Job Summary
About the Role
The deepest hands-on agent-architecture authority in the Forward Deployed organisation. You design how multiple agents coordinate, fail safely, and scale under uncertainty — then deploy them in client environments. This is an orchestration-architect role: the value is less in writing code and more in designing systems that decide, act, and recover without constant supervision, with the right guardrails and human oversight.
Key Responsibilities
What You'll Do
- Architect production-grade autonomous agent and multi-agent systems for enterprise clients
- Design orchestration, state management, and persistent-memory patterns for long-running autonomous tasks
- Lead client technical discovery and own end-to-end reference architectures, prototypes, and code reviews
- Build with Claude (API, Claude Code, MCP) and modern agent frameworks; set the engineering standards FDEs follow
- Embed guardrails, observability, evaluation harnesses, and human-in-the-loop controls
- Model token cost and design for AI unit economics from the architecture stage — know when not to use an agent
- Feed reusable patterns back into the CoE blueprint library and MSAB
Skill Requirements
What We're Looking For
- 8+ years in software/systems engineering, with hands-on production experience building LLM and agentic systems
- Strong distributed-systems intuition, security-model design, and cost-modeling ability
- Hands-on with at least one agent framework (Anthropic SDK, LangGraph, CrewAI, Autogen, or Semantic Kernel)
- Demonstrated eval-design rigor — the ability to test judgment, not just correctness
- Experience with state management and persistent-memory systems for autonomous tasks (e.g. Zep, Mem0)
- Excellent communication; able to engage client CTOs and lead cross-functional delivery
Other Requirements
Nice to Have (future-ready)
- Strong opinions on agent-vs-no-agent trade-offs and safe-failure design
- Experience with MCP integration patterns and multimodal / real-time RAG pipelines
- Published reference architectures, open-source contributions, or eval frameworks
Key Performance Parameters
01 Reference architectures — ≥5 reusable agent reference architectures adopted by delivery teams
02 Win contribution — Architecture cited as a differentiator in ≥3 competitive wins per period
03 Prototype velocity — Time-to-first-working-agent-prototype <4 weeks per engagement
04 Quality — Zero Tier-1 architecture review failures; eval coverage on all deployed agents
05 Cost discipline — Token / unit-economics modelling built into every architecture proposal