Pathway 01 / Build
Generative AI Application Engineer
Strengthen the model, retrieval, evaluation, and product layer around your agents.
Explore generative AI →Career pathway / Workflow engineering
Agentic AI Automation Engineer
Design supervised agents that connect tools, systems, and approvals around a business process people can measure.
- Target roles
- Automation engineer, AI workflow builder
- Learner level
- Intermediate
- Format
- 10 business days, online-first cohort
- Commitment
- 6–8 hours each weekday; live sessions in CT
- Prerequisites
- Python basics, HTTP APIs, and comfort reading system diagrams
- Certificate
- Certificate of pathway completion
The study plan
Six layers of supervised autonomy.
Days 01–02Automation discovery
Find the process worth improving and the boundary where a person should stay in control.
- Process mapping, handoffs, and exception paths
- ROI, risk, and repeatability scoring
- Human-in-the-loop boundaries and approval design
Deliverable: an automation opportunity assessment with baseline measures and a risk register.
Day 03Agent architecture
Build an agent that can reason through a task without hiding its state or escape routes.
- Tools, planning loops, state, memory, and structured decisions
- Task contracts, schemas, and deterministic escape hatches
- Permission boundaries and graceful stop conditions
Deliverable: a single-agent task runner with explicit state and a human approval checkpoint.
Days 04–05Integrations and orchestration
Connect the agent to real systems in a way that survives retries and partial failure.
- APIs, webhooks, queues, authentication, and secrets
- Retries, idempotency, timeouts, and dead-letter handling
- Event boundaries and observable workflow state
Deliverable: a multi-system workflow with retry behavior and an operator-facing runbook.
Days 06–07Multi-agent patterns
Use multiple specialized agents only when coordination creates real value.
- Routing, delegation, shared context, and concurrency
- Conflict resolution, quality handoffs, and bounded roles
- When a deterministic workflow is the better design
Deliverable: a research-and-review agent team with a visible handoff trace.
Days 08–09Safety and evaluation
Probe the workflow before asking other people to rely on it.
- Permissions, sandboxing, and prompt injection defenses
- Audit trails, data boundaries, and unsafe tool-use tests
- Task-completion tests, regression suites, and human review
Deliverable: an agent red-team report with findings, mitigations, and residual risk.
Day 10Operating agents in production
Make autonomy observable, interruptible, and accountable in the operating environment.
- Monitoring, cost and latency budgets, and approval UX
- Incident response, rollback, and process owner communication
- Measuring time saved, quality, and exception rate honestly
Capstone: a supervised business-process automation with measurable ROI, safety case, and portfolio narrative.
By the end
Autonomy with a clear owner.
You will be able to map a process, choose where agents help, instrument the handoffs, and show whether the result is safer and more useful than the starting point.
- 01
Choose the right boundary
Separate repeatable work from judgment that deserves an accountable human.
- 02
Design for interruption
Build approval points, stop conditions, retries, and audit trails into the happy path.
- 03
Test the adversary
Use red-team scenarios and task-completion tests to make failures concrete.
- 04
Measure the process
Connect system traces to quality, exception rate, time saved, and cost.
“The red-team report became the strongest part of my portfolio because it showed judgment, not just orchestration.”
Mentor profile
Amara Okafor
Workflow systems architect focused on human approvals, resilient integrations, and the operational reality of agentic products.
Before you begin
Quick answers
Is this a course about autonomous agents doing everything?
No. The work centers supervised systems: clear permissions, useful approvals, observable state, deterministic fallbacks, and honest measurement of the process.
Will I receive a certificate?
Yes, a certificate of pathway completion when the required work is submitted. It is not a professional license and does not guarantee employment or income.
Related pathways
Build around the workflow.
Pathway 05 / Assure
Responsible AI, Security & Governance
Add risk, privacy, threat modeling, and assurance to the systems you automate.
Explore responsible AI →