When does class start/end?
Class hours may vary, please reach out to contact@ascendientlearning.com if you have any questions.
Designing Agentic AI Systems
Design agentic AI systems with bounded autonomy, named failure modes, and clear ownership. Agentic fit is sized against blast radius, reversibility, human-review boundaries, and per-step cost and...
Read More
Duration
3 days
Course Code
GAI-2701
Available Formats
Classroom
Overview
Course Description
Design agentic AI systems with bounded autonomy, named failure modes, and clear ownership. Agentic fit is sized against blast radius, reversibility, human-review boundaries, and per-step cost and latency budgets that keep multi-agent loops from runaway economics. Topology and reasoning vocabulary span single-agent, supervisor-worker, peer-collaboration, and hybrid topologies plus the patterns inside one agent — ReAct, Reflexion, Plan-and-Execute, evaluator-optimizer — alongside event-driven orchestration, shared versus isolated state, summary and episodic memory, and hand-off contracts that survive agent boundaries. Integration, failure modes, and observability draw on tool design, the protocol stack including MCP, A2A, and ACP, the Lethal Trifecta rule, prompt injection, runaway loops, sandboxed execution, circuit-breakers, human escalation, trace design, evaluation harnesses, and metrics for non-deterministic behaviour. Governance and ROI work through ownership boundaries between platform, application, and risk teams, change-control patterns, business-case sizing, and the anti-patterns that separate real fit from cargo-cult adoption. Hands-on labs produce topology selection workshops, hand-off contract drafts, threat-modelling exercises, observability designs, and an end-to-end agentic-system review. The course is designed for solution architects, AI architects, and technical leads designing agentic systems in production.
Skills Gained
By the end of this course, participants will be able to:
- Assess agentic fit against autonomy, blast radius, and reversibility constraints
- Select an agent topology plus orchestration pattern grounded in documented design criteria
- Design context, memory, and hand-off contracts that survive agent boundaries
- Threat-model adversarial plus operational failure modes for agentic systems
- Specify observability and evaluation that fits non-deterministic agent behaviour
- Articulate the ROI case and named anti-patterns that gate agentic investments
Who Can Benefit
This course is designed for:
- Solution Architects
- AI Architects
- Technical Leads
Prerequisites
Participants should enter this course with:
- GAI-1701 or equivalent
- Familiarity with agentic AI concepts
Organizational Objectives
This course assists organizations to:
- Reduce agentic-project failure rate through a shared autonomy-boundary rubric applied at design time
- Lower production-risk surface by threat-modelling agents before deployment
- Improve governance clarity through documented ownership boundaries between platform, application, and risk teams
- Build architecture capability for agentic systems across the AI-architect function
- Reduce wasted spend by gating agentic proposals against named anti-patterns
Software
All attendees must have a modern web browser and an Internet connection.
Course Details
Course Details
Assessing Agentic Fit and Autonomy Boundaries
By the end of this module, you will be able to separate tasks that benefit from agency from those better served by classical pipelines, size autonomy against blast radius and reversibility, and set per-step cost and latency budgets that keep multi-agent loops from runaway economics.
- Agentic fit versus pipeline fit — the test for when agency adds value
- Blast radius, reversibility, and the human-review boundary
- Per-step cost and latency budgets — max-depth, token-budget propagation, model-tiering
- Compensating transactions and idempotency for irreversible tool calls
- Workflow patterns (prompt chaining, routing, parallelisation) as agency-free alternatives
- The “start with the simplest workflow” rule before reaching for an agent
- <b>Hands-on Lab:</b> Apply the agentic-fit test plus a budget worksheet to three candidate workloads and produce an autonomy-boundary recommendation for each that names blast radius, reversibility, and per-step budgets.
Designing Agent Topology and Orchestration
By the end of this module, you will be able to select an agent topology and orchestration pattern grounded in documented design criteria, name the reasoning patterns that live inside one agent, and choose the simplest topology that satisfies the workload.
- Topologies across agents — single-agent, supervisor-worker, peer collaboration, hybrid
- Reasoning patterns inside one agent — ReAct, Reflexion, Plan-and-Execute, evaluator-optimizer
- Orchestration shapes — sequential, conditional, hierarchical, event-driven
- Start-with-supervisor then graduate to swarm — when topology change is justified
- Multi-agent cost arithmetic — 5-15× token cost over single-agent equivalents
- Open-source and vendor agent frameworks as reference implementations
- <b>Hands-on Lab:</b> Run a topology selection workshop on a candidate workload that picks one topology and one reasoning pattern, names the criteria driving the choice, and rejects two alternatives with reasons.
Designing Context and Memory Across Agent Boundaries
By the end of this module, you will be able to design context and memory that survive agent boundaries, choose between shared and isolated state, and write hand-off contracts that survive both summary loss and bad-faith hand-offs.
- Shared versus isolated state across agents
- Summary memory, episodic memory, and hybrid memory patterns
- Hand-off contracts — objective, output format, tool scope, task boundary
- Context engineering — what to load, what to defer, what to forget
- Memory services as platform primitives versus per-agent stores
- Failure modes — duplicate work, wrong-era investigations, summary loss
- <b>Hands-on Lab:</b> Draft hand-off contracts for a three-agent workflow with explicit objectives, output formats, tool scopes, and task boundaries that prevent the duplicate-work and wrong-era failure modes.
Integrating Agents with Tools, Protocols, APIs, and Events
By the end of this module, you will be able to integrate agents with external tools through the agent-tool protocol stack, design tool interfaces that constrain agent capability, and apply OAuth 2.1 scopes and per-tool ACLs that survive a hostile prompt.
- The three-layer protocol stack — MCP for tools, A2A for agent peers, ACP for local orchestration
- Tool design — explicit objectives, output schemas, side-effect contracts
- OAuth 2.1, PKCE, token exchange, and fine-grained per-tool ACLs
- Event-driven coupling versus synchronous tool calls
- Function-calling JSON schemas, OpenAPI, and AsyncAPI as tool-exposure conventions
- MCP gateways as enforcement points
- <b>Hands-on Lab:</b> Design the tool-integration layer for a candidate agent including the protocol choice for each surface, the OAuth scope set per tool, and the ACL boundary each tool will enforce.
Handling Failure Modes, Adversarial Threats, and Resilience
By the end of this module, you will be able to threat-model an agentic system against current adversarial and operational failure modes, apply the Lethal Trifecta rule, and design resilience through sandboxed execution, circuit-breakers, and structured escalation.
- The Lethal Trifecta — private data, untrusted content, external communication
- OWASP Top 10 for Agentic Applications and MITRE ATLAS for agent threats
- Prompt injection, runaway loops, and tool-call hijacking
- Sandboxed execution for generated code — microVMs, isolated runtimes
- Circuit-breakers and rate caps as architectural primitives
- Structured human escalation and trust calibration over a session
- <b>Hands-on Lab:</b> Threat-model a candidate agentic system using the Lethal Trifecta plus the OWASP Agentic Top 10 and produce a resilience design that names sandboxes, circuit-breakers, and escalation triggers.
Observing and Evaluating Agentic Systems
By the end of this module, you will be able to specify observability and evaluation that fit non-deterministic agent behaviour, name the metrics that matter, and design trace structures that survive long-running multi-step trajectories.
- Trace design for agents — trajectory spans, tool spans, hand-off spans, evaluator spans
- OpenTelemetry GenAI semantic conventions as the emerging trace schema
- Agent metrics — tool-call accuracy, plan validity, hand-off integrity, refusal calibration
- Cost-per-resolution, p95 latency, runaway-loop rate as operational SLOs
- Evaluation harnesses — offline benchmarks, judge-LLM rubrics, online judges on traces
- Agent benchmarks as reference points — BFCL, GAIA, SWE-bench, ?-bench
- <b>Hands-on Lab:</b> Design observability and evaluation for a candidate agentic system including the trace shape, the metric set, the SLO targets, and one offline plus one online evaluation that gates production.
Governing Agentic Systems - Ownership and Team Boundaries
By the end of this module, you will be able to define ownership boundaries between platform, application, and risk teams for an agentic system, apply Team Topologies interaction modes, and design change-control that fits non-deterministic behaviour.
- Ownership boundaries — platform team, application team, risk team
- Team Topologies interaction modes applied to agent platforms
- Change-control patterns for prompts, tools, models, and agent topologies
- Approval gates that match autonomy boundaries from Module 1
- Agent identity and provenance — know-your-agent, signed credentials
- Governance for agent-to-agent workflows across organisations
- <b>Hands-on Lab:</b> Draft an ownership-and-change-control plan for a candidate agentic system that names the responsible team for each artefact, the interaction mode between teams, and the approval gates per change class.
Framing ROI and Anti-Patterns for Agentic Investments
By the end of this module, you will be able to size an agentic-system business case honestly, name the anti-patterns that separate real fit from cargo-cult adoption, and produce an end-to-end review that gates an agentic investment.
- ROI sizing for agentic systems — cost-per-resolution versus alternative pipelines
- The token-cost multiplier — 5-15× observed in multi-agent production
- Anti-patterns — agents-where-pipelines-fit, supervisor-with-no-veto, eval-after-deploy, tool-sprawl
- Industry signals — analyst forecasts on agentic cancellation rates and failure-rate framing
- Static-context and complacency as agentic anti-patterns
- Synthesising the course — end-to-end agentic-system review as gating artefact
- <b>Hands-on Lab:</b> Run an end-to-end agentic-system review on a candidate workload that pairs a ROI estimate with a named-anti-pattern check and produces a Go, No-Go, or Reshape recommendation.
Schedule
FAQ
Does the course schedule include a Lunchbreak?
Classes typically include a 1-hour lunch break around midday. However, the exact break times and duration can vary depending on the specific class. Your instructor will provide detailed information at the start of the course.
What languages are used to deliver training?
Most courses are conducted in English, unless otherwise specified. Some courses will have the word "FRENCH" marked in red beside the scheduled date(s) indicating the language of instruction.
What does GTR stand for?
GTR stands for Guaranteed to Run; if you see a course with this status, it means this event is confirmed to run. View our GTR page to see our full list of Guaranteed to Run courses.
How do I find an Ascendient Learning training location?
We have training locations across the United States and Canada - View a complete list of classroom training locations.
Which delivery formats are available?
At Ascendient Learning, we offer training that is Instructor-Led, Online, Virtual, and Self-Paced.
Does Ascendient Learning deliver group training?
Yes, we provide training for groups, individuals and private on sites. View our group training page for more information.
What does vendor-authorized training mean?
As a vendor-authorized training partner, we offer a curriculum that our partners have vetted. We use the same course materials and facilitate the same labs as our vendor-delivered training. These courses are considered the gold standard and, as such, are priced accordingly.
Is the training too basic, or will you go deep into technology?
It depends on your requirements, your role in your company, and your depth of knowledge. The good news about many of our learning paths, you can start from the fundamentals to highly specialized training.
How up-to-date are your courses and support materials?
We continuously work with our vendors to evaluate and refresh course material to reflect the latest training courses and best practices.
Are your instructors seasoned trainers who have deep knowledge of the training topic?
Ascendient Learning instructors have an average of 27 years of practical IT experience and have also served as consultants for an average of 15 years. To stay current, instructors spend at least 25 percent of their time learning new, emerging technologies and courses.
Do you provide hands-on training and exercises in an actual lab environment?
Lab access is dependent on the vendor and the type of training you sign up for. However, many of our top vendors will provide lab access to students to test and practice. The course description will specify lab access.
Will you customize the training for our company’s specific needs and goals?
We will work with you to identify training needs and areas of growth. We offer a variety of training methods, such as private group training, on-site of your choice, and virtually. We provide courses and certifications that are aligned with your business goals.
How do I get started with certification?
Getting started on a certification pathway depends on your goals and the vendor you choose to get certified in. Many vendors offer entry-level IT certification to advanced IT certification that can boost your career. To get access to certification vouchers and discounts, please contact info@ascendientlearning.com.
Will I get access to content after I complete a course?
You will get access to the PDF of course books and guides, but access to the recording and slides will depend on the vendor and type of training you receive.
How do I request a W9 for Ascendient Learning?
View our filing status and how to request a W9.
Reviews
I like their training. A lot of material covered. The labs are very good. l learned a lot.
I think the platform is very good and look forward to taking my next course in early October.
The training was good but needed the basic skills of maximo before getting deep in the configuration of it.
the interface was super easy to use and the instructions to get ready for the course was also very simple and easy to understand.
Great and very intuitive. Better than the traditional hit the wrong button/lose points.
