OpenAI Deployment Company ↗
How are workflows selected, integrated and fed back to products?
W01 · Read with the module and assignment ↗READING GUIDE / PRIMARY RESOURCES
Further reading and technical resources
Find primary cases and technical documentation by course stage. Each resource includes a reading question and module so reading leads to a design decision or assignment.
01 / Framing and discovery
Understand how delivery teams find valuable problems and turn observations into scope and ownership.
Introducing OpenAI Frontier ↗
How do logs, documents and code form an investigation workflow?
W01 · Read with the module and assignment ↗Claude for the financial industry: A practical deployment guide (AIG case) ↗
Which expert tasks involve extraction, checking or judgment?
W02 · Read with the module and assignment ↗AIG 2025 Annual Report · The Future of Insurance ↗
How can underwriting and claims share concepts and capabilities?
W02 · Read with the module and assignment ↗Building effective agents ↗
When is a fixed workflow enough, and when is model-directed tool use useful?
W01 · Read with the module and assignment ↗02 / Data, ontology and tools
Read alongside your data contract and object model: how to connect, represent and authorize.
Airbus launches Skywise ↗
How do connected business objects support multiple applications?
W05 · Read with the module and assignment ↗How enterprises are driving AI transformation with Claude ↗
How does document generation combine retrieval, templates, versions and expertise?
W04 · Read with the module and assignment ↗Ontology overview ↗
How do objects, links and actions represent work?
W05 · Read with the module and assignment ↗Function calling ↗
Who owns tool arguments, execution and result handling?
W05 · Read with the module and assignment ↗MCP architecture overview ↗
What do hosts, clients and servers each do?
W05 · Read with the module and assignment ↗中华人民共和国个人信息保护法 ↗
What should a qualified reviewer check for minimization, access and de-identification?
W06 · Read with the module and assignment ↗03 / Evaluation, execution and operations
Translate methods into evaluation fixtures, tool constraints, trace fields and recovery checks.
Morgan Stanley uses AI evals ↗
How does expert feedback become evals, retrieval improvements and regression?
W07 · Read with the module and assignment ↗Klarna AI assistant: first-month results ↗
How should conversation handling, repeat contact and human support be assessed together?
W08 · Read with the module and assignment ↗TGH selects Palantir for connected care coordination ↗
How does resource state enter accountable operational coordination?
W09 · Read with the module and assignment ↗Demystifying evals for AI agents ↗
How do tasks, traces, outcomes and graders form an evaluation?
W07 · Read with the module and assignment ↗MCP security best practices ↗
How do connectors handle authorization, tokens and trust boundaries?
W06 · Read with the module and assignment ↗LLM01:2025 Prompt Injection ↗
How can instructions in documents or tool results cross trust boundaries?
W09 · Read with the module and assignment ↗Generative AI semantic conventions ↗
How can model calls and tool executions be linked to a business task?
W09 · Read with the module and assignment ↗AI Risk Management Framework ↗
How does risk identification become measurable operational requirements?
W09 · Read with the module and assignment ↗04 / Value and service productization
Separate functioning systems, adoption and increments; prepare reproducible evidence for acceptance and pricing.
从卖工具到卖效果：RaaS 如何重塑 ToB 市场格局？ ↗
What metrics, attribution and shared responsibility does outcome pricing require?
W10 · Read with the module and assignment ↗Difference-in-Differences ↗
What identification conditions are needed beyond calculating differences?
W10 · Read with the module and assignment ↗What to do about parallel trends when you only have baseline data? ↗
How should limited baseline data affect attribution conclusions?
W10 · Read with the module and assignment ↗Attribution, exercise data and version checks
Public context uses primary company or institutional sources. Names identify case subjects, not AIelsen partnerships or accreditation. FDE exercises are teaching designs, not undocumented project histories.
Exercise values and identifiers are synthetic. Do not add customer identities, personal information, private strategy or keys to worksheets, notes or exports. Records are local; review them before sharing. Hosts may keep ordinary access logs.
Resources were compiled September 7–8, 2026. Dates distinguish publication from access. Verify interfaces and specifications for your implementation version; qualified specialists should review legal, clinical and financial judgments.