# Mrinal Wadhwa

> CTO at Autonomy, Ockam, and Fybr.

Some things I've created:

- [Autonomy](https://github.com/build-trust/autonomy), a platform to create AI products that use swarms of intelligent agents to autonomously get deep, complex work done.
- [Ockam](https://github.com/build-trust/ockam), open source protocols and libraries to add cryptographic identities, mutual authentication, fine-grained authorization, and end-to-end encryption to autonomous and distributed systems.
- [Unramble](https://github.com/mrinalwadhwa/unramble), an open source app that makes it fast and simple to type with your voice.

## Recent Talks and Interviews

- [Agent Swarms: live demo at GitHub HQ of 5000 parallel AI agents collaborating on a deep code review.](https://www.youtube.com/watch?v=uDby5FKQN9k)
- [Talk at Snowflake on how to create long-horizon agents that can autonomously do complex work.](https://www.youtube.com/watch?v=0Ifm-W2Frv0)
- [Software engineering is getting harder, a conversation with ScaleKit's CTO on their podcast.](https://www.youtube.com/watch?v=rkXnJkkvnks)
- [Guest lectures at IIM Bangalore on how AI agents are disrupting industries, and my advice for leaders.](https://www.linkedin.com/posts/mrinalwadhwa_guest-lectures-on-agents-at-iim-bangalore-activity-7390975540800811008-l0k2)
- [Cryptographic identities and how to build secure agent swarms. A chat with GM of AI at 1Password.](https://www.youtube.com/watch?v=ZXSKPnBnV3w)
- [Trustful communication in distributed systems, a conversation with Snyk's CTO on their podcast.](https://www.youtube.com/watch?v=SF1arse7B5M)
- [Context engineering and how to make autonomous agents that succeed, a talk at Snowflake.](https://www.youtube.com/watch?v=t9jTkEhcNUg)
- [Shipping an AI product live, in 10 minutes, at the MCP demo night at WorkOS.](https://www.youtube.com/watch?v=ZJVcOy9QNp4)

## Background

For 21 years, my work was focused on designing distributed systems and products that handle large volumes of data.

Early on, I worked on high-availability and replication for network-attached storage at EMC, where I learned how to make complex distributed systems reliable.

After that, I spent a few years at SAP, focused on large-scale enterprise data, processes, and systems.

Learnings from these early years apply directly to my work now on autonomous AI agents that collaborate across networks and companies to automate tedious business processes.

In 2014, I became CTO of a hardware company that was transitioning into a platform for the Internet of Things. I designed that system and led the deployment of thousands of sensors and controllers embedded in city infrastructure, ports, and more worldwide.

Many of our sensors were machine-learning based and operated in extremely noisy environments, like city streets with power lines, buses, and interference, which taught me how to ship reliable products on top of probabilistic and non-deterministic behavior, a challenge that's acute in all AI products.

Because these systems were embedded in critical infrastructure, like cities and ports, security and trust had to be foundational, not bolted on.

That experience led me to build [Ockam](https://github.com/build-trust/ockam), a collection of protocols and an open-source Rust library for cryptographic identities, trust, and end-to-end encrypted communication in distributed systems.

Autonomy brings together all those experiences to empower teams to ship autonomous products that are dependable, secure, and highly scalable.

## Recent Articles

- [The flywheels of a software factory](/articles/the-flywheels-of-a-software-factory/index.md): A software factory begins by separating work that needs human attention from work agents can carry out autonomously. But orchestration alone only increases throughput. The deeper advantage comes from reinforcing loops that use product behavior, failures, agent traces, and engineering review to improve both the product and the factory.
- [How I built a self-improving software factory](/articles/how-i-built-a-self-improving-software-factory/index.md): Fluent is an open-source factory that autonomously turns your team's vision, ideas, bug reports, user feedback, production logs, and agent traces into working software.
- [Agent Sandboxes, how to create millions of them](/articles/agent-sandboxes/index.md): Long-horizon agents need their own workspace: a filesystem and a shell. A journey through linux isolation primitives and how we create millions of sandboxed workspaces in milliseconds.
- [Agent Identities, everything you need to know](/articles/agent-identities/index.md): Agents have raised the stakes: they take autonomous actions. Without cryptographic identity, we can't authenticate requests, authorize actions, or attribute decisions.
- [Agent Swarms, like the one Cursor created](/articles/agent-swarms-like-the-one-cursor-created/index.md): Deep autonomous work requires infrastructure, not just frameworks or harnesses. A demo of distributed AI agents reviewing Vue.js core.
- [How to make agents that succeed](/articles/how-to-make-agents-that-succeed/index.md): Controlled iteration is the only way to ship a reliable autonomous product and actually improve it over time.

## All Pages

- [Home](/index.md)
- [Articles](/articles/index.md)