Shashi Jagtap is the founder and chief executive of Superagentic AI, where he develops infrastructure that helps autonomous coding agents navigate complex software systems. Before founding the company, he spent nearly six years at Apple working on Xcode automation, XCTest and continuous-integration infrastructure.
Jagtap built his early career around mobile DevOps and test automation, publishing technical work on iOS simulators, release pipelines and Apple-platform tooling through XCBlog. He founded XCTEQ in 2018 and joined Apple’s Xcode organization the following year.
After leaving Apple in April 2025, he founded London Agentic AI, a practitioner community focused on agent frameworks, interoperability and evaluation, and incorporated Superagentic AI on April 28. His account of the company’s first year frames its mission around Agent Experience: improving the tools, protocols, memory, evaluation and execution environments available to autonomous agents.
- Treat codebases as programmable environments. Jagtap applies Recursive Language Models to repositories whose source files, tests, dependencies and configuration overwhelm conventional prompting. Agents inspect repository structure through generated code inside a REPL, return bounded observations and delegate focused questions to additional model calls. His AI Engineer World’s Fair session applies this architecture to unfamiliar monorepos, repository onboarding and root-cause analysis.
- Make coding-agent decisions inspectable. His open-source RLM Code implements recursive reasoning as an independent coding-agent harness with sandboxed execution, configurable recursion depth, spending limits and replayable agent trajectories. Portable JSONL traces capture generated code, delegated queries, token usage and final outputs, allowing engineers to investigate how an agent reached its conclusions.
- Optimize the complete agent environment. Jagtap’s SuperOpt research examines how prompts, tool interfaces, retrieval pipelines, execution protocols and persistent memory jointly influence agent reliability. His SuperRadar project helps practitioners evaluate the expanding ecosystem of agent frameworks, orchestration systems and coding tools.
