Kumar D Ajay

Building Agentic Coding Harnesses and Workflows
2x Google Summer of Code

[Kumar D Ajay @KumarDAjay1]

All of this on "General coding assistance" mode and not the Spec driven development!!! Preposterous

[Kumar D Ajay @KumarDAjay1]

Is it just me or is Kiro the most useless thing ever built for coding?

Asked it to dump RAG results in a file, 4 lines of code btw, and it created a design doc, implementation plan, to-do list, spawned 5 subagents and got stuck in resolving dependencies all on Opus 4.8...

[Kumar D Ajay @KumarDAjay1]

Hey fellow AI developers on @X

What's something you recently found out about AI that blew your mind away?

I'll start: You can set the temperature to max and LLMs will respond in multiple different languages to philosophical questions like, "What is life?"

[Kumar D Ajay @KumarDAjay1]

@aanthonymax Old MNCs that started with Angular.js then migrated Angular and made it their standard still choose it even for new projects just because they have developers who know it

[Kumar D Ajay @KumarDAjay1]

Is it just me or do all of these 35B-MoE models just miserably fail at editing large yaml files where indentation is key?

I've tried qwen3.6, orntih-1 and now ornith-1.5 also fails to indent the edits properly.

I've tried these models with Cline, Aider and GitHub Copilot.

[Ornith @ornith_]

Aloha! 🌺Introducing Ornith-1.5, a family of open-source LLMs spanning 9B Dense, 35B MoE, and 397B MoE, trained with self-improving strategies.

It achieves state-of-the-art performance among open-source models of comparable size and delivers performance comparable to Claude Opus 4.8 across reasoning, agentic, and coding tasks:
✅Terminal-Bench 2.1 (86.1)
✅SWE-Bench (86 on verified, 65.1 on pro, 79.6 on Multilingual)
✅DeepSWE (56)
✅HLE (44.6)
✅ClawEval (81.4)
✅Tool Decathlon (71.2)

Ornith-1.5 takes a major step toward training foundation models through end-to-end self-improvement, extending the self-scaffolding strategies introduced in Ornith-1.0 into a more complete self-improvement loop: the model proposes new tasks, generates task-specific scaffolds, and produces solution rollouts for reinforcement learning, continuously creating new learning experiences from which it can improve.

All models, along with their quantized versions (FP8, GGUF, MLX, and NVFP4), have been released under the MIT License, enabling unrestricted commercial and research use.
📘Tech Blog: https://t.co/OZ63scRWLB
🤗Huggingface: https://t.co/mGJLwhrQOM

[Kumar D Ajay @KumarDAjay1]

Everyone's worried about Claude watermark and then there's some dude who created this...

[Kumar D Ajay @KumarDAjay1]

@sports_reactss Hope it reaches the right candidate

[Kumar D Ajay @KumarDAjay1]

@Kyle_Structure 7 is too less but anyways, hello 👋

[Kumar D Ajay @KumarDAjay1]

@Aftersevencvn @X MLX but for windows machines

[Kumar D Ajay @KumarDAjay1]

@Sharmaa_ji06 Cfbr

[Kumar D Ajay @KumarDAjay1]

@technicalyblond Hello

[Kumar D Ajay @KumarDAjay1]

@PrathameshBarv2 @X @grok Hope you find a job soon

[Kumar D Ajay @KumarDAjay1]

@maheshnani122 Reach++

[Kumar D Ajay @KumarDAjay1]

@garimaajuneja Hello

[Kumar D Ajay @KumarDAjay1]

@ShinigamiSukuna @X Hello

[Kumar D Ajay @KumarDAjay1]

@vineeth_agi May this reach the right people

[Kumar D Ajay @KumarDAjay1]

@mindinpanic @X @elonmusk Trying to build a lightweight harness that works with 9B to 35B-A3B models without overwhelming them with too much unnecessary context

[Kumar D Ajay @KumarDAjay1]

Hey @X Algorithm and @elonmusk

l'm looking to #connect with people interested in:

Agentic AI
Coding Harnesses
Local LLMs
SaaS
Frontend
Backend
Full-stack
DevOps
ML
Data Science
LeetCode & DSA
Freelancing
Building in public

If that's you, let's connect!

[Kumar D Ajay @KumarDAjay1]

@_VIDUSHI_SAINI @X Done 🙌

[Kumar D Ajay @KumarDAjay1]

@Kyle_Structure @X That's me