At CI&T, we help large enterprises transform the potential of AI into real business impact with AI Deployment, AI-native execution, and tech-integrated business solutions.
With 30 years of experience in technological transformation, we accelerate innovation with expertise in Agentic SDLC, Application modernization, Data & AI, Martech and Business strategy.
We are 8,000 CI&Ters across more than 25 countries, collaborating to build solutions with real impact. AI is already part of how we work, evolve, and innovate every day.
You will own the technical direction of the platform's agent SDK (`florence-agentic-ai`) and its framework adapters, the agent and MCP templates, and the runtime contract every agent on the platform depends on.
You decide what the SDK's public API is, how it changes without breaking dozens of teams, and how safety behaviors (loop guards, retry ownership, failure codes, delegation limits) hold the same way across Strands, Claude Agent SDK and LangGraph, over HTTP and A2A.
Responsibilities:
- Own the SDK as a product: public API, semantic versioning, deprecation policy, migration guides and release hygiene (builds you can reproduce, bounded dependencies, publishing from `main` only).
- Design and evolve the adapter and host contracts so that agents behave the same whatever the framework (Strands, Claude Agent SDK, LangGraph) or deployment target (AgentCore Runtime, AgentCore Harness, ECS Fargate).
- Lead multi-agent orchestration patterns: supervisor/worker and planner/executor/verifier designs, declared-peer delegation, A2A protocol support, and ending agent loops by design. Build runtime safety into the SDK: call-chain and hop limits, session cycle budgets, timeouts, per-peer circuit breakers, structured failures that alerts can page on.
- Define the extension model (hooks, per-request settings, escape hatches) and turn repeated escape-hatch use into supported features.
- Own the parity and contract test suite that gates every SDK and template release.
- Lead design reviews and RCAs. Turn incidents into lasting platform controls, not one-off fixes.
- Act as the leader of the teama and also mentor engineers
- Split large refactors into parallel tracks and deliver against fixed release dates.
Requirements:
- Advanced English proficiency (C1 or above), with autonomy to communicate directly with international stakeholders
- Extensive years of professional experiecne with software engineering and leading teams
- Experience building libraries, SDKs or platforms used by other engineering teams.
- Expert Python (async, typing, packaging, dependency management, pytest at scale).
- Hands-on production experience building LLM agents with at least one of: Strands Agents, Claude Agent SDK, LangGraph, OpenAI Agents SDK, CrewAI or similar.
- You understand the agent loop: tool calling, context management, turn limits, streaming. Experience with Model Context Protocol (MCP) servers and clients, and with agent-to-agent communication (A2A or equivalent).
- Track record of managing breaking changes in a widely used internal or open-source library: versioning, compatibility shims, migration tooling.
- Strong distributed systems fundamentals: timeouts, retries, idempotency, circuit breakers, backpressure, failure propagation.
- AWS experience, ideally Amazon Bedrock and Bedrock AgentCore (Runtime, Gateway, Memory, Identity).
- Clear technical writing: design docs, ADRs, migration guides.
Nice to Have:
#LI-LO1