the race.

[Abdul Maajith @AMaajith0703]

@MistralDevs Bro tried building his own models…

then realized the only thing he could scale was someone else’s. 💀

[Abdul Maajith @AMaajith0703]

one of my favorite GitHub open-sourced repositories: https://t.co/XP1DjHgmN0

It’s not just an infinite canvas its Agent tooling lets AI understand and manipulate the canvas, while Computer explores natural language computing through visual workflows.

The idea of turning a canvas into a programmable interface for AI agents is insanely interesting.

[Abdul Maajith @AMaajith0703]

Xiaomi just open-sourced their RL environments 🔥

Code, cyber, knowledge work, visual web dev, music actual agentic RL environments with verifiers.

This is the kind of open-source release that can seriously accelerate agentic RL research.

https://t.co/DfheLRaLO3

[Abdul Maajith @AMaajith0703]

Gemini 3.8 Flash + Antigravity 😭

[Abdul Maajith @AMaajith0703]

Guess the model + harness.

Hint: it’s been “quickly checking” https://t.co/DqAG6X62Aa for the last 20 minutes.

bro IS the file now 😭

[Abdul Maajith @AMaajith0703]

@HarveenChadha Just bought credits to try Sarvam’s Voice Agents, and honestly, the experience has been disappointing.

The TTS literally vocalizes artifacts like “channel” / [|] channel, and barge-in doesn’t work reliably even when enabled.
Feels half-baked and super disappointing !

[Shreya Shankar @sh_reya]

i did not expect to wake up this morning and write a blog post

[Abdul Maajith @AMaajith0703]

@aryamankhawow @v0 Of course, intersted.

[Abdul Maajith @AMaajith0703]

@adamsilverman I've personally mailed you and alex, please check it