Arnab Saha

Engineering Manager @atlanhq. Building AI agents, multi-agent systems, and observability tooling. Writing about what works and what breaks.

[Atlan @AtlanHQ]

Introducing Context Conference, the community conference where you contribute to shaping context as a discipline.

Join us live on Oct 28!

[Arnab Saha @arniesaha]

https://t.co/LvtoziL7oG

[Prukalpa ✨ @prukalpa]

Launch alert! 🚀 Excited to announce the launch of Becoming Frontier today.

For the past 16 months, we’ve been on a quest to build @AtlanHQ into a truly frontier company. 16 months of learning, unlearning, failing, and experimenting. Today we're sharing all of it.

Becoming Frontier is our inside look at rebuilding Atlan into a truly AI-native company.

[Arnab Saha @arniesaha]

https://t.co/W7dSuBv8Yo

[Arnab Saha @arniesaha]

We built this for a real reliability engineering problem: investigating production alerts faster and more consistently.

Running an agent near real incidents taught us the model is only one part. The loop around it matters just as much.

Proud of the team for what they shipped.

[Atlan @AtlanHQ]

We had an agent loop running in early 2026. It still failed.

11,000 alerts a month and the agent investigated all of them with the same intensity, burning the API budget on noise and learning nothing between runs.

So we built the scaffolding around it: admission filtering, memory of past wrong hypotheses, and cost discipline per run.

85% of alerts now suppressed before a model runs. Investigations down from 10+ min to ~2 min at a third of the cost.

By @sharlok7, @jain_nikhila, & @arniesaha

[Arnab Saha @arniesaha]

@PlayStationIN Woohoo! Just saw this! Thanks @PlayStationIN sent you a DM

[Arnab Saha @arniesaha]

@FanboiiApple @PlayStationIN Hey man. It’s a generic one.