Other

AI Workflow Engineer – – AI-Native Operations

Improving Corporate ServicesHouston, TX🇺🇸United StatesPosted 25 Aug 2026

Quick Overview

Seniority

Mid Senior

Work mode

Hybrid

Location

Houston, TX, United States

Posted

2 days ago

AzureComplianceDatabricks

Job Description

Role Overview

Improving is seeking a hands-on AI Workflow Engineer to serve as an embedded builder within our AI-Native Operations initiative. This person will turn defined operational opportunities and technical specifications into production-ready AI agents and workflows. This role owns execution: identifying the specific operational target, building the agent or workflow, testing and hardening it, and shipping it into production.

What You Will Build

This role will develop and deliver both Stage 3 tasks and Stage 4 workflows:

Stage 3 – Task: An individual uses AI to complete a defined task that includes a written evaluation rubric, measurable before-and-after evidence, and a version-controlled task library.
Stage 4 – Workflow: Multiple AI-driven steps are connected into an end-to-end workflow and surrounded by deterministic validation, adversarial testing, governance controls, and mandatory human punch-out points that cannot be bypassed.

You will deploy and register these solutions in Agent 365 and build them within the architectural tier assigned by Improving’s technical leadership:

Bedrock
Foundational
Functional
Transactional
Operational

Key Responsibilities

Build and ship production-ready Stage 3 tasks and Stage 4 AI workflows.
Implement chained, multi-step AI processes based on architecture and technical direction established by Improving’s technical leadership.
Build deterministic validation around AI-generated outputs rather than relying exclusively on prompts or model behavior.
Establish adversarial tests that deliberately attempt to break, confuse, or bypass the workflow.
Implement governance rules, security controls, and mandatory human punch-out points that cannot be circumvented.
Test workflows end to end using realistic scenarios, failure conditions, and edge cases.
Deploy and register agents in Agent 365.
Maintain accurate agent inventory, ownership, lifecycle status, and compliance reporting.
Build solutions using Azure AI Foundry, Copilot Studio, Databricks, and related technologies in accordance with established architecture decisions.
Use LangChain or LangGraph to develop the chained, branching, and multi-step logic supporting Stage 4 workflows.
Work with Entra ID to support appropriate identity, access, and security requirements.
Coach individual Improvers in applying Stage 3 rigor, including written evaluation rubrics, before-and-after evidence, and version-controlled task libraries.
Document and transfer completed workflows so the owning team can operate and support them independently.
Escalate architecture, platform placement, security, and governance questions to the designated technical leader before implementing workarounds or making independent design decisions.
Deliver working solutions within days or weeks whenever the scope allows.

Required Background

Hands-on software engineering experience delivering and supporting production software—not only prototypes, demonstrations, or proofs of concept.
Experience building, testing, deploying, and maintaining AI-enabled applications, agents, or automated workflows.
Working knowledge of Azure AI Foundry, Microsoft Copilot Studio, Microsoft Entra ID, and Databricks.
Experience using LangChain or LangGraph to build chained, multi-step agentic workflows.
Ability to develop deterministic validation and evaluation logic around AI-generated output.
Understanding of AI evaluation methods, failure conditions, adversarial testing, and human-in-the-loop controls.
Experience integrating applications with APIs, enterprise data sources, identity systems, and business processes.
Ability to work effectively within an architecture and technical direction established by a lead architect.
Willingness to raise architectural or governance concerns while avoiding unsupported workarounds or independent architectural decisions.
Strong bias toward execution, rapid iteration, and shipping production-ready solutions.
Ability to document completed solutions and prepare operational teams to own them.

What Success Looks Like

Success is measured by a shipped workflow—not a proposal, presentation, architecture diagram, or prototype.

A successful solution:

Runs reliably from beginning to end in a production environment.
Includes working deterministic validation, governance controls, and mandatory human punch-out points.
Has been tested against realistic attempts to break, manipulate, or bypass its controls.
Handles failures, uncertainty, and exceptions safely.
Is properly deployed, registered, documented, and maintained in Agent 365.
Meets the architecture and placement requirements established by Improving’s technical leadership.
Can be operated and supported by the owning team without requiring the builder to remain involved.