Corpus-Complete Vocabulary Normalization

Snapshot: corpus-complete-20260711 (2026-07-11) · source commit 1737631198b184e05ba0a343d67df6fa9e71b938 Scope: 1780 approved schema-v2 extractions; 1754 deduped sources; 991 deduped agentic projects.

Corpus-complete analysis output. Counts are generated from extraction.json; citations link to deterministic report.md views. This is an input to, not the output of, Phase 4 synthesis.

This logged normalization run inventories the complete corpus vocabulary. It changes analysis counts only; extraction files and report.md files are not edited.

Only reviewer-confirmed aliases are merged into aggregate counts. Accepted vocabulary remains canonical. Proposed and out-of-vocabulary names stay visible for reviewer adjudication.

Decisions

Original name Analysis treatment Status Raw usages Deduped projects Rationale Recommendation / competing mappings
cost-aware-serving cost-aware-routing normalized 1 1 Reviewer-confirmed direct alias for cost-aware model serving and routing. binding decision
feedback-loop-iteration eval-driven-development normalized 2 2 Reviewer-confirmed mapping for the two observed iterative evaluation and improvement usages. binding decision
guardrails safety-policy-enforcement normalized 1 1 Direct alias found by design review; count generic guardrail usage under safety and policy enforcement. binding decision
human-in-the-loop human-in-the-loop-gate normalized 1 1 Direct alias found by design review; count it under the accepted gate vocabulary. binding decision
model-fallback fallback-degradation normalized 1 1 Reviewer-confirmed direct alias for model fallback and graceful degradation behavior. binding decision
async-message-orchestration keep as proposed pending review review_required 27 27 The complete usage list supports a possible distinct messaging boundary, but semantic overlap with durable-execution must be adjudicated. The numeric threshold is comfortably met; review whether usages consistently describe durable message-centered coordination rather than generic durable execution.
Candidates: durable-execution
code-execution-sandbox keep as proposed pending review review_required 1 1 Singleton with a potentially distinct execution-isolation boundary. Keep visible for boundary review rather than folding a concrete isolation architecture into generic policy enforcement.
Candidates: safety-policy-enforcement, tool-use-orchestration
fallback-strategy keep as proposed pending review review_required 1 1 Singleton whose generic name could span more than one accepted boundary. Likely alias of fallback-degradation; confirm that the usage describes graceful fallback rather than model routing.
Candidates: fallback-degradation, model-routing
fine-tuning-feedback-loop keep as proposed pending review review_required 1 1 Singleton below the acceptance threshold whose boundary overlaps two accepted patterns. Review whether the usage is fine-tuning-adaptation plus data-quality-loop rather than a distinct pattern.
Candidates: fine-tuning-adaptation, data-quality-loop
forward-deployed-engineering keep as proposed pending review review_required 3 3 The numeric acceptance threshold is now met, but semantic coherence still requires reviewer adjudication. The complete corpus reaches three projects; reviewer should decide whether all usages share a distinct deploy-with-the-customer engineering boundary.
incremental-rollout keep as proposed pending review review_required 1 1 Singleton below threshold with an apparent accepted equivalent. Likely alias of canary-rollout; confirm the rollout includes staged exposure or comparison.
Candidates: canary-rollout
latency-aware-model-routing keep as proposed pending review review_required 1 1 Singleton compound term that would fragment two accepted building blocks. Review as a composition of model-routing and latency-optimization rather than accepting a new compound name.
Candidates: model-routing, latency-optimization
model-output-rebasing keep as proposed pending review review_required 1 1 Singleton describing adjustment of generated outputs against a changing base; no confirmed canonical equivalent yet. Keep proposed pending evidence from additional projects.
Candidates: structured-output-contract, data-quality-loop
multi-stage-retrieval keep as proposed pending review review_required 1 1 Singleton whose canonical mapping depends on which retrieval stages are actually distinct. Review whether the usage is retrieval-ranking, rag-retrieval, or prompt-chaining before normalization.
Candidates: rag-retrieval, retrieval-ranking, prompt-chaining
source-citation-grounding keep as proposed pending review review_required 1 1 Singleton describing source-linked answer grounding; potentially useful but below the three-project acceptance threshold. Keep proposed for reviewer boundary review; consider rag-retrieval only if citation fidelity is not a distinct reusable control.
Candidates: rag-retrieval, safety-policy-enforcement
structured-output-validation keep as proposed pending review review_required 2 2 Two usages are below threshold and may describe validation rather than generation-time schema constraint. Likely direct alias of structured-output-contract; reviewer should confirm from both usages.
Candidates: structured-output-contract
workflow-decomposition keep as proposed pending review review_required 16 16 The current name implies step decomposition while known usages may instead describe limiting automation scope. The numeric threshold is met; reviewer should decide whether usages are true step decomposition or should be renamed/normalized toward thin-slice automation.
Candidates: tool-use-orchestration, prompt-chaining, thin-slice-automation
workflow-orchestration keep as proposed pending review review_required 1 1 The name is broader than the accepted vocabulary boundaries, so its actual topology determines the canonical mapping. Review the single usage and map to tool-use-orchestration only if orchestration is specifically tool/API execution.
Candidates: tool-use-orchestration, durable-execution, multi-agent-orchestration

Complete Vocabulary Inventory

Accepted names are inventoried by aggregate usage count. Proposed, aliased, and out-of-vocabulary names are expanded in the usage-review section below.

Name Vocabulary status Raw usages Deduped projects Agentic usages Analysis treatment
agent-memory accepted 192 190 185 canonical accepted name
async-message-orchestration proposed 27 27 22 review required; not merged
batch-offline-processing accepted 144 143 41 canonical accepted name
canary-rollout accepted 60 60 27 canonical accepted name
code-execution-sandbox out_of_vocabulary 1 1 1 review required; not merged
confidence-thresholding accepted 74 74 39 canonical accepted name
context-compaction accepted 119 117 100 canonical accepted name
context-engineering accepted 125 124 83 canonical accepted name
cost-aware-routing accepted 279 278 112 canonical accepted name
cost-aware-serving alias 1 1 0 normalized to cost-aware-routing
data-quality-loop accepted 234 233 98 canonical accepted name
durable-execution accepted 125 122 96 canonical accepted name
eval-driven-development accepted 860 848 482 canonical accepted name
fallback-degradation accepted 105 103 61 canonical accepted name
fallback-strategy out_of_vocabulary 1 1 1 review required; not merged
feedback-loop-iteration alias 2 2 2 normalized to eval-driven-development
fine-tuning-adaptation accepted 256 253 90 canonical accepted name
fine-tuning-feedback-loop out_of_vocabulary 1 1 0 review required; not merged
forward-deployed-engineering proposed 3 3 3 review required; not merged
guardrails alias 1 1 1 normalized to safety-policy-enforcement
guardrails-io-filtering accepted 179 175 101 canonical accepted name
human-in-the-loop alias 1 1 1 normalized to human-in-the-loop-gate
human-in-the-loop-gate accepted 634 626 401 canonical accepted name
hybrid-search accepted 75 75 39 canonical accepted name
incremental-rollout out_of_vocabulary 1 1 1 review required; not merged
latency-aware-model-routing out_of_vocabulary 1 1 0 review required; not merged
latency-optimization accepted 230 227 107 canonical accepted name
llm-as-judge accepted 218 213 124 canonical accepted name
model-distillation accepted 40 38 11 canonical accepted name
model-fallback alias 1 1 0 normalized to fallback-degradation
model-output-rebasing out_of_vocabulary 1 1 0 review required; not merged
model-routing accepted 323 322 169 canonical accepted name
multi-agent-orchestration accepted 338 334 336 canonical accepted name
multi-stage-retrieval out_of_vocabulary 1 1 0 review required; not merged
observability-tracing accepted 505 497 334 canonical accepted name
privacy-preserving-processing accepted 207 206 88 canonical accepted name
prompt-chaining accepted 109 108 53 canonical accepted name
prompt-versioning accepted 143 141 61 canonical accepted name
query-rewriting accepted 63 62 32 canonical accepted name
rag-retrieval accepted 567 560 306 canonical accepted name
retrieval-ranking accepted 147 146 49 canonical accepted name
safety-policy-enforcement accepted 357 353 214 canonical accepted name
semantic-caching accepted 69 68 31 canonical accepted name
source-citation-grounding out_of_vocabulary 1 1 0 review required; not merged
structured-output-contract accepted 318 315 159 canonical accepted name
structured-output-validation out_of_vocabulary 2 2 2 review required; not merged
synthetic-data-generation accepted 83 83 31 canonical accepted name
tool-use-orchestration accepted 729 717 655 canonical accepted name
workflow-decomposition proposed 16 16 14 review required; not merged
workflow-orchestration out_of_vocabulary 1 1 0 review required; not merged

Proposed, Alias, and OOV Usage Review

Original name Treatment Project Agentic How applied Evidence cue
async-message-orchestration review required ai-powered-marketing-compliance-monitoring-at-scale (performline, 2025) no S3 events, EventBridge, SQS queues, Lambda workers, result queues, and DLQs decouple high-volume extraction from downstream compliance checks. "EventBridge uses event-driven processing to route Amazon S3 events to Amazon SQS."
async-message-orchestration review required ai-powered-search-and-agent-automation-for-digital-asset-management (bynder, 2026) yes Bynder fans out upload and agent execution work through events, SQS, and Kafka so search and agent domains can process independently. "we publish an event on Kafka for each of the agents that needs to be executed."
async-message-orchestration review required architecture-patterns-for-production-ai-systems-lessons-from-building-and-failing-with-generative-ai-products (outropy, 2024) yes Semantic events decouple agents by publishing well-defined domain events to a bus instead of point-to-point calls. "using semantic events. There's different definitions of semantic events."
async-message-orchestration review required automating-supplier-ticket-management-with-llm-agents (wayfair, 2025) yes Ticket creation is decoupled from automation work through webhook events, Pub/Sub, and a consumer that invokes the graph. "a webhook publishes an event to a Pub/Sub topic. Our Pub/Sub consumer listens for these events"
async-message-orchestration review required automating-trading-card-copywriting-with-multi-agent-generative-ai (fanatics_collectibles, 2026) yes Long-running card-copy jobs are submitted from the UI to background workers and later surfaced through status/progress polling. "Instead of UI doing the work, the UI submits the work as a job in the background. A worker processes independently"
async-message-orchestration review required automating-workflows-with-ai-agents-across-the-organization (notion, 2026) yes Agents are triggered from Slack channels, run continuously, and post responses or tasks back into the appropriate channel/database. "Since it’s a Custom Agent, it runs continuously with no hand-holding from the team."
async-message-orchestration review required autonomous-ai-agents-for-engineering-tasks-with-closed-loop-feedback-systems (brex, 2026) yes Tasks enter from multiple trigger systems and are routed to agents that work independently until a PR is ready. "Tasks come in from Slack, Linear, GitHub, or a cron job. The orchestrator routes them to the agent pool."
async-message-orchestration review required building-a-production-ready-ai-phone-call-assistant-with-multi-modal-processing (realchar, 2023) yes RealChar uses an event bus so parallel systems can subscribe to relevant events and publish responses without blocking each other. "everything happening in midc is going to publish into a event bus"
async-message-orchestration review required building-a-self-healing-software-factory-with-ai-agents (software_factory, 2026) yes Slack threads, GitHub issues, scheduled planners, Sentry events, and PR reviews form asynchronous work queues for agents. "you can have an automation just run on it um whenever there is a new bug being opened and really have this autonomous flow"
async-message-orchestration review required building-an-enterprise-grade-ai-agent-for-recruiting-at-scale (linkedin, 2025) yes Each assistant instance has an identity and mailbox, and the lifecycle runs through asynchronous messages. "lifecycle is orchestrated entirely through asynchronous message passing."
async-message-orchestration review required building-durable-and-reliable-ai-agents-at-scale-with-dapr-workflows (humanlayer, 2026) yes Dapr pub/sub provides discovery and message delivery for distributed agents without broker-specific code, allowing asynchronous collaborative workflows. "How do you make them discover each other beyond a single process?"
async-message-orchestration review required building-linkedin-s-first-production-agent-hiring-assistant-platform-and-architecture (linkedin, 2025) yes All user-agent and agent-agent communication is modeled as messages on LinkedIn's production messaging platform, inheriting its delivery, retry, retargeting, and thread semantic... "all the communication between users and agents, and agents and agents should be modeled as a message."
async-message-orchestration review required building-modular-and-scalable-rag-systems-with-hybrid-batch-incremental-processing (bell, 2023) no The incremental path uses Pub/Sub messages plus Apache Beam windowing to coordinate document-change events and process only the most recent action, avoiding race conditions from... "the largest addition to to this architecture is the pub subtopic which listens to all the changes which occur in our knowledge based bucket"
async-message-orchestration review required building-production-grade-ai-agents-with-distributed-architecture-and-error-recovery (parcha, 2023) yes Parcha moved long-running agents to asynchronous processes with pub/sub, REST polling, SSE, Slack threads, RQ workers, and future webhooks. "We now run our agents as long-running processes asynchronously. A web service can still trigger agents, but instead of communicating bi-directionally through WebSockets, they po..."
async-message-orchestration review required building-production-ready-customer-support-ai-agents-challenges-and-solutions (gradient_labs, year n/a) yes The support agent is modeled as a state machine that reacts to conversation starts, new customer messages, missing signals, timers, and external state changes. "building we built AI Agents from a high sort of medium and lowlevel perspective from the high level perspective we architect our AI agents as state machines so in one of our blo..."
async-message-orchestration review required building-production-scale-ai-agents-with-extended-genai-tech-stack (linkedin, 2025) yes The core orchestration substrate is LinkedIn's messaging platform, which sequences long-lived tasks, supports retries, and can stream or split responses. "Long lived tasks could be broken down into a sequence of messages with guaranteed first in first out (FIFO) delivery"
async-message-orchestration review required enterprise-scale-cloud-event-management-with-generative-ai-for-operational-intelligence (fidelity_investments, 2024) yes CENTS uses event-driven routing over EventBridge rather than scheduled polling. "now essentially an event-driven system that gets this in real time out to the users"
async-message-orchestration review required intelligent-document-processing-for-education-quality-assessment-reports (bqa, 2025) no S3 events and SQS queues decouple ingestion, extraction, summarization, and assessment stages. "An event notification is sent to an Amazon Simple Queue Service (Amazon SQS) queue"
async-message-orchestration review required orchestrating-fleet-scale-ai-coding-agents-with-temporal-workflows (macroscope, 2026) yes Murmur turns PR comments, CI failures, and review bot findings into queued events that wake the VM and start another agent turn. "this VM will, um see a new event get queued up. And that event is a PR comment from me."
async-message-orchestration review required orchestrating-multi-agent-code-review-systems-with-improved-observability-and-reliability (cubic, 2026) yes Agents are connected through event-driven architecture and triggered from pull-request-related events. "offering built-in support for parallel execution and enabling agents to connect through its [event-driven architecture](h..."
async-message-orchestration review required parallel-asynchronous-ai-coding-agents-for-development-workflows (google, 2025) yes Jules is explicitly asynchronous, with tasks running in the background while developers continue other work and later inspect plans, tests, and PRs. "Jules is an asynchronous coding agent meant to run in the background and do all those tasks that you don't want to do in parallel"
async-message-orchestration review required platform-driven-ai-agent-orchestration-for-large-scale-engineering (linkedin, 2025) yes LinkedIn treats online, nearline, and batch invocations differently, with asynchronous execution and targeted human notifications for non-interactive work. "The developer submits a spec, and from there on out, everything happens async. The orchestrator spins up a sandbox with the agent, with identity and permissions and a consistent..."
async-message-orchestration review required production-agent-platform-architecture-for-multi-agent-systems (linkedin, year n/a) yes LinkedIn models ambient, long-running agent work as messaging, including user-to-agent and agent-to-agent messages with retry. "model longrunning asynchronous flows. We modeled this as precisely a messaging problem."
async-message-orchestration review required production-deployment-challenges-and-infrastructure-gaps-for-multi-agent-ai-systems (getonstack, 2025) yes The proposed fix uses message queues, retries, and dead-letter handling for reliable A2A communication. "message queue systems for reliable A2A communication, distributed tracing for tracking message flows across agents"
async-message-orchestration review required semantic-search-for-aviation-safety-reports-using-embeddings-and-hybrid-search (beams, 2025) no SQS queues orchestrate Lambda-based report processing, including spike handling for bulk customer uploads. "different steps run in like individual um containers and everything is being orchestrated by uh an SQSQ."
async-message-orchestration review required unified-pipeline-for-ai-token-spend-management-and-attribution (ramp, 2026) no Kafka decouples bursty webhook ingestion from storage and provides replayability. "Event Streaming with Kafka: The architecture uses Kafka as a critical decoupling layer between ingestion and storage. This choice addresses a fundamental challenge in LLMOps mon..."
async-message-orchestration review required voice-ai-agent-development-and-production-challenges (various_(canonical,_prosus,_deepmind), 2023) yes Slow actions such as purchases can be detached from the live call and completed with SMS follow-up. "For particularly slow processes like completing a purchase, the agent informs the user upfront, allows them to disconnect, and follows up via SMS when complete."
code-execution-sandbox review required mcp-protocol-development-and-agent-ai-foundation-launch (anthropic_/openai/_goose, 2025) yes Programmatic MCP/code mode lets a model compose multiple API-style calls in sandboxed code to reduce token and latency overhead. "the model generates you know code and what you effectively doing is this composability that the the model would have done anyway"
cost-aware-serving cost-aware-routing video-super-resolution-at-scale-for-ads-and-generative-ai-content (meta, 2024) no Meta targets VSR only to videos likely to benefit and uses CPU infrastructure where latency does not require GPUs. "The discovery that only medium-to-high quality input videos benefit from VSR enables targeted deployment that optimizes this tradeoff."
fallback-strategy review required managing-memory-and-scaling-issues-in-production-ai-agent-systems (gradient_labs, 2025) yes When latency looked like an LLM provider incident, the team adjusted its LLM failover system to route around provider degradation. "one of the many LLM model providers that we use might be about to declare an incident. On that day, that was indeed the case, and so we adjusted our LLM fail-over system"
feedback-loop-iteration eval-driven-development building-claude-code-scaling-ai-powered-development-from-terminal-prototype-to-production (anthropic, 2024) yes Features were identified and tuned from dogfooding, GitHub issues, Slack feedback, direct observation, and user revolt, rather than only from roadmap assumptions. "You try a thing, you give it to users, you talk to users, you learn, and then eventually you might end up at a good idea."
feedback-loop-iteration eval-driven-development building-custom-tracing-tools-and-development-infrastructure-for-ai-powered-meeting-notes (granola, 2026) yes Granola emphasizes tracing and preview infrastructure as feedback loops for iterating with LLMs rather than relying on one-shot prompts. "So, essentially, this this is basically a long talk to tell you that the answer isn't to one-shot better. It's about figuring out how you can make that feedback loop where it ki..."
fine-tuning-feedback-loop review required video-content-summarization-and-metadata-enrichment-for-streaming-platform (paramount+, 2023) no The roadmap includes human preference datasets, reward models, and scheduled fine-tuning updates for generated summaries. "The architecture supports continuous feedback loops essential for adaptive personalization systems. On Google Cloud, the reinforcement learning fine-tuning process is supported ..."
forward-deployed-engineering review required forward-deployed-engineering-for-enterprise-llm-adoption (openai_/ramp/nominal/_dataland, 2026) yes The companies use embedded technical teams to own customer outcomes, translate enterprise workflows into AI systems, and feed deployment learning back into product and model dev... "the fundamental thing about FTE is like embedding with specialists and understanding their tasks and then making sure the model can solve those tasks."
forward-deployed-engineering review required forward-deployed-engineering-for-enterprise-llm-deployments (openai, 2025) yes OpenAI uses embedded engineers to bridge enterprise workflows and model/product capabilities, making production deployments repeatable while surfacing product and research learn... "embedding with them, understanding their domain really deeply, and like sitting with the users on the ground and getting them to properly like adopt technology."
forward-deployed-engineering review required pivoting-from-gpu-infrastructure-to-building-an-ai-powered-development-environment (windsurf, 2023) yes Windsurf uses deployed engineers who work directly with customers to make sure enterprise deployments get value from the technology. "we also have these deployed engineer uh like sort of roles similar to what you said uh that get their hands really dirty with the technology"
guardrails safety-policy-enforcement building-evaluation-systems-for-ai-powered-healthcare-at-scale (sword_health, 2026) yes Some evaluators become real-time guardrails that prevent unwanted patient-facing content; medical-advice checks may use LLM-as-judge and failures trigger fallback or retries. "some uh evaluators can uh be used as uh guard wells um and they act as real-time safety checks that prevent unwanted content from reaching the end uh user."
human-in-the-loop human-in-the-loop-gate building-claude-code-scaling-ai-powered-development-from-terminal-prototype-to-production (anthropic, 2024) yes Plan mode and verbose output kept humans in the loop at the specification and observation points where agent execution could otherwise go off track. "Once the plan is good, and sometimes it takes a little back and forth, they just get clawed to execute."
incremental-rollout review required mainframe-to-cloud-migration-with-ai-powered-code-transformation (mercedes-benz, 2025) yes The facade allowed controlled cutover, rollback, and staged movement from mainframe to cloud rather than a single high-risk switch. "Control cut-over, we've said that already, like you can control the systems, when should they change? And then it's your decision to turn over;"
latency-aware-model-routing review required llm-powered-search-relevance-re-ranking-system (leboncoin, 2023) no The architecture routes expensive ad encoding to offline batch computation and keeps only query encoding and scoring in the online path. "having a two-towers model architecture allows us to compute the ads and queries representations at different times."
model-fallback fallback-degradation video-content-summarization-and-metadata-enrichment-for-streaming-platform (paramount+, 2023) no The transcription system can fall back to speech-to-text models, including a containerized Whisper service, when transcripts are absent or poor. "The speech-to-text (STT) process serves as a fallback method when transcripts don't exist or are of insufficient quality."
model-output-rebasing review required production-scale-nlp-suggestion-system-with-real-time-text-processing (grammarly, 2022) no The client rebases active model suggestions onto every user edit so positions and applicability remain correct without backend re-inference. "The rebase procedure is the technical heart of how Grammarly maintains suggestion accuracy as documents evolve."
multi-stage-retrieval review required llm-powered-search-relevance-re-ranking-system (leboncoin, 2023) no ElasticSearch performs first-stage retrieval, and the neural re-ranker only rescored top-k candidates before final ranking. "a pool of ads is first retrieved and ranked by our ElasticSearch (with a TF-IDF li..."
source-citation-grounding review required source-grounded-llm-assistant-with-multi-modal-output-capabilities (google_/_notebookllm, 2024) no NotebookLM's citation model tracks where answer content comes from and supports fact checking. "because of our citations model. The fact that we know exactly where every piece of the answer comes from"
structured-output-validation review required multi-agent-ai-system-for-automated-test-case-generation-in-payment-systems (amazon_amet_payments, 2026) yes Pydantic schemas validate generated test-case objects before they propagate into test-management workflows. "Pydantic automatically validates LLM responses against defined schemas to facilitate type correctness and required field presence."
structured-output-validation review required multi-agent-ai-system-for-investment-thesis-validation-using-devil-s-advocate (linqalpha, 2026) yes The rebuttal prompt requires JSON fields for assumptions, counterarguments, citation metadata, and materiality risk flags. "Step 3. Structured Output For each assumption, output in JSON with the following fields:"
workflow-decomposition review required ai-powered-digital-co-workers-for-customer-support-and-business-process-automation (neople, 2024) yes Neople guides customers to automate concrete workflow slices instead of attempting to encode every edge case at once. "we really look at the concrete workflows, specific workflows with our customers and guide them to start with those rather than trying to automate the whole thing. And that reall..."
workflow-decomposition review required ai-sales-representatives-for-inbound-lead-conversion (showme, 2025) yes The sales funnel is represented as workflow steps and smaller agent responsibilities such as greeting, qualification, pitching, and follow-up. "you need to divide in different agents actually inside it with different prompts and one is going to take over the the greetings and first discovery."
workflow-decomposition review required architecture-and-production-patterns-of-autonomous-coding-agents (anthropic, 2025) yes Research, documentation reading, testing, and code review can run in subagents with isolated contexts. "researcher docs reader testr runner code reviewer"
workflow-decomposition review required architecture-patterns-for-production-ai-systems-lessons-from-building-and-failing-with-generative-ai-products (outropy, 2024) yes Outropy split Slack summarization into smaller semantic workflow steps such as discussion extraction, deduplication, ranking, and personalized summary. "one tiny pipeline that summarized discussion from our Slack channels. Then we would deduplicate discussions"
workflow-decomposition review required automating-workflows-with-ai-agents-across-the-organization (notion, 2026) yes Employees automate narrow repeatable workflows such as Q&A, feedback routing, web-team triage, style-guide feedback, and weekly work tracking. "Describe a workflow once, set a trigger or schedule, and the work happens automatically whether you’re online or not."
workflow-decomposition review required building-a-notion-style-note-taking-app-with-ai-agents-and-automated-software-development (software_factory, 2026) yes The feature planner decomposes a long product spec into single-concern GitHub issues so agents can work sequentially on foundations before more complex features. "the feature uh planner, which is right here. Uh it does take the product spec and translates it into the individual uh issues"
workflow-decomposition review required building-production-ready-ai-agents-for-internal-workflow-automation (vercel, 2025) yes The talk decomposes business processes into trigger, steps, outputs, and replaceable agentic substeps; the same decomposition is used for lead, abuse, and analytics agents. "Workflows. This is the simplest possible workflow. It has no branches or loops, but it could, right?"
workflow-decomposition review required building-production-ready-llm-agents-with-state-management-and-workflow-engineering (renovai, 2023) yes Workflow engineering decomposes production control into router-plus-code, plan-and-execute, state-machine, or autonomous patterns depending on how much structure is needed. "in workflows engineering we decide what the human design and which kind of autonomy do we get to give to the agent itself"
workflow-decomposition review required multi-agent-architecture-for-automating-commercial-real-estate-development-workflows (build.inc, 2025) yes Build.inc decomposes diligence into single-purpose tasks and subgraphs to make the workflow modular and testable. "Breaking workflows into smaller, single-purpose tasks enables more efficient and accurate execution."
workflow-decomposition review required private-equity-ai-transformation-lessons-from-portfolio-companies (pwc_/warburg_pincus/_abrigo, 2025) no The panel repeatedly scopes AI to narrow workflow slices such as credit memos, alert dispositions, document checking, and catalog image enrichment rather than broad transformation. "inject AI into credit memos or alert dispositions, wherever that might be."
workflow-decomposition review required production-llm-systems-at-scale-lessons-from-financial-services-legal-tech-and-ml-infrastructure (nubank,_harvey_ai,_galileo_and_convirza, 2024) yes The cases repeatedly break broad workflows into narrow modules with explicit contracts, allowing independent model selection, evaluation, and human intervention. "Every organization emphasizes breaking monolithic systems into narrow, specialized components. This enables targeted optimization, clearer evaluation, and easier maintenance."
workflow-decomposition review required production-ready-llm-integration-using-retrieval-augmented-generation-and-custom-react-implementation (buzzfeed, 2023) yes When prompts became too complex, Buzzfeed broke instructions into multiple API calls that informed application control flow. "Their solution was to break complex instructions into multiple API calls that informed the control flow of the application."
workflow-decomposition review required productionizing-llm-powered-data-governance-with-langchain-and-langsmith (grab, 2024) no Grab splits one broad governance classifier into PII and non-PII subtasks and chunks wide tables. "• Task decomposition: Split the model into two parts—one for PII tags and another for all other tag types. This follows the principle of breaking complex tasks into smaller, mor..."
workflow-decomposition review required replacing-complex-feature-implementation-with-prompt-based-skills-git-worktrees-in-production (cursor, 2026) yes Cursor decomposed the feature into existing primitives: skills/commands, subagents, git worktrees, setup scripts, and parent-agent comparison. "there are two primitives that we could use to effectively allow Cursor users to use work trees by simply leveraging two primitives. One is agent skills and the other are sub age..."
workflow-decomposition review required scaling-ai-coding-agents-through-automated-verification-and-specification-driven-development (factory_ai, 2025) yes Factory says robust validation enables large modernization projects to be decomposed into parallel subtasks for multiple agents. "decomposing a large-scale modernization project uh into a bunch of different subtasks"
workflow-decomposition review required scaling-financial-software-with-genai-and-production-ml (ramp, 2023) yes Ramp decomposes sales and finance workflows into data, decision, and automation pieces instead of automating whole jobs at once. "look at where can you understand someone's workflow and decompose it to either build a data platform, in some cases more generative AI based automation"
workflow-orchestration review required video-content-summarization-and-metadata-enrichment-for-streaming-platform (paramount+, 2023) no Transcript storage events trigger a multi-stage generation workflow that moves from transcription to summary/metadata generation and personalization integration. "The generation phase is triggered by events when transcripts land in storage buckets. The generation process stores outputs in Firestore as key-value pairs (with content ID as t..."

Reviewer Checks