Enterprise workflows

Published engineering accounts

Forward deployed engineering use cases

What does an FDE deliver with a customer? Start with the problem, the engineering work and the evidence available for accepting the result.

Each entry distinguishes source-named FDE involvement from related engineering work. Reported outcomes have not been independently verified here. These are not Nuvepro customer cases or evidence of a partnership.

22 accounts are included. Cross-linked existing workflow cases represent no additional deployment. Practice questions are proposals for discussion, not implemented workflows or executed environments.

Sources reviewed 2026-11-18.

Source explicitly names FDE involvement

Practitioner corrections become bounded tax-agent engineering work

Crete accounting firms · OpenAI and Thrive Holdings

Provider and partner-reported operating pilot

The customer problem

Complex returns require extraction from messy client files and mapping across files.

The engineering work described

OpenAI FDEs, researchers and Thrive engineers built Tax AI with Crete accountants. The published improvement design investigates recurring failures, scopes bounded fixes and reruns evaluations before engineering review. Its fair-rental-days task is representative, not an inspected change history.

Data and systems

Client files and notes, tax-engine mappings, traces and evaluations; the representative design includes a bounded writable Codex worktree.

Human checks and acceptance

Practitioners approve filings; engineers own shipping. Ambiguity returns to the product team.

Reported result

The source reports 7,000 pilot returns, about 50% greater throughput, and rental-property work reaching 90% precision and recall in six weeks.

Evidence limits

No customer artifacts, test sets or metric denominators were inspected. The illustrative scaffold does not establish that every path is production code. Agent-assisted improvement does not establish model-weight training or autonomous shipping.

A proposed practice question

How could a synthetic exercise connect an accountant's correction to trace evidence, a bounded engineering change, independent regression checks and separate practitioner and engineering acceptance?

No environment has been built or exercised for this question. Domain agreement, design and independent acceptance checks remain to be established.

Source explicitly names FDE involvement

Embedded chip-verification engineering and shared Codex delivery

Unnamed semiconductor company · OpenAI

Provider-reported production and early adoption

The customer problem

Chip verification required customer-specific evaluation and repeated integration.

The engineering work described

FDEs embedded with verification engineers on directed-test generation, debugging and architecture assistance, using jointly developed evaluations and custom harnesses. The account describes later consolidation onto a shared Codex harness, skills and plugins, with team-specific connectors and fresh evaluations.

Data and systems

Custom verification harnesses, shared Codex skills and plugins, team-specific connectors and MCP.

Human checks and acceptance

Customer engineers co-developed evaluations for useful tests and correct failure diagnosis. Detailed criteria and formal approval evidence are undisclosed.

Reported result

OpenAI reports production adoption within three to six months and about 48% less directed-test time. These results were not independently reproduced.

Evidence limits

The customer identity, source artifacts, datasets and metric method are unavailable. Results are provider claims; no implementation or evaluation was independently run. This is the existing workflow-library case, viewed through its FDE work, not an additional deployment.

Existing case: Read this case in the workflow library. This cross-link adds no independent deployment.

A proposed practice question

What evidence would a domain expert and delivery engineer require before replacing a custom verification harness with reusable skills, connectors and task-specific evaluations?

No environment has been built or exercised for this question. Domain agreement, design and independent acceptance checks remain to be established.

Source explicitly names FDE involvement

FDE-supported Confluence content operations

Expedia · Atlassian

Provider-reported operating delivery with limited implementation detail

The customer problem

Content operations at enterprise scale; the precise starting workflow is undisclosed.

The engineering work described

Atlassian attributes AI-supported Confluence content operations at Expedia to its FDE team. Case-specific implementation steps are unavailable. The article's general FDE discovery workshops do not establish Expedia followed every described program step.

Data and systems

Confluence content. Exact connectors, records and agent architecture are undisclosed.

Human checks and acceptance

Formal test criteria, authorization and archival review controls are undisclosed.

Reported result

Atlassian reports 600,000 archived pages and 3,200 hours saved annually, accompanied by an Expedia engineering-manager endorsement.

Evidence limits

These are attributed vendor claims, not independent measurements. No customer system or output was inspected. Available detail cannot establish a technical or approval workflow; general program activity and anonymous examples are not Expedia-specific evidence.

A proposed practice question

Before designing a synthetic content-lifecycle exercise, what discovery evidence would establish retention rules, archival permissions, accountable reviewers and independent acceptance checks?

No environment has been built or exercised for this question. Domain agreement, design and independent acceptance checks remain to be established.

Source explicitly names FDE involvement

FDE-built regulatory review with cited verdicts

Unnamed compliance project · Kloia

Provider-reported project; production acceptance and outcomes undisclosed

The customer problem

Move a compliance AI prototype toward usable delivery under regulatory constraints.

The engineering work described

Kloia describes a recent FDE project with seven agents, coverage of 330 regulatory clauses and a citation attached to each verdict. It does not disclose the agent roles, integration sequence or validation method.

Data and systems

Regulatory clauses and cited verdicts; exact source documents, software and customer systems undisclosed.

Human checks and acceptance

Reviewer responsibilities, authorization boundaries and final acceptance criteria are undisclosed.

Reported result

Reducing a five-day review to under one day is explicitly a target, not an achieved result.

Evidence limits

This is a promotional provider account of an unnamed project, not an inspected implementation or independent compliance assessment. No deployment or correctness evidence was reviewed.

A proposed practice question

What independent reference, clause coverage checks and qualified reviewer approval would a synthetic compliance exercise need before accepting a cited verdict?

No environment has been built or exercised for this question. Domain agreement, design and independent acceptance checks remain to be established.

Source explicitly names FDE involvement

Customer-specific voice delivery and model evaluation

Unnamed telecommunications customer · OpenAI

Provider-reported production deployment

The customer problem

Extend a customer-service platform into voice while meeting requirements for accents, tone, latency, function calling and instruction following.

The engineering work described

FDEs worked with the customer to build the platform and define requirements. Research addressed capability gaps; successive model snapshots were tested against customer requirements over six months.

Data and systems

An existing customer-service platform and voice-model snapshots. Integrations, source records and evaluation datasets are undisclosed.

Human checks and acceptance

FDEs and the customer scoped requirements; Research improved the model. Formal sign-off and operating escalation controls are undisclosed.

Reported result

OpenAI reports key evaluations improving from around 60% to 80% and then above 90%, with the resulting model supporting the customer's deployment.

Evidence limits

Provider-reported evaluation scores are not a general accuracy rate or independently reproduced business impact. Customer identity, test definitions and artifacts are unavailable. Other customers' call-handling results are separate.

A proposed practice question

In a proposed synthetic exercise, which customer scenarios, tool-call checks and escalation tests would justify acceptance of a revised voice model?

No environment has been built or exercised for this question. Domain agreement, design and independent acceptance checks remain to be established.

Source explicitly names FDE involvement

Embedded engineering for FOX fan-facing AI

FOX Corporation · Databricks

Provider and customer-reported operating delivery

The customer problem

Build fan-facing AI experiences across FOX Sports and FOX One.

The engineering work described

FOX's CTO describes Databricks FDEs building alongside FOX engineers. Discovery steps and implementation sequence are undisclosed.

Data and systems

Lakebase, AI Search, Databricks Apps and Model Serving; underlying datasets undisclosed.

Human checks and acceptance

Customer CTO endorsement; evaluation criteria and formal acceptance procedures undisclosed.

Reported result

The CTO reports approximately twice the app time among Sports AI users. This is an engagement comparison, not proof of causal impact.

Evidence limits

Provider article with attributed customer testimony. No artifacts or measurements independently reviewed. Summary search metrics are not substituted for the CTO's engagement claim.

A proposed practice question

What baseline, relevance checks, traffic tests and owner acceptance would a proposed synthetic fan-search exercise require?

No environment has been built or exercised for this question. Domain agreement, design and independent acceptance checks remain to be established.

Source explicitly names FDE involvement

FDE-led migration of JPMC risk data

JPMC Consumer and Community Banking · Databricks

Provider and customer-reported completed migration

The customer problem

Modernize the CCB Risk data foundation supporting subsequent AI work.

The engineering work described

JPMC's Consumer and Community Banking CTO attributes data and notebook migration plus user training to Databricks FDEs. Detailed pipeline changes are undisclosed.

Data and systems

CCB Risk data and notebooks; source platforms and sensitive-data controls undisclosed.

Human checks and acceptance

Customer CTO endorsement; reconciliation, authorization and formal acceptance procedures undisclosed.

Reported result

The CCB CTO reports 5+ petabytes and 500+ notebooks migrated in four months, with 600+ users trained.

Evidence limits

Migration results are attributed testimony, not independently verified. They do not establish a deployed AI application or measured downstream business impact.

A proposed practice question

How would a proposed synthetic migration prove reconciliation, notebook behavior, access controls, rollback and separate owner acceptance?

No environment has been built or exercised for this question. Domain agreement, design and independent acceptance checks remain to be established.

Related engineering work; FDE involvement not established

Context engineering for Qualcomm support workflows

Qualcomm · Context

Provider-reported production rollout

The customer problem

Help customer engineers answer support questions that depend on documentation, proprietary code and earlier troubleshooting.

The engineering work described

Context worked with a Qualcomm pilot team to select real support tasks for evaluation. It iterated on retrieval and shared context before expanding rollout.

Data and systems

Qualcomm virtual private cloud; indexed support documents, proprietary code surfaced through a metadata directory, and more than 100 connected Git repositories.

Human checks and acceptance

Context built the held-out suite with the pilot team and scored tasks against task-specific rubrics. The suite gated retrieval, context and routing changes. Formal owner sign-off is undisclosed.

Reported result

Context reports a rise from 23% to 98% on its held-out task suite with the same model, and rollout to 85 teams.

Evidence limits

Provider-reported suite results are not general accuracy or independently verified. Databricks involvement in this implementation is unestablished.

A proposed practice question

Can a proposed support sandbox improve held-out troubleshooting results while preventing unauthorized repository reads?

No environment has been built or exercised for this question. Domain agreement, design and independent acceptance checks remain to be established.

Source explicitly names FDE involvement

Deliver an Oracle-dialect text-to-SQL model

Unnamed large enterprise · Human FDE described by the research authors

Research authors report a completed production delivery

The customer problem

Generate Oracle SQL from natural-language questions within the internal stakeholder’s data-access, query compilation and CPU latency constraints.

The engineering work described

A human FDE worked through stakeholder requirements, tested more than 100 adapters across 200 evaluations, and delivered a Ministral 3 8B model with a tuned adapter and prompt over four months.

Data and systems

Oracle schemas and queries, public training benchmarks, a restricted production workload and a CPU serving target.

Human checks and acceptance

The research team validated the requirements against engagement records. The evaluation checked development execution accuracy, production query failures and p90 CPU latency. Formal stakeholder sign-off is undisclosed.

Reported result

The paper reports 79.9% Spider development execution accuracy, six failures on the SQL compilation check across 100 production questions and 9.26 seconds p90 for the shipped model under the study evaluator.

Evidence limits

The production workload lacked reference SQL, so compilation did not establish answer correctness. The paper’s separate agent experiment used a simulated customer. Its runs were research evaluations.

A proposed practice question

What evidence should a proposed exercise require for permitted training data, valid Oracle SQL, correct query results and acceptable CPU latency?

No environment has been built or exercised for this question. Domain agreement, design and independent acceptance checks remain to be established.

Source explicitly names FDE involvement

Verify appliance model numbers before looking up parts

Encompass Supply Chain Solutions · HappyRobot

Provider reports operating voice support and expansion to more brands

The customer problem

Parts callers need compatible stock and prices. One misheard character in a model number can return the wrong part.

The engineering work described

Call tracking identified repetitive work. Encompass started with Midea. A HappyRobot FDE worked with the team to refine transcription, validate model numbers and stress-test confusing formats before expanding.

Data and systems

Cisco IVR, product model numbers, parts catalog, and APIs for live inventory and pricing.

Human checks and acceptance

Support operations counted transfers to live agents as failures. A failed lookup escalated to a person. Formal release sign-off and the test set are undisclosed.

Reported result

HappyRobot reports roughly 1,700 LG-related calls a week and about 64% resolved without human intervention.

Evidence limits

The reported resolution rate measures handling without transfer. Independent part-match accuracy, the reporting window and formal acceptance records are unavailable.

A proposed practice question

How could a proposed voice exercise detect a single-character mismatch, confirm the right part and escalate an unresolved lookup?

No environment has been built or exercised for this question. Domain agreement, design and independent acceptance checks remain to be established.

Source explicitly names FDE involvement

Personalize subscriber meals with an identity-aware agent platform

CookUnity (identified in the provider FAQ; the case headline anonymizes the client) · phData

Provider reports a production-ready platform delivered in seven weeks; the initial rollout targeted 100,000 subscribers.

The customer problem

CookUnity’s bot and meal-ranking model lived in separate codebases. Subscribers needed recommendations that reflect changing preferences and goals while respecting diet and allergen rules.

The engineering work described

phData embedded one FDE with three CookUnity engineers for seven weeks. It connected the existing bot and ranking model through AgentCore Gateway, unified them in AgentCore Runtime, applied subscriber-level identity controls and layered Bedrock Guardrails with application checks. It provisioned infrastructure with Terraform and monitored traces through AgentCore Observability.

Data and systems

AWS profiles, order history and menus; bot, ranking model, AgentCore Runtime/Gateway/Identity, Lambda, ECS, S3, VPC, IAM and Terraform.

Human checks and acceptance

CookUnity engineers retained day-to-day ownership. phData cites Bedrock Model Evaluations and traces but publishes no eval set, failure rates, human escalation path or formal owner sign-off.

Reported result

phData reports production-ready delivery in seven weeks, with an initial 100,000-subscriber rollout target. It reports 99%+ diet/allergen compliance and 60%+ fewer meal-selection support contacts; some acceptance and retention figures are estimates or projections.

Evidence limits

The story headline anonymizes the client; its FAQ names CookUnity. Results are provider-reported; denominators, evaluation data and independent audit are unavailable. Formal owner acceptance is undisclosed.

A proposed practice question

How would you test changed meal preferences against dietary and allergen constraints, identity-scoped records, tool failures and escalation, and what evidence should a nutrition owner require before accepting recommendations for release?

No environment has been built or exercised for this question. Domain agreement, design and independent acceptance checks remain to be established.

Source explicitly names FDE involvement

Standardize finance reporting across a retailer’s operational systems

Unnamed European multinational retailer · G7 Consulting Group

Provider-reported 20-week engagement with later expansion into Treasury and Supply Chain Finance.

The customer problem

Finance teams reconciled inconsistent KPI definitions across 20+ operational systems and prepared weekly reports manually, delaying forecasts and executive decisions.

The engineering work described

G7 says an embedded Forward Deployed Finance Engineer worked with finance and technical stakeholders to create governed models and a semantic layer, then iteratively built AI-assisted summaries and reporting with users.

Data and systems

ERP, inventory, CRM and payment data; governed metrics for revenue, margin, inventory, cash flow, working capital, opex and forecast accuracy.

Human checks and acceptance

Finance professionals review and approve generated executive narratives. The source says solutions were designed, tested and refined with finance users, but gives no formal acceptance criteria or evaluation method.

Reported result

G7 reports 60% less manual reporting and 50% faster executive reporting within 20 weeks, plus daily dashboards and consistent KPIs. Provider-reported, not independently verified.

Evidence limits

The customer is unnamed and the source is provider-authored. Data models, test sets, metric denominators and independent evidence are unavailable. Finance review is described, but formal business-owner sign-off is not.

A proposed practice question

How could a synthetic finance-data exercise reconcile inconsistent KPIs, trace AI narratives to governed data and require finance approval against measurable checks? Not built or executed.

No environment has been built or exercised for this question. Domain agreement, design and independent acceptance checks remain to be established.

Source explicitly names FDE involvement

Move an insurance claims AI pilot into an owned production workflow

Unnamed insurance claims operations team · Cohash

Provider-reported pilot-to-production delivery and internal handover.

The customer problem

A claims AI pilot used a personal key and copied production data in a sandbox, had no evaluation set and depended on a platform scheduled for retirement.

The engineering work described

Cohash says its FDEs worked with claims staff to choose one workflow and metric, built expected/hard/edge evaluation cases before prompt changes, connected live tools with scoped identity, added human review for costly decisions, tracing and spend limits, then prepared CI and runbooks.

Data and systems

Claims cases, live APIs and MCP tools under workflow-scoped identity, tracing, spend limits, CI evaluations and runbooks.

Human checks and acceptance

The provider says the business owner and internal engineering team now operate the workflow and its evaluations run in CI. Owner identity, acceptance criteria and operating evidence are undisclosed.

Reported result

Cohash reports the workflow moved from sandbox pilot to internal production ownership; no quantitative business outcomes are published. Provider-reported, not independently verified.

Evidence limits

The client, publication date and implementation artifacts are not disclosed. Evaluation thresholds and production evidence are unavailable; no metrics or independent verification are provided.

A proposed practice question

How would a synthetic claims workflow prove representative evaluation, identity-bounded tool access, human-review thresholds and a usable engineering handover? Not built or executed.

No environment has been built or exercised for this question. Domain agreement, design and independent acceptance checks remain to be established.

Source explicitly names FDE involvement

Recover hotel bookings through property-specific voice AI

Unnamed 50+ property hospitality chain (two properties disclosed) · Dextr AI

Dextr reports a live chain deployment; the published results cover two properties two months after launch.

The customer problem

Two properties were 32–41% below prior-year voice bookings before peak season. After-hours calls rang out, while booking provenance, payment status and property rules sat across separate systems and teams.

The engineering work described

Dextr says a dedicated FDE, AI strategist and operators held twice-weekly sessions with the PMS provider, contact center and property operations. They versioned a shared reservation ontology and connected the agent to live system-of-record data, booking, payment, routing and escalation workflows.

Data and systems

PMS rates, availability, inventory and payments; contact-center conversations; property-level policies; OTA-versus-direct booking provenance; event and group blocks.

Human checks and acceptance

Property operations codified guest lists, block rules and escalation thresholds; humans retained complex bookings. The account describes human routing but no named owner sign-off criteria or formal acceptance records.

Reported result

Dextr reports 3.5–3.9 times more calls answered, 53–62% of booking revenue from AI and 46–70% more revenue than each property's pre-launch trajectory over two months. Vendor-reported, not independently verified.

Evidence limits

The chain and properties are anonymized. Dextr provides no underlying call, revenue or counterfactual data, evaluation set, customer artifacts or independent verification. The comparison uses a pre-launch trend; attribution and formal owner acceptance remain unverified.

A proposed practice question

How would a synthetic hotel reservation exercise reconcile PMS and contact-center state, preserve booking provenance, encode property policies and compare a bounded rollout against a pre-agreed baseline? Not built or executed.

No environment has been built or exercised for this question. Domain agreement, design and independent acceptance checks remain to be established.

Related engineering work; FDE involvement not established

Rebuild logistics billing while live downstream workflows continue

Unnamed U.S. delivery orchestration platform · Limestone Digital

Provider reports production release by months five to six and continued development after one year.

The customer problem

The legacy invoice platform could not safely support shipper and provider billing. Live delivery workflows and downstream consumers had to keep running through a financial-logic rebuild.

The engineering work described

Two AI-augmented engineers first mapped the legacy repository into a knowledge graph. They maintained specs as requirements changed, corrected AI-generated plans, then built invoicing, cost calculation, disputes, approvals and event-driven synchronization. A human reviewer gated every merge.

Data and systems

Legacy Knack invoicing, delivery data and downstream consumers; replacement .NET/C# platform on Kubernetes with event-driven synchronization.

Human checks and acceptance

Reviewers had to verify each change against its spec and explain and debug it without AI. Limestone reports a client tech-lead endorsement; formal business-owner acceptance criteria are undisclosed.

Reported result

Limestone reports 122 PRs in 90 days, 330+ by production around month six, and two client bonuses. Provider-reported delivery measures; billing correctness and business impact were not independently verified.

Evidence limits

Client and artifacts are undisclosed. No invoice samples, reconciliation results, tests, defect rates or cutover records are published. The specific team is not shown to hold FDE titles; FDE-like reflects customer-specific discovery and production engineering.

A proposed practice question

How could a synthetic billing migration map legacy behavior, preserve live downstream contracts, reconcile invoices independently and require named customer and engineering acceptance at cutover? Not built or executed.

No environment has been built or exercised for this question. Domain agreement, design and independent acceptance checks remain to be established.

Source explicitly names FDE involvement

Resolve delayed-order requests through a WhatsApp service agent

Riachuelo / Midway · Genesys

Genesys reports production use with ongoing plans to expand virtual agents across voice and chat.

The customer problem

Late deliveries were the main reason customers contacted Riachuelo's e-commerce support. Employees had to fetch tracking details or create a level-two support ticket for routine order-status requests.

The engineering work described

Riachuelo says it built an Agentic Virtual Agent for its WhatsApp journey using scripts from strong consultants. Genesys says forward-deployed engineers worked with Riachuelo to move agentic use cases into production; the source does not attribute each technical task to a named engineer.

Data and systems

WhatsApp service journey, order tracking links, level-two ticket creation, self-service options and Genesys Cloud customer-engagement platform.

Human checks and acceptance

Consultants remain available for exceptions and requests needing judgment. The customer story describes routing critical situations and connecting customers to people, but publishes no formal acceptance criteria or escalation thresholds.

Reported result

Genesys reports the agent retains 84% of these interactions in self-service, compared with 30% for the traditional chatbot. Metrics are provider-reported; measurement definitions and independent verification are unavailable.

Evidence limits

The provider-published account quotes a Riachuelo CX leader but lacks the evaluation set, denominator, period, error rate, integration artifacts or formal owner acceptance. FDE attribution comes from a separate Genesys announcement.

A proposed practice question

How could a synthetic order-support exercise verify order-to-tracking links, ticket creation, customer-safe answers and human routing against independently checked cases? Not built or executed.

No environment has been built or exercised for this question. Domain agreement, design and independent acceptance checks remain to be established.

Related engineering work; FDE involvement not established

Normalize carrier rates and prepare freight quotations

Unnamed international freight forwarder · Tyler Kim

Practitioner-reported operating implementation, January 2026

The customer problem

Operators maintained incompatible carrier rate sheets and assembled quotations from incomplete shipping requests.

The engineering work described

Kim describes customer-specific discovery, operator-authored SOPs, rate extraction with deterministic checks, and an Outlook quoting assistant revised after live failures. Operators can supply context and correct individual steps.

Data and systems

Carrier PDFs, spreadsheets and images; rate tables, calculation tools and Outlook.

Human checks and acceptance

Operators accept draft rates. Early draft quotes require review; later sending authority and formal owner acceptance are undisclosed.

Reported result

The practitioner reports reduced quotation effort; no independent measurement was inspected.

Evidence limits

The customer, raw eval cases and denominators are unavailable. Formal FDE involvement is not established. Three articles describe one deployment; source dates are not deployment-verification dates.

Existing case: Read this case in the workflow library. This cross-link adds no independent deployment.

A proposed practice question

How would your team normalize test carrier sheets, draft a quotation with missing shipment facts, independently verify charges and obtain an authorized operator decision before release?

No environment has been built or exercised for this question. Domain agreement, design and independent acceptance checks remain to be established.

Source explicitly names FDE involvement

Reconcile deal facts and assemble franchise agreements

Unnamed global hospitality brand · Flank

Vendor-reported operating contract drafting; CRM reconciliation remains early

The customer problem

Off-system deal changes left franchise documents and Salesforce records inconsistent.

The engineering work described

Flank VP Product Paul Lacey names Jason as the forward-deployed engineer. The team reconciled deal evidence into structured facts, then used deterministic templates to assemble agreement packs, iterating with customer users.

Data and systems

Salesforce, emails, MOUs, corporate records and existing contract templates.

Human checks and acceptance

Users resolve conflicting or missing facts. Legal acceptance criteria and CRM write-approval controls are undisclosed.

Reported result

Flank reports same-day drafts and expanded customer scope; these are vendor claims.

Evidence limits

The customer is unnamed. No independent results, error dataset or formal owner sign-off was inspected. CRM reconciliation has no concrete performance data; drafting does not establish general legal correctness.

Existing case: Read this case in the workflow library. This cross-link adds no independent deployment.

A proposed practice question

How would your team reconcile conflicting test deal records, assemble a franchise pack from approved templates, independently verify every variable and route unresolved legal decisions to the responsible reviewer?

No environment has been built or exercised for this question. Domain agreement, design and independent acceptance checks remain to be established.

Related engineering work; FDE involvement not established

Bring proprietary flight-data decoders into Spark

Eclipse Aerospace · Databricks with Eclipse Aerospace

Vendor-reported customer-specific data engineering; no AI deployment established

The customer problem

Proprietary telemetry required a separate decode/storage step.

The engineering work described

Eclipse supplied decoders and domain knowledge; Databricks supplied platform/Spark integration in joint connector development. FDE-like is our interpretation, not a named role.

Data and systems

Proprietary files, existing Rust decoders, Python, Arrow, Spark and downstream tables.

Human checks and acceptance

Customer acceptance and malformed-record release/recovery authority are undisclosed.

Reported result

The vendor reports a simpler ingestion path; numerical gains are not independently verified.

Evidence limits

No code, benchmarks, records or sign-off inspected. The 12 November primary and 14 November roundup are not implementation-launch dates. This links the same workflow case, not another deployment.

Existing case: Read this case in the workflow library. This cross-link adds no independent deployment.

A proposed practice question

How would your engineering owner independently check synthetic decoded fields and malformed-record handling before accepting an exact diagnostic and simulated reprocessing plan?

No environment has been built or exercised for this question. Domain agreement, design and independent acceptance checks remain to be established.

Source explicitly names FDE involvement

Modernize a legacy insurance policy administration module

Unnamed insurer in the Americas · IBM Consulting

Author-reported production use for a selected module

The customer problem

An insurer needed to modernize a desktop policy system whose business rules were largely undocumented in SQL Server procedures. Earlier AI-generated code had changed premium calculations.

The engineering work described

IBM describes a four-person FDE pod working with business specialists. Parallel pipelines extracted procedure logic and rebuilt the interface and services, with human reviews.

Data and systems

VB.NET and SQL Server; a React and Spring Boot application on AWS, PostgreSQL, third-party administrator feeds, and connections to claims and billing. The author reports 62 procedures converted for the selected module; the wider application had about 300.

Human checks and acceptance

Business specialists helped interpret rules. The article describes human decision points and manual release approval; it does not identify the approving owner or publish acceptance criteria.

Reported result

The author reports 99.2% first-pass functional parity and about 40 days to production versus a 14–16 week traditional estimate. The estimate is not an observed control, and neither result is independently verified.

Evidence limits

The customer is unnamed. Test corpus, calculation methods, acceptance records and independent operating evidence are unavailable. Claims are attributed to the author.

A proposed practice question

How might a synthetic team extract legacy policy rules, compare outputs, investigate discrepancies and obtain a business owner’s approval before a controlled release?

No environment has been built or exercised for this question. Domain agreement, design and independent acceptance checks remain to be established.

Source explicitly names FDE involvement

Migrate enterprise software assets with guided matching

Unnamed enterprise customer of a global technology company · Deloitte

Provider reports a functioning tool; production rollout is undisclosed

The customer problem

Enterprise customers faced lengthy manual reconstruction of legacy digital assets before they could move to a new software platform.

The engineering work described

A Deloitte FDE and designer embedded with one customer, gathered requirements and daily feedback, and built an IT-guided tool that matches legacy assets to target items, explains matches and flags edge cases for review.

Data and systems

A database of legacy-platform assets, the new software environment and AI-generated matching with compatibility scores.

Human checks and acceptance

Customer stakeholders provided requirements and daily feedback; an IT user guides the migration and edge cases are flagged for human review. Formal owner acceptance and test criteria are undisclosed.

Reported result

Deloitte reports a functioning tool built in weeks and says it could reduce a twelve-month migration to weeks. This is a potential outcome, not an independently measured result.

Evidence limits

Both organizations are unnamed. The provider does not publish a test set, match accuracy, production rollout or formal sign-off. Reported outcomes are not independently verified.

A proposed practice question

How would a synthetic migration exercise compare legacy assets with target objects, check uncertain matches and obtain an IT owner’s acceptance of a reconciled plan?

No environment has been built or exercised for this question. Domain agreement, design and independent acceptance checks remain to be established.

Source explicitly names FDE involvement

Connect retail service agents to live orders and returns

Unnamed global retail brand · Entrans

Provider-reported phased production rollout

The customer problem

FAQ chatbots could not resolve live order and returns questions across disconnected commerce systems.

The engineering work described

An Entrans FDE pod studied conversations, APIs and data flows with ecommerce and service teams. It built agents for product discovery, order status and starting returns, with policy guardrails and contextual human handoff.

Data and systems

Product catalogue, order management and CRM connected through secure APIs and MCP tools in the customer’s cloud and repositories.

Human checks and acceptance

The provider describes handoff of complex cases to service agents and says usage and answer quality guided expansion. Evaluation criteria, authorization thresholds and formal owner sign-off are undisclosed.

Reported result

Entrans reports production go-live on live commerce systems and a reusable foundation for further workflows. Numerical operating results are not provided.

Evidence limits

The customer is unnamed and the account is vendor-authored. Production status and outcomes are not independently verified; evaluation artifacts and formal acceptance are unavailable.

A proposed practice question

How would a synthetic service exercise verify order status, apply return policy and hand off an exception with context to an authorized reviewer?

No environment has been built or exercised for this question. Domain agreement, design and independent acceptance checks remain to be established.