AWS Jam: new Kiro and AgentCore training workshops

On September 30, 2026, AWS announced three new AWS Jam experiences for agentic AI development. For developers and technical leads, the useful decision is which practical training matches their work: development with Kiro, deployment with AgentCore, or multi-agent systems. These workshops are not presented as new AWS Certified certifications.

This analysis, checked on October 20, 2026, draws on the official AWS announcement (opens in a new tab) and the related course outlines. The announcement dates from September; it is not a launch announced today. Here is how to compare the options before allocating time or training budget.

Three workshops for three different problems

The announcement lists the following formats. Challenge counts and durations refer to the Jam workshops, rather than certification exams.

The workshops can run independently or alongside their corresponding courses. Distinguish a practice session from a course that teaches the concepts: their durations are not automatically combined into a particular commercial package.

Our recommendation is to start with an observable difficulty. Consider Kiro if generated changes lack clear acceptance criteria. Look at AgentCore if your prototype works but its access decisions remain difficult to explain. Examine the advanced option if several agents already coordinate work and diagnosing their behaviour has become a problem.

Kiro: prepare specifications and change reviews

The official Spec-Driven Development with Kiro course outline (opens in a new tab) describes two days of intermediate training. It covers feature and bug-fix specifications, steering files, MCP, hooks and team governance. Recommended foundations include programming, Git, REST APIs and generative AI concepts.

That outline helps explain the workshop's direction; it is not a guaranteed list of Jam challenges. To assess your own needs, choose a recent correction from your project and ask what you wrote down before changing the code.

Can you describe the current behaviour, the expected behaviour and the cases that must remain unchanged? Do you have an example that reproduces the problem? Can you explain how a reviewer will determine that the correction is sufficient? Answering those questions is a more useful preparation activity than collecting generic prompts.

For a personal exercise, select a narrow defect, such as a search feature handling an empty value incorrectly. Write down the expected result and a neighbouring case that must remain unchanged. Then compare your specification with the proposed change. This OnLevelUp exercise does not reproduce AWS material: its purpose is to reveal assumptions that you usually leave unstated.

AgentCore: move from a prototype to verifiable behaviour

The Building Agentic AI with Amazon Bedrock AgentCore outline (opens in a new tab) describes an intermediate, one-day course and recommends Agentic AI Foundations. Its scope includes execution, identity, tools, memory and observing agent behaviour.

To prepare for this area, we suggest drawing the path of a simple request: user, agent, tool, data and result. At each boundary, record the expected permission and the evidence available. Your personal sketch should make your reasoning reviewable; it should not be presented as an official AWS architecture.

Imagine an assistant that reads fictional orders. State two constraints: it may read an order's status but cannot modify the order, and it must explain unavailability without inventing a result. Identify the observations you would need to distinguish denied access, an unavailable source and a misinterpreted response.

That preparation separates obtaining a convincing answer from demonstrating controlled behaviour. Arrive with a precise question, such as: “What evidence is missing from my diagnosis when a tool call fails?” You can then look for an answer in the concepts actually taught, rather than assuming that a demonstration proves every operational requirement.

Advanced training: justify multi-agent complexity

The Building Advanced Agentic Systems on AWS outline (opens in a new tab) addresses coordination, context, security and observability. It recommends Agentic AI Foundations and Building Agentic AI with Amazon Bedrock AgentCore.

Before choosing this option, check whether your problem needs multiple agents. Our suggested preparation is to compare two solutions on paper: a straightforward process with a single responsible component, followed by a division of work across several components. Describe both the benefit and the additional diagnostic difficulty created by the second approach.

Consider a fictional technical report assembled from documents. One component extracts facts, another compares them and a third proposes a synthesis. What happens when their conclusions disagree? Where do you find the source of a claim? Who determines that a step must be repeated? These are preparation questions, rather than descriptions of official Jam exercises.

Your ability to discuss those trade-offs is a useful signal for choosing the advanced level. If you are still learning permissions or how an individual call travels through a system, more foundational practice may offer greater immediate value.

Jam, microcredentials and certifications serve different goals

The official AWS immersive learning overview (opens in a new tab) separates Jam challenges from skills assessments. For Jam, it describes open-ended console problems, points and clues. It presents microcredentials separately as practical validations that complement certifications.

Our French article on AWS microcredentials for agent identities and data covers a different announcement. A team result in a workshop should not be described as earning those credentials.

State your objective in one sentence: learn with support, validate a targeted skill, or prepare for a broader certification. Our AWS AI certification path overview addresses the third goal. A workshop can support practical learning, but registering does not guarantee exam coverage or a passing result.

Make the session useful for the whole team

Here is an OnLevelUp approach to preparing for a group workshop. Set one learning outcome per participant, then plan to rotate responsibilities: one person operates, another states the hypothesis and a third checks the evidence. This is a teaching suggestion, not an AWS Jam rule.

When progress stops, record the symptom before trying a correction. Write down what you expect the change to do and what would make you reject the hypothesis. After receiving help, restate the reasoning and identify the missing knowledge. That produces a usable revision list instead of leaving the leaderboard as your only memory of the session.

Afterwards, revisit a different problem from your own work. Explain the approach without retaining or disclosing confidential challenge details. A short review with a colleague can check whether you can transfer the reasoning to a new situation. Keep the review focused on decisions and evidence, rather than on who entered the most commands during the workshop.

Confirm the exact offer before booking

The announcement directs readers to AWS Classroom Training (opens in a new tab) to arrange these workshops. The catalog also describes an optional one-day Jam for certain private training deliveries. Do not treat that arrangement as identical to every Jam event or to challenges available through a Skill Builder Team subscription.

Ask for the exact workshop name, language, dates, delivery format, prerequisites and applicable price. Check separately whether your quote includes the corresponding course and the Jam. The consulted sources do not establish French session dates or a single universal price for these three additions.

Choose the workshop that addresses your next concrete difficulty. Judge its value afterwards by what you can explain, verify and reproduce on your own case. A one-day training experience can be useful without being presented as a promise of certification or complete production readiness.