From customer service chatbots to AI-assisted coding tools and document summarization systems, AI employees are already embedded in workflows across support, sales, product, and engineering.
While the front-end footprint of AI is growing, its role in actual operations remains shallow. Many current implementations rely on prompt-based assistants or isolated features, which are great at responding but weak at executing.
These tools often stop at the UI layer as they cannot take context-aware actions within live systems and frequently depend on human supervision for follow-through.
Krish Chelikavada and Keon Woo Kim, co-founders of Om Labs, are approaching AI from the opposite end: as an operational hire instead of an interface enhancement or reactive tool. They are developing autonomous AI employees that integrate into the production system and function as effective teammates.
Inside Om Labs, they’ve developed modular orchestration layers, memory-aware systems, and retrieval pipelines that make continuous AI execution possible. These components power real-time, contextualized decisions in areas like customer operation and testing, helping teams scale without expanding human headcount.
Krish and Keon’s experience building secure, low-latency, multi-agent crypto infrastructure has shaped how they now approach AI: as a systems problem first, not just a UX one.
A Systems-Born Founding Story: From MPC Wallets to AI Agents
Before entering the AI space, the duo built one of the fastest crypto infrastructure stacks in the industry, 0xPass. This distributed key management system, built on multi-party computation (MPC), achieved sub-second latency, outperforming leading protocols and attracting demand from companies serving over 10 million wallet connections.
Keon Woo Kim, whose engineering background includes building Uber’s high-volume incentive systems, led the design of 0xPass’s backend. He implemented a cryptographic framework based on DKLS23 and multi-party ECDSA signing, which allowed sensitive crypto operations to be securely executed across distributed nodes, maintaining performance without compromising security.
Krish Chelikavada’s background in product design, enterprise product strategy, and technical fundraising was vital in securing $1.9M in pre-seed capital from AllianceDAO, Balaji Srinivasan, and Soma Capital at a time when crypto venture capital has significantly stalled due to market downturn.
Despite gaining a significant foothold in the space, both founders saw a bigger opportunity: crypto was slowing while AI was accelerating. Rather than stay in a shrinking market, they made a deliberate call to pivot, turning down potential acquisition discussions and seed rounds to trust in their conviction that it was time to build in AI.
Support3: The AI Support Agent That Works Where Developers Live
The team’s first launch was Support3, an autonomous support agent built for Web3.
Instead of acting like a helpdesk chatbot, Support3 integrates directly into platforms like Discord, Slack, and Telegram, behaving like a real team member by engaging in multi-turn conversations that mirror the flow of real human support.
It monitors public and private channels in search of messages that appear to be support requests, such as bug reports, feature requests, and general questions. Once it finds them, Support3 analyzes the content (mainly, the nature of the problem and its priority) and tags it appropriately, helping teams avoid overlooking anything, even across high-volume or fragmented channels.
Support3 is powered by a memory-aware architecture that can retain context more thoroughly than regular models, so when a similar issue comes up, it replies immediately. For more technical questions that need follow-up, the agent scans the interface to collect the proper context before responding or creating a support ticket, ensuring the issue is tracked correctly and escalated.
The system tracks which questions are asked most often and highlights recurring bugs or pain points. As a result, internal teams across product and documentation can access these insights to tailor their user experience processes better and prevent future support issues. It even drafts FAQs and summarizes support trends to give teams the right tools to stay ahead of problems.
Within months of launch, Support3 was adopted by several industry unicorns that aim to use it to scale support without adding hires.
Jina: Autonomous QA at the Infrastructure Layer
Following the success of Support3, Krish and Keon introduced Jina, an autonomous QA agent that mimics how a real user would behave to stress-test software.
Unlike traditional tools that rely on brittle scripts, users can input a request through plain English text (like “complete checkout” or “book a flight”), and Jina will execute flows using a vision-based model that can directly interact with interfaces just like a real user.
When issues are found, Jina captures detailed context (steps taken, errors encountered, affected flows, and suggested fixes) and files reports directly to product or engineering teams.
Since it is not tied to CSS or DOM selectors, it automatically analyzes and readjusts to the flow of the UI and adapts when labels change or layouts shift, reducing the need for constant maintenance. It also embeds into CI/CD pipelines and runs tests on every build, making it a hands-off, always-on QA layer.
This system-level approach makes Jina ideal for fast-moving SaaS teams, internal dev tools, and user-facing applications where quality needs to keep pace with deployment.
The Underlying Principles Behind The Designs
Despite appealing to different use cases, both Support and Jina3 follow the same principles:
Latency and autonomy can both be directly integrated into production workflows, enabling quick and reliable decision-making without human approval.
Interface minimalism and functional depth: rather than introducing new interfaces, they can integrate through existing APIs, as well as message queues and permissioned task handlers that fit directly into day-to-day workflows.
Memory and grounding: both agents operate with persistent memory and retrieval-augmented generation (RAG), which provides a deeper contextual truth to its outputs and ensures it’s not only relying on surface-level prediction.
By following these baseline principles, the team is building full-stack agents with memory management, logic layers, and grounding via retrieval pipelines to guarantee predictable and accurate results.
A Shift from Tools to Workers
AI employees as a “copilot” has become a familiar idea, but Om Labs is pushing toward a more radical shift: agents as coworkers.
With Support3 and Jina, they’re building AI employees that businesses can embed inside core functions, whether that means running 24/7 QA cycles or answering Discord tickets like a tier-one support rep.
For developers and technical leaders, this invites opportunity: one where work isn’t just run by machines but delegated to them with full trust in their capabilities.
Rashan is a seasoned technology journalist and visionary leader serving as the Editor-in-Chief of DevX.com, a leading online publication focused on software development, programming languages, and emerging technologies. With his deep expertise in the tech industry and her passion for empowering developers, Rashan has transformed DevX.com into a vibrant hub of knowledge and innovation. Reach out to Rashan at [email protected]
