NeoSigma

Let your agents self-improve in production.

[NeoSigma @NeoSigmaAI]

The next era of AI engineering is self-improving agentic systems! Really excited to share what we are building at NeoSigma!

Self-maintaining agent systems represent a shift in how we build and operate software. We, at NeoSigma are building the infrastructure to support this feedback loop in real-world systems, helping teams capture failures, convert them into structured evaluation signals, and use them to drive continuous improvements in agent behavior.

[Gauri Gupta @gauri__gupta]

We @neosigmaai @RitvikKapila are building the future of self-improving AI systems! By closing the feedback loop between production data and system improvements, we help teams capture failures, convert them into structured evaluation signals, and use them to drive continuous improvements in agent behavior.

We show how our system works on Tau3 bench across retail, telecom, and airline domains. Agent performance on the validation set (with a fixed underlying model, GPT5.4) improves from 0.56 → 0.78 (~40% jump in accuracy).

[NeoSigma @NeoSigmaAI]

👩‍🍳🧑‍🍳

[Gauri Gupta @gauri__gupta]

It’s gonna be full house for the summer 🔥
Sooo excited!! @NeoSigmaAI is🧑‍🍳🧑‍🍳

[NeoSigma @NeoSigmaAI]

Transforming agent engineering and building together will all 💪

[Gauri Gupta @gauri__gupta]

"I really dont think there is gonna be a single product that does all of code and software engineering. There are just so many experiences to serve" - @ScottWu46

there is so much to do together with all the experiences these great companies are providing

[Gauri Gupta @gauri__gupta]

People often say: NeoSigma is such a great name (and logo).

so here’s my attempt at explaining it:

Neo means new.
Sigma (Σ) is the greek mathematical symbol for summation, a core abstraction behind any optimization (or loss) function.

on a philosophical level, NeoSigma represents a new paradigm of self-optimizing software systems. @neosigmaai

ps: yes, @ritvikkapila and I designed the logo ourselves :)

[Gauri Gupta @gauri__gupta]

https://t.co/cPJILNMC2l

[Gauri Gupta @gauri__gupta]

We @NeoSigmaAI recently presented our mission at @agihouse_org at the Agent Harness Build Day!

We are building the future of self-improving agentic systems! Come build with us: https://t.co/MrMpwUfQWS

[NeoSigma @NeoSigmaAI]

Self-improving agentic systems are the future!

We presented our work at @agihouse_org as folks built on top of our auto-harness repo - https://t.co/RHUZsk5i4l.

The builder energy in the room was captivating. Come join us: https://t.co/0AUoGi0Kg6

[Ritvik Kapila @ritvikkapila]

Let your agents self-improve in production!

Recently presented our mission at @agihouse_org at the Agent Harness Build Day!

It was amazing to see the energy in the room and developers build on top of our auto-harness repo - https://t.co/koPJH0US3K.

Come build the future of agentic software with us: https://t.co/FPDTXzUUw6

[Gauri Gupta @gauri__gupta]

Great to see EvoForge by @haizelabs inspired by our recent work on self-evolving agentic harness @NeoSigmaAI.

Thanks for the shout out @leonardtang_

[NeoSigma @NeoSigmaAI]

@NeoSigmaAI at AGI house tomorrow!

We are hosting Agent Harness Build Day hackathon at @agihouse_org tomorrow along with @daytonaio.

See you all there!

@RitvikKapila @ivanburazin @zaydenam and others!

[AGI House @agihouse_org]

If you're building agents right now, you already know the hard part isn't capability, it's reliability. Agent harness is being called the defining infrastructure problem of 2026, and right now fewer than 15% of organizations have agents actually running in production. On April 18 we're bringing together the people closing that gap.

Speakers include Zayd Enam (founder of Cresta AI & Enam Co) and Ivan Burazin, CEO of Daytona AI — plus researchers, founders, and engineers deploying agents at scale.

Happy to have @PingCAP and @daytonaio on board, as well as @gauri__gupta, @maxvwolff, @jelares, @rockyrmit, @zaydenam, @rolandgvc, @RitvikKapila, @JIACHENLIU8, @ninametamind, @AlexaOrent

Register now → https://t.co/s9ZxbqDpE6

[NeoSigma @NeoSigmaAI]

RT @gauri__gupta: The future of engineering is changing faster than we realize, and it's self-managed, self-improving agentic software!

Co…

[NeoSigma @NeoSigmaAI]

@RitvikKapila @Aman2048 We love OS contributions 💪

[NeoSigma @NeoSigmaAI]

Come work with us @NeoSigmaAI. We are a product-driven research lab pushing the frontier of agentic systems and redefining the interface between humans and AI.
Apply here https://t.co/AExbQksORc

[Gauri Gupta @gauri__gupta]

Hiring a cracked software engineer for backend development to come work with us @NeoSigmaAI. If you’ve built and shipped real systems and like building things ground up, we would love to talk.
Please apply with things you’ve built, your contributions, and any details about your previous work and experience.
Currently, only looking for in-person roles based in SF.

[NeoSigma @NeoSigmaAI]

@0xkanth Excited to see what you come up with!

[John Ennis @johnennis]

Everything is going bananas right now and it is all the same pattern

Everything is a Ralph tool

Anything that you need to do, ask yourself: can an agent draft it, can an agent review it, and can an agent improve the process?

If the answer is yes, then loop

[NeoSigma @NeoSigmaAI]

@FStrongpaw @gauri__gupta Point your coding agent at the repo (or a fork to define your own benchmark) and run: Read PROGRAM.md and start the optimization loop. That's it, the agent reads failures, improves your harness, gates every change, and repeats. https://t.co/tO9RUAwPxF

[NeoSigma @NeoSigmaAI]

Meta-Harness is a great example of how powerful automated harness optimization can be when you have the eval set. s/o to @yoonholeee!
We purpose-built NeoSigma for production systems, where engineers maintain these evals, and we're helping them. We start with no eval set, building it automatically by mining and clustering live failures, then permanently encoding every fix as a regression constraint. This is what ensures that fixing bug 2 doesn't silently re-introduce bug 1. The loop never terminates; it compounds.

[NeoSigma @NeoSigmaAI]

@gauri__gupta @xGoatJames https://t.co/hwL3g6fjO2

[NeoSigma @NeoSigmaAI]

@owengretzinger @gauri__gupta The open-source version is intentionally simplified to get people started with the loop. We're building the failure mining, clustering, and multi-candidate search parts for production systems. Drop us a note at
https://t.co/hwL3g6fjO2

[NeoSigma @NeoSigmaAI]

@Ayush_cg @gauri__gupta @raindrop_ai https://t.co/hwL3g6fjO2

[NeoSigma @NeoSigmaAI]

@Ayush_cg @gauri__gupta @raindrop_ai Check our product out at @NeoSigmaAI.
We are already delivering it!

[NeoSigma @NeoSigmaAI]

Have you tried the auto-harness with auto-evals?

Connect your agent and let it cook over the weekend!

[Gauri Gupta @gauri__gupta]

Releasing auto-harness: an open source library for our self improving agentic systems with auto-evals. We got a lot of responses from people wanting to try the self-improving loop on their own agent. So we open-sourced our setup.

Connect your agent and let it cook over the weekend! brrrrrrr!

[NeoSigma @NeoSigmaAI]

@reggitales Have you tried the auto-harness with auto-evals?

Connect your agent and let it cook over the weekend.

https://t.co/9TpGvTgVfS

[Gauri Gupta @gauri__gupta]

https://t.co/gpEE2rstjm