AWS ML Enablement Workshop: Working Backwards for Cross-Functional AI Teams

What ML Enablement Workshop Is and Who It Targets

ML Enablement Workshop addresses a specific gap in how organizations adopt AI/ML: engineering and business teams typically work in isolation, and the result is either technically sound work with weak product fit or well-framed product ideas that engineering cannot build quickly. The workshop brings both sides together and runs them through Amazon's Working Backwards process, which starts from a hypothetical press release describing the finished customer experience and works backward to define what must be built.

The target participants are product managers, business-side stakeholders, and engineers working on the same product who have not yet agreed on what to build or how to validate it. The README describes the goal as forming a cross-functional team that can generate a working prototype and a validated hypothesis within six months. This is not a general machine learning training course; it is a planning and alignment process that uses generative AI as a tool throughout.

The Three-Session Structure: Day0, Day1, and Day2

The workshop runs in three sessions totaling about 7.5 hours.

Day0 is a one-hour session, which the README notes can be run remotely. It covers the purpose and roles of the workshop and checks that the organizational prerequisites are in place before the main program begins.

Day1, called the practical session, runs for 3.5 hours. Participants practice the Working Backwards process with generative AI to produce multiple press release drafts for proposed products or features. They also build a mock that can be shown to real users for feedback. The README points to a yourwork/ directory in the repository as a reference for how that day's output is structured.

Day2, the improvement session, runs for 3 hours. Teams incorporate feedback gathered from the mocks built on Day1, update their press releases, and lay out the first steps toward a production release within the following three to six months. The output of Day2 is a concrete plan, not just a refined idea.

Working Backwards with Generative AI and Kiro

A distinctive aspect of ML Enablement Workshop v3.x is its integration of Kiro, which the README describes as a spec-driven AI development tool, into the Working Backwards process. The intent is that the press releases and hypotheses generated during the workshop are written in a form that feeds directly into Kiro's specification-driven implementation flow, reducing the gap between product planning and engineering work.

The README provides a full example of Working Backwards outputs generated entirely by generative AI, using the theme of improving English conversation skills for business meetings. The example demonstrates how generative AI can handle the mechanical work of filling in press release templates, so participants can focus on evaluating the quality of the hypothesis and identifying questions the mock should answer. The README notes this example was produced without any human editing, as a demonstration of what the tooling can produce.

Python Dependencies and Repository Layout

The repository is primarily a Jupyter Notebook project (the primary language is listed as Python) with workshop guidance documents in the docs/ directory, template materials in the yourwork/ directory, and supporting code in the notebooks/, model/, and data/ directories. A community version of the workshop is also included for broader organizational AI/ML literacy programs.

For contributors or facilitators who need to run the code components, the project declares its Python environment in a pyproject.toml:

[project]
name = "aws-ml-enablement-workshop"
version = "0.1.0"
requires-python = ">=3.12"
dependencies = [
    "markitdown-mcp>=0.0.1a4",
    "mcp[cli]>=1.8.1",
    "pandas>=3.0.0",
    "openpyxl>=3.1.0",
    "python-docx>=1.2.0",
    "click>=8.3.1",
    "tabulate>=0.9.0"
]

Python 3.12 or later is required. A uv.lock file is present alongside the pyproject.toml, indicating that uv is the intended package manager. An environment.yml is also present at the top level for conda-based environments. The README does not document how to run the notebooks or what outputs they produce.

Customer Outcomes: What Organizations Have Reported

The README includes verbatim feedback from several Japanese organizations that have run the workshop. The most quantified outcome comes from MUFG (Mitsubishi UFJ Bank), which presented at re:Invent 2024. The README describes how a joint workshop between the bank's sales and engineering teams led to a 10x increase in lead generation and a 30% improvement in conversion within three months of running the prototype program. MUFG described the process as helping define the target sales experience and milestone, then rapidly building and validating it.

Other organizations reported outcomes that are harder to quantify but suggest where the workshop adds value: BASE ran it across five product teams simultaneously and reported concentrated discussions between engineers and product managers on generative AI use cases; PKSHA Technology noted that the two-part structure gave participants time to revise and improve their own plans rather than just receiving a framework; MoneyForward credited the workshop with creating communication between product managers, developers, and researchers that had not previously existed.

These reports are customer statements included in the README, not independently verified outcomes. The MUFG numbers come from a re:Invent talk cited in the README.

Limitations: Language, Facilitation, and Scope

The most significant practical constraint for non-Japanese teams is language. The README, the docs/ directory, and the workshop guide materials are written in Japanese. A team running the workshop without a Japanese-reading facilitator will need to translate the facilitation guides, which are the core of what the repository provides. The code and pyproject.toml are in English, but the workshop process documentation is not.

The workshop is not a self-serve tool. It requires a facilitator to run the sessions, participants who can commit uninterrupted blocks of time, and organizational conditions where a Day0 preparation has established shared goals. Teams looking for a training course they can take individually or at their own pace will find this repository does not serve that use case.

The repository also does not cover the implementation work after Day2. It helps teams define what to build and validate the hypothesis with a mock, but the path from a validated press release to a shipped product is outside the scope of the material. The Kiro integration described in the README is one bridge toward implementation, but the README does not document that integration in detail.

Comparison with Design Sprint and AWS Training Programs

The closest widely-known alternative is the Google Ventures Design Sprint, a five-day process for solving product problems and testing solutions with users. The principal difference is emphasis: Design Sprint focuses on identifying and validating a single problem-solution pair quickly, while ML Enablement Workshop is built around Amazon's Working Backwards method and specifically incorporates generative AI as an active participant in the ideation and prototyping steps. A team already familiar with Working Backwards would find ML Enablement Workshop more directly applicable.

AWS also provides formal ML training through AWS Skill Builder and certification paths, but those are individual learning tracks for technical skills rather than facilitated cross-functional workshops. They cover service APIs, model training, and infrastructure, but do not address the business and product planning work that ML Enablement Workshop targets.

Maintenance, Releases, and License

The repository is under the MIT-0 license, which is Amazon's zero-attribution variant of MIT. It permits use, modification, and distribution without requiring attribution. The most recent releases are v3.1.0 from October 2025 and v3.0.0 from September 2025. The last push to the repository was on 2026-09-27, indicating the project is under active maintenance.

A book co-authored by the workshop team was published in 2024: 「事例でわかるMLOps 機械学習の成果をスケールさせる処方箋」, which covers MLOps practices and includes the workshop approach in a chapter on overcoming the 80% failure rate of machine learning projects.

Editorial conclusion

ML Enablement Workshop is the right fit for organizations that want a structured, facilitated process for combining business and engineering perspectives on AI/ML product development, and are willing to commit a cross-functional team to a 7.5-hour program. Teams that need a self-serve learning tool rather than a facilitation guide, or whose primary developers are not comfortable with Japanese-language materials, will find the experience rough without a Japanese-speaking facilitator. Before scheduling, confirm that a Day0 preparation session can be completed to check organizational readiness, and review the Kiro spec-driven workflow if you intend to carry the outputs directly into implementation.

Frequently asked questions

Official sources

1. aws-samples/aws-ml-enablement-workshop on GitHub
2. Issues
3. License: MIT-0
4. README
5. Releases