Ajay Prakash

Prakash works on platforms and infrastructure for enterprise AI agents at LinkedIn. His QCon biography describes earlier leadership of AI platform and product work for Sales Navigator and his current focus on the agent platform. His personal writing and presentations explain how existing coding agents can work with a company’s internal systems and practices.[1][5]

Serve instructions alongside tools

Prakash coauthored LinkedIn’s account of Contextual Agent Playbooks & Tools with Nikhilesh Payyavuala. CAPT exposes both operational tools and reusable instructions through MCP. A playbook supplies the organization-specific steps for a task, then directs the agent to the relevant tools. The design augments existing coding assistants rather than requiring a separate model trained on every internal workflow.[2]

Put workflow coordination in code

Stratos makes agents, MCP tools and decision models available as TypeScript workflow components. Code handles branches, loops, parallel execution and passing data between steps, while agents handle the reasoning tasks assigned to them. Saved workflows can be reused and revised using traces from earlier runs. Its repository labels the project early-stage software and specifies its license.[3]

In his conference talk, Prakash explains why a growing inventory of tools and instructions can overwhelm an agent’s context. LinkedIn’s approach uses search, schema retrieval and execution as a smaller interface to the underlying inventory. Smaller playbooks can reference one another so the agent discovers context progressively. He also describes agents proposing repository changes when they encounter outdated instructions, preserving a path for review.[4]