Applied AI Engineer

Twenty•Arlington, VA
•$159,000 - $263,000•Onsite

About The Position

Twenty is hiring an Applied AI Engineer to build and maintain an agent environment with an interaction model similar to Claude Code and Codex. Users will provide tasks and context, review the agent’s work, monitor its actions, and intervene when needed. The role focuses on the agent harness: assembling context, executing tool calls, tracking progress, coordinating delegated work, carrying information across sessions, checking results, and recovering from failures. The engineer will write and maintain production software in Python, TypeScript, or Rust; investigate failed runs; test harness changes; and work with users and domain experts to improve the system. The work includes building agent execution loops, managing context, orchestrating workflows, defining tool interfaces, providing user controls, handling failures and access permissions, evaluating changes, and operating the system with secure traces, versioned prompts and workflow code, controlled releases, and rollback support. The position is onsite and full-time in a mission-focused company working on offensive cyber operations for the U.S. and its allies. U.S. citizenship is required; no active clearance is required for this role.

Requirements

  • At least four years of experience building and maintaining production software, including hands-on work on an LLM application that calls tools or performs multi-step tasks.
  • Strong production experience in at least one of Python, TypeScript, or Rust, with ability to work in the others as needed.
  • Ability to build services, work with asynchronous execution and external APIs, manage persistent storage, and write automated tests.
  • Experience writing or substantially changing an agent execution loop and explaining what the model controls, what the application enforces, and when users must decide.
  • Experience investigating agent failures involving tool use, lost context, or incomplete tasks.
  • Experience with retrieval, context assembly, summarization, or memory for an LLM application, including working within token limits and testing context retention.
  • Understanding of timeouts, retries, concurrency, and idempotency, with ability to design and verify recovery from partial workflow failures.
  • Experience building tests for model-driven workflows and using execution traces to investigate failures; ability to account for run variation and compare automated grading with human assessments.
  • Understanding of prompt injection and risks from instructions embedded in retrieved documents or tool outputs.
  • Ability to restrict tool access, protect credentials, and implement validation and approval steps for sensitive actions.
  • U.S. citizenship.

Nice To Haves

  • Experience with long-running or multi-agent workflows, including checkpoint recovery, task delegation, and shared state.
  • Experience building retrieval or memory systems with source tracking, freshness checks, and access controls tied to the requesting user’s permissions.
  • Experience running controlled comparisons or ablation studies to measure the effects of individual components.
  • Experience with self-hosted inference, inference optimization, or task-specific fine-tuning, including measuring effects on task performance and operating cost.

Responsibilities

  • Build agent harnesses that call models, run tools, record results, determine next steps, set time and cost limits, define stopping conditions, and detect unproductive repetition.
  • Manage the context supplied to models, including instructions, retrieved records, prior actions, and tool results; incorporate retrieval and summarization and investigate context-related failures.
  • Orchestrate model calls, application code, and human review, including branching, parallel execution, checkpoints, persistent task state, and delegated work.
  • Define tool inputs, outputs, and error responses; validate arguments and results; and implement retries that avoid duplicate actions.
  • Build user controls for inspecting progress and tool activity, correcting instructions, approving actions, stopping runs, and resuming unfinished work.
  • Detect invalid tool calls, dropped requirements, and inaccurate success reports; implement retry limits, cancellation, escalation, permission enforcement, and approval requirements.
  • Evaluate changes with representative tasks and production failures; measure completion, severity, human intervention, latency, and cost; and run regression tests.
  • Operate the system with traces, secret redaction, access restrictions, versioned prompts and workflow code, limited-traffic release testing, and rollback support.
  • Investigate failed runs and work directly with domain experts to understand task requirements, failure cases, acceptable errors, review needs, and user expectations.

Benefits

  • Medical, dental, and vision plan options.
  • Life and AD&D coverage options.
  • Disability coverage options.
  • Paid parental leave for eligible full-time employees: 12 weeks for birthing parents, 4 weeks for non-birthing parents, and 6 weeks for adoptive, foster, or intended parents through surrogacy.
  • Paid holidays and flexible PTO.
  • 401(k) with pre-tax and Roth options.
  • HSA/FSA options and dependent care FSA.
© 2026 Teal Labs, Inc
Privacy PolicyTerms of Service