Mayank Agarwal

Bag More 9s with @resolveai

Earlier this week, I shared our vision for AI for Prod: a comprehensive, always-on platform where you can run software with agents. Today, we are excited to announce Resolve AI's new capabilities, including: - Background agents that run operational tasks - A new agent architecture that delivers 2x investigation quality - New agentic capabilities like governed actions - New ways to work with agents in UI or terminal The future of software engineering is a closed loop between writing software and running it: continuous and self-improving, creating greater reliability and a world where every engineer can ship faster and operate production fluently. This is what we've been building toward since day one, and today's release is a major step. Thank you to the Resolve team, our customers, and our partners for pushing us here.

Today we announced Resolve AI Labs, backed by an additional $40 million in our Series A Extension at a $1.5 billion valuation, led by DST Global and Salesforce Ventures. This brings our total funding to more than $190M. Production environments generate the most complex data in software: logs, metrics, traces, topology, change history. All of it noisy, ephemeral, and constantly shifting. In incidents, this complexity becomes a crisis. Resolve sits inside the world's most complex production environments, like Coinbase, DoorDash, and Salesforce. Two years operating inside them is the foundation Labs is built on. The next generation of AI agents will operate across long horizons, watch production continuously, and act autonomously. Resolve AI Labs is our dedicated research arm pursuing frontier model performance at a fraction of the cost, purpose-built for production reasoning. Leading it is Dhruv Mahajan, who led post-training for large-scale Llama foundation models at Meta. We couldn't be more excited to have him.

We started Resolve at the beginning of the applied AI era. We’ve learned a lot since then - and this is one of the most important lessons we learned building, that changed our trajectory.

Continual learning is one of the hardest and most critical features to build for AI agents in production. Every company has deep institutional knowledge fragmented across docs, teams, and the minds of tenured engineers. Much of it has never been written down. This is the single biggest barrier to useful AI agents in production. We think about it as three types of knowledge: Organizational Vocabulary: naming conventions, shorthand, acronyms. This is the most straightforward to teach. Tool-Specific Knowledge: how to navigate systems, where data lives, how it connects. It's time-consuming to map manually and varies across companies. Decision-Trace Knowledge: the hardest to capture. The intuition that separates system experts from everyone else. Pattern recognition built over years, rarely documented, almost always context-dependent. Across all three, Resolve learns on its own, explicitly and implicitly, but always verifies with the humans who know the system best. Everything it learns is viewable, editable, and version-controlled. Resolve is already becoming the single source of truth for organizational context in production environments. More in my Introducing AI for Prod blog, link in replies.

My co-founder @spirosx and our investors Sebastian from @lightspeedvp & @saammotamedi from @GreylockVC talking about AI for Prod