Lead AI/Agent Architect

Future Works is a US-focused, AI-native professional services firm that builds operational AI and data systems. They are seeking a Lead AI/Agent Architect to design and build an AI agent system for managing complex transaction workflows for a large enterprise client, ensuring a seamless integration of human oversight and automation.

Responsibilities:

• Agent Orchestration Design: Own the orchestrator agent that holds the deal state machine and routes work to one specialist agent per stage, defining stage handoff protocols and cross-stage integration
• Tool-Use Architecture: Design the typed tool layer (template population, market-data query, financial analytics, clause extraction, document assembly) as the integration seam that binds to sample data in the sandbox and to live systems in production, so the move to production is a connector change rather than a rebuild
• Human-in-the-Loop Protocol: Define the mandatory approval gates (engagement agreement issuance, LOI execution, lease signature, invoice issuance) and review checkpoints (requirements-brief sign-off, short-list, negotiation positioning), ensuring the agent never executes binding legal or financial actions without broker confirmation
• Secure Sandbox & Data Governance: Own the sandbox data model and test scenarios, and define the security, compliance, and data-governance approach: data residency, PII handling, audit logging, and a zero-retention policy with no training on client or deal data
• Feasibility Validation: Validate the technical feasibility of the automation hypotheses stage by stage and translate the automation boundary findings into architecture
• Extensibility by Design: Architect the system with explicit extensibility to Agency (Landlord Rep) workflow and other brokerage lines, producing a platform foundation rather than a point solution
• Cross-Functional Technical Leadership: Lead and direct the AI/ML Engineers through the build phases, managing orchestration and cross-stage integration while engineers build per-stage capabilities in parallel