New opportunity

Senior AI Engineer / Agent Development Lead

Acosta Group · Oak Brook, United States

About this role

Acosta Group seeks a Senior AI Engineer / Agent Development Lead in Oak Brook, IL, to build reusable AI agents, copilots and workflow automations for priority enterprise and business-unit use cases. Working under AI Architecture and AI Solutions Delivery leadership, the role supports an AI center-of-excellence model with central guardrails and federated delivery.

Responsibilities

Design, build, test and support agents, copilots, RAG workflows, prompts, APIs, integrations, evaluation assets and reusable AI components using approved enterprise platforms and governance standards.
Partner with AI delivery, architecture, data, technology, cyber and business stakeholders to implement use cases. Establish development patterns that improve delivery speed, consistency, security and maintainability.
Help move prototypes into production through testing, deployment, monitoring, documentation, release management and operational handoff. Integrate approved enterprise data sources, applications, collaboration and workflow tools, APIs and knowledge repositories.
Apply responsible AI and security requirements for data handling, access controls, human oversight and auditability. Troubleshoot agent behavior, prompts, retrieval, integrations, workflows and performance; share technical guidance and examples with internal teams, citizen developers and delivery partners.

Qualifications

At least 7 years of hands-on experience in software, AI, automation, data or cloud engineering, or enterprise application development; at least 3 years building AI-enabled applications, automations, agents, copilots, RAG or ML-enabled systems, APIs, enterprise integrations or intelligent workflow orchestration.
Strong programming, scripting, integration, testing, debugging and deployment skills. Practical familiarity with GenAI, LLMs, retrieval, embeddings, prompt engineering, tool calling, multi-step reasoning, evaluation, monitoring and responsible AI guardrails. Hands-on experience designing agentic workflows with enterprise data access, human-in-the-loop controls and exception handling.
Familiarity with agent platforms or frameworks such as Microsoft Copilot Studio, Azure AI Foundry, Semantic Kernel, AutoGen, LangGraph, LangChain, CrewAI, LlamaIndex or OpenAI Agents SDK. Experience with cloud, enterprise or data platforms, automation tools, or low-code/no-code or collaboration platforms. Strong problem-solving, documentation and communication skills are required.

This is a full-time Corporate position with a stated salary range of $105,000–$140,000. The posting notes possible lifting of materials or equipment weighing less than 15 pounds.

For full details about this role, click “Apply on company site” to view the employer’s original posting.

Skills for this role

AI agentsAgent orchestrationGenerative AILarge language modelsCopilotsRetrieval-augmented generationPrompt engineeringEmbeddingsTool callingHuman-in-the-loop designAI evaluationAI monitoringResponsible AIWorkflow automationAPI integrationSoftware testingDebuggingDeploymentMicrosoft Copilot StudioAzure AI FoundrySemantic KernelAutoGenLangGraphLangChainCrewAILlamaIndexOpenAI Agents SDKCloud platformsData engineeringAccess controls

Checking your profile…

Preparation suggestions, not the employer’s actual interview questions. Always check the original posting for current requirements.

Understand the role

What matters most

Build reusable agent and RAG capabilities
The posting calls for hands-on design, testing, and support of agents, copilots, RAG workflows, prompts, APIs, evaluation assets, and reusable components for enterprise and business-unit use cases.
Scale delivery within enterprise guardrails
The role supports a center-of-excellence model with central guardrails and federated execution. It calls for approved platforms and development patterns that improve speed, consistency, security, and maintainability.
Move solutions into production
The posting emphasizes testing, deployment, monitoring, documentation, release management, operational handoff, and troubleshooting agent behavior, retrieval, integrations, and performance.
Coordinate secure integrations and enable other teams
The role integrates approved data sources and applications, applies access controls, human oversight, and auditability, and shares working patterns with internal teams, citizen developers, and delivery partners.

Likely challenges to think through

Reuse versus use-case fit
Preparation hypothesis: shared components may speed parallel delivery but fit different business-unit workflows imperfectly. Prepare to explain how you would define a reusable core, allow bounded variation, and test each use case against its requirements.
Agent autonomy versus enterprise control
Preparation hypothesis: tool use and multi-step reasoning may conflict with data-access limits and human oversight. Prepare a design walkthrough covering permission boundaries, exception handling, approval points, and auditable actions.
Prototype speed versus operational reliability
Preparation hypothesis: a working demo may not yet be supportable in production. Prepare to describe how evaluation, failure testing, monitoring, documentation, and handoff would expose and address that gap.

Practice your answers

Technical practice

How would you design an agentic workflow that plans steps, calls tools, and handles exceptions?

What to coverSuggested answer focus: Describe how you would bound the agent’s decisions, validate tool inputs and outputs, and route failures or uncertain actions to a person. Explain the trade-off between flexible reasoning and predictable, supportable workflows.

How would you build a RAG workflow over an approved enterprise knowledge repository?

What to coverSuggested answer focus: Cover document preparation, embeddings, retrieval, access-aware results, and how the response uses retrieved material. Explain how you would investigate stale, irrelevant, or missing results.

How would you improve a copilot whose prompts produce inconsistent answers or tool calls?

What to coverSuggested answer focus: Discuss prompt structure, clear tool definitions, test cases, and inspection of the steps leading to an answer. Distinguish a prompt problem from a retrieval, integration, or model-behavior problem.

How would you integrate an agent with enterprise applications and APIs?

What to coverSuggested answer focus: Describe approved data access, authentication, input validation, error handling, and the limits of what the agent may do. Explain how you would make integration failures visible without exposing sensitive data.

Where would you place human oversight and responsible-AI controls in an agent workflow?

What to coverSuggested answer focus: Identify actions or uncertain outputs that warrant review, alongside data-handling, access-control, and auditability needs. Explain the trade-off between automation speed and appropriate oversight.

How would you evaluate an agent or RAG workflow before release?

What to coverSuggested answer focus: Propose representative test cases and measures for answer quality, retrieval, tool use, failures, and guardrail behavior. Explain how you would use the results to decide what needs improvement without claiming a specific company release threshold.

How would you diagnose a slow or incorrect multi-step agent run?

What to coverSuggested answer focus: Trace the prompt, retrieval results, tool calls, integration responses, and timing to isolate the fault. Discuss fixes and how you would test that they improve behavior without creating new failures.

What would you do to move an AI prototype into production support?

What to coverSuggested answer focus: Cover testing, deployment, monitoring, documentation, release management, and operational handoff, all named in the posting. Explain what evidence would help the receiving team troubleshoot and maintain it.

How would you design reusable components for teams delivering different AI use cases?

What to coverSuggested answer focus: Discuss shared patterns for integrations, evaluation, controls, and documentation while allowing use-case-specific configuration. Explain how you would balance faster federated delivery with central standards and maintainability.

How would you choose an approved agent platform or framework for a use case?

What to coverSuggested answer focus: Compare options against orchestration needs, enterprise integration, governance, testing, and operational support rather than assuming one named framework is required. Explain how you would work within approved platforms and existing templates.

Behavioural practice

Tell me about a time you turned an ambiguous business request into a working technical solution.

What to coverSuggested practice: Use a real situation, the questions you asked, and the actions you took to define scope and document the solution. Give the result and explain how you handled remaining uncertainty.

Describe a time you aligned engineering, security, data, and business stakeholders on a delivery decision.

What to coverSuggested practice: Set out the situation, competing priorities, and your actions to reach a workable decision. State the result and how you communicated the trade-offs.

Give an example of handing a prototype or new capability over for ongoing support.

What to coverSuggested practice: Explain the situation, your actions on testing, documentation, monitoring, or handoff, and the result. Be specific about what made the capability supportable.

Tell me about a time you helped another team adopt a reusable technical pattern.

What to coverSuggested practice: Describe the audience and problem, the guidance or working example you shared, and the result. Explain how you made the pattern useful without ignoring that team’s needs.

Describe a time you investigated a difficult failure in an AI, automation, or integrated workflow.

What to coverSuggested practice: Use a real situation/action/result example showing how you narrowed down the cause and coordinated a fix. Include how you communicated the outcome and reduced the chance of recurrence.

Sharpen your knowledge

Topics to prepare

Agent orchestration and control flow
The posting calls for planning, multi-step reasoning, tool calling, human oversight and exception handling. Sketch an agent workflow and explain when it should call a tool, request approval, retry or stop.
Retrieval and embeddings
RAG workflows, retrieval and embeddings are explicitly listed. Trace how a question moves through retrieval to an answer, then identify where irrelevant results or unsupported claims could arise.
Evaluation and operational monitoring
The role includes evaluation assets, monitoring and troubleshooting. Define what you would measure for answer quality, tool failures and workflow performance, and how a failed case would be investigated.
Responsible AI and enterprise security
The posting specifies data handling, access controls, human-in-the-loop design and auditability. Review a proposed agent workflow and mark where permissions, approval and auditable records are needed.
Reusable delivery under central guardrails
Acosta Group describes central guardrails with federated execution and asks for reusable, maintainable patterns. Compare a one-off prototype with a reusable component, noting which standards, documentation and handoff details would need to be shared.

Skills to revise

Implement an agent workflow
The role requires hands-on agentic workflows. As a practice exercise, implement a small multi-step flow with a tool call, an approval point and an exception path; test each branch.
Build and test a RAG path
The posting calls for RAG workflows and prompt development. As practice, connect a small permitted document set to a retrieval-backed prompt and test how it handles relevant, irrelevant and missing information.
Integrate an API with controlled data access
API integration and approved enterprise data access are explicit responsibilities. Practice a mock integration that checks access before a call and handles timeouts and denied requests without exposing data.
Debug with repeatable evaluations
The role involves testing and troubleshooting prompts, retrieval, integrations and performance. Create a compact test set, reproduce a failure, isolate its cause and record the before-and-after result.
Prepare a production handoff
The posting calls for deployment, monitoring, documentation, release management and operational handoff. Draft a sample release checklist covering tests, monitoring signals, known exceptions and support instructions.

Your preparation checklist

Map evidence to the role
The posting asks for 7+ years of relevant engineering experience and 3+ years building AI-enabled systems. Prepare a concise timeline and specific examples of programming, APIs, integrations, testing, debugging and deployment; identify any gaps without overstating experience.
Prepare an agent project walkthrough
Choose a real project, if available, and explain the business requirement, your contribution, architecture, tool calls or retrieval, enterprise data access, exception handling and outcome. Bring a simple diagram and distinguish what you built from what the team built.
Practice a governed workflow design
As an interview exercise, sketch an agent or copilot that uses an approved knowledge source and API. Show access controls, data handling, human review, audit records and how it would fit central guardrails with federated delivery; label platform choices as proposals, not Acosta standards.
Show how you evaluate and troubleshoot
Prepare a small test-and-evaluation plan covering prompt quality, retrieval relevance, agent behavior, integration failures and performance. Explain what evidence you would inspect, how you would reproduce failures and when you would escalate or require human intervention.
Outline a production handoff
The role supports prototypes moving into production. Prepare an example checklist for testing, deployment, monitoring, documentation, release management and operational ownership, noting which parts you have personally handled.
Explain a reusable development pattern
Prepare one concrete pattern or component you could share with engineers or citizen developers, such as a guarded tool-integration template. Show how documentation and examples could improve consistency, security and maintainability across use cases.
Draft questions for the employer
Ask which enterprise platforms and governance standards are approved, how Architecture and Solutions Delivery divide decisions, what use cases are priorities, and what evidence is needed for production handoff. These are questions to clarify, not assumptions about Acosta’s processes.

Availability last checked: Oct 17, 2026. Status can change after a check; confirm on the employer’s website. Report an issue · Our checking policy

KEEP EXPLORING

Similar AI jobs

Related skills and specializations in United States. Matched to this role, not your profile.

Explore more jobs

Finding similar opportunities…