Topic
Source-linked workflows that name builder. Part of the AI agent use cases on UsecaseBoard, or browse all use cases.
OpenAI’s practical guide frames agents for complex judgment (fraud/refunds), brittle rule piles (vendor security reviews), and unstructured docs (home insurance claims)—not for simple deterministic checklists.
Official Agents API cookbook: connect Slack threads to persistent agent sessions, workplace tools, and isolated sandboxes so workplace chat becomes a durable agent runtime.
Product Hunt’s Cronloop lets you describe a recurring job in Markdown, pick Codex or Claude Code, connect tools, and run agents on a loop with durable memory that can self-improve each run.
MarkItDown documents an OCR plugin that accepts the same model-client configuration used for image descriptions and adds OCR during file conversion.
The CrewAI project expands a brief, selects from supplied templates, and adapts the chosen page; its README requires users to provide appropriately licensed template files.
The CrewAI example assigns agents distinct coding and review responsibilities and provides configuration and commands for running a game-building crew.
The Academy example combines Claude's document and image understanding with coding to generate a portfolio page reflecting supplied professional material and visual preferences.
The project-management guide generates user acceptance tests for a website form, organizes the cases in Sheets, and drafts instructions for the colleague running them.
The example distinguishes waiting reviews, requested re-reviews, and approved changes and prepares reminders containing the work's size and blocking context.
The scenario distinguishes customer-visible changes from internal work and drafts a recurring changelog with links back to the merged pull requests.
Lindy's scenario identifies actionable bugs in an on-call channel, checks existing issues, and attaches new reports to the appropriate ticket.
The example narrows a funnel change by cohort and authentication path, then correlates analytics, tickets, and deployment history to investigate likely causes.
The published agent watches new Jira issues, checks configured required fields, and posts a comment requesting missing information.
A developer leaves Codex CLI running for hours to add debug prints and test hypotheses against a deterministic front-door unit test until it can explain the bug with evidence.
A company built a Grok Bot purchasing agent that uses Ramp to mint a different virtual credit-card number for each transaction instead of sharing one corporate card.
Agent Use Cases records a failed Cursor attempt to scaffold a second module because it missed the four or five reference files. The useful lesson is to make reference-context inspection explicit before generation.
Anthropic's published PDF skill supplies tool recipes for reading tables, OCR of scanned pages, combining outputs, and producing files, giving an agent concrete operations for document work.
CrewAI's Instagram example uses specialist agents to turn an idea into a post and documents a local Ollama/OpenHermes setup alongside browser and search tool configuration.
CrewAI's Markdown reviewer gives an assistant a deterministic linting tool, then asks it to turn the tool's errors into readable change recommendations rather than relying on model-only syntax judgment.
Microsoft's Data Formulator combines AI data transformation with visual chart editing and Data Threads, letting analysts compare alternative questions without losing the path through their analysis.
LangChain's enrichment template searches and reads the web, extracts facts into a supplied schema, and checks completeness, with configurable models and tools in LangGraph Studio.
Open Deep Research provides a public LangGraph implementation with separate research, summarization, compression, and report roles, configurable search providers, MCP support, and an evaluation workflow.
A LangChain sample demonstrates parallel trip-input preparation, a join barrier, weather research, a human interrupt for preferences, and itinerary assembly using the same graph locally and when deployed.
Anthropic's recipe has Claude find themes across Intercom, survey CSVs, and interview PDFs, then create an auditable Excel workbook with classifications, trends, quotations, and feature requests.
Anthropic's Cowork recipe combines incident-channel messages, alert chronology, logs, and a local template to produce an evidence-linked postmortem and action tracker.
Zapier's purchase-request template matches requested software against an approved Google Sheet, suggests existing alternatives, and escalates genuinely new requests with researched context rather than authorizing purchases.
Zapier's Brand Voice Analyzer examines tone, sentence structure, vocabulary, formatting, and sensitivity risks, then produces a structured style guide and examples for review.
This n8n template extracts invoice PDFs with GPT-4o, checks invoice numbers and totals, assesses category and risk, and uses a wait-and-resume approval path before recording the final status.
An n8n template combines an Ollama agent, three Supabase vector knowledge bases, Postgres memory, request validation, deduplication, and confidence-based escalation across Slack, Gmail, and callbacks.
An appraiser in South Korea reports spending two months building a report-writing program with Claude Code, then having colleagues test it and discussing adoption within the firm.
A builder reports that a two-court booking system became more reliable after moving availability checks, resource assignment, and booking IDs out of the LLM and into deterministic n8n workflows.
Operator keeps a multi-bot Grok setup clean by giving each bot one job, a never-do list that points to the right sibling bot, shared how-to skills, and a dedicated weekly bot that audits those skills for anything stale.
Agent Use Cases summarizes an HN user's ChatGPT-plus-GitHub-MCP setup: per-repository AGENTS.md files point to saved discussion summaries, decisions, and research notes for later sessions.
PersonalAgentEval tests everyday tasks against accounts with substantial prior history. Its X report highlights a pharmacy task where most agents omitted a previously promised caregiver notification.
A Muse user reports describing a simple object and its approximate size, receiving a model file that opened in their printer software.
This workflow records agent outcomes in an n8n Data Table with linked SHA-256 hashes, provides a verification endpoint, and prepares daily integrity reports.
Agent Use Cases describes getonup, an open-source Cloudflare Workers project for publishing agent-built static pages on an owned domain with a dated index and Zero Trust access.
Jason Lemkin reports that Muse flagged an expiring domain whose expected automatic renewal had failed after an account migration; the team renewed it themselves.
Agent Use Cases describes ldraw-nova: an agent plans submodels, writes generator scripts, searches parts, checks collisions and gaps, and iterates on renders before exporting CAD files. The source notes substantial model and runtime limitations.
A developer describes a pipeline in which Gemini saves meeting transcripts to Drive, Grok Bot turns them into a PRD, and Cursor implements it; exceptions feed back into subsequent coding tasks.
The author reports that Muse installed PyTorch and Depth Anything V2 to estimate distance from a robot's camera and used a sanity check before a movement command.
An X post describes an engineer's Grok Bot template that starts from an API's data shapes, builds an MCP-and-skills scaffold, and proves the resulting Cursor plugin locally.
OpenAI's Cookbook provides a synthetic support-agent optimization sprint comparing prompts, tool controls, model routing, caching, and offline follow-up; its default simulation is not a production performance benchmark.
TikTok's US LIVE Diagnostics playbook walks sellers through an AI recap, funnel-step diagnosis, and follow-up prompts that turn session metrics into script, product, and pacing changes.
OpenAI's data-analyst example combines schema discovery, metric context, read-only SQL, and explicitly saved corrections in one persistent investigation; database execution stays in the application.
OpenAI's SRE example ties an alert webhook to a persistent agent session and Slack thread, investigates bundled telemetry and code-change fixtures, and records rollback decisions without performing deployments.
The Gardener Grok Bot template focuses on small maintenance changes that remove provably unused or duplicate code without changing behavior.
TikTok's US Seller Center guide documents an agentic assistant that uses shop context to diagnose issues and guide actions, including Shopify products failing to sync.
The workflow scores incoming mail for prompt-injection patterns, routes high-risk messages to review, and uses Gemini for structured triage of the remaining messages.
Frigade's HN maker post describes observing an app's existing requests, deriving typed endpoint recipes, and letting the app owner enable them as agent tools; changes can trigger recipe updates.
A builder's LinkedIn post describes preprocessing and routing chat inquiries for an educational institute so the LLM receives the relevant request type instead of every raw message indiscriminately.
A practical Copilot CLI guide demonstrates remote session access through GitHub's web or mobile app while the process remains on the original computer, useful for reviewing progress and steering a running task.
An Indie Hackers walkthrough connects a Jotform request, n8n planning and content generation, a GitHub branch, a Vercel preview, an approval step, and post-change monitoring.
VulnScout assembles public vulnerability identifiers, exploitation-status signals, and remediation references into a structured research brief.
Make's memory guide provides a concrete read-normalize-reason-route-write pattern keyed to a customer or case, with module-level inspection and recovery paths for missing or malformed state.
The template runs question-and-reference-answer cases through a support agent and uses a separate GPT-4 judge to score open-ended responses in n8n's evaluation view.
A Technically writer documents a month-long Claude Code experiment managing a Meta Ads account: fresh sessions read prior logs, pull performance data, decide, act or abstain, and commit the updated history.
The small template accepts a public GitHub skill link and helps add its instructions to Gumloop for a repeatable agent task.
A founder describes a working beta that scans a repository, asks product questions, and writes a behavior contract distinguishing confirmed decisions from unsettled assumptions.
Christopher Johnson describes keeping most of a 13-agent content pipeline hierarchical while giving market, technology, and policy researchers peer messaging, followed by a synthesizer producing the same downstream brief format.
AgentMail's Product Hunt page describes per-agent inboxes, incoming-message events, threaded replies, semantic search, and attachment extraction. This lead is an infrastructure recipe rather than a measured customer result.
Jam's creator reports using a named agent to build a campaign dashboard, draft channel-specific content, and operate marketing tasks, with persistent logs and a shared desktop interface for several coding runtimes.
A public Discord showcase introduces Glimpse, a small native webview library and Pi integration for agent status interfaces. The linked README documents bidirectional JSON messages and an optional cursor-following companion.
Make's Post-Call Hero case study describes a transcript-driven agent that drafts follow-up email, prepares CRM changes, and can create stakeholder Slack channels, supported by fixed context files and a limited tool set.
A public LiveKit Slack Q&A explains waiting for a participant and checking participant kind before beginning a SIP interaction, with a minimal JobContext example and a link to lifecycle documentation.
A Lobsters introduction points to Cua-Bench, an open-source framework for evaluating computer-use agents across operating-system and UI variations rather than assuming one successful interface generalizes.
A product manager's tutorial walks through a Claude Code project with a lightweight specification, a Google Sheets question source, and browser feedback used to repair the data connection.
The template connects a locally hosted Signal API bridge to an OpenAI-powered n8n assistant, allowing additional tools to be added to the chat workflow.
Jerry Liu links a worked example by Tuana Celik that classifies incoming email and attachments as invoices or expenses, then routes them through LlamaCloud extraction and LlamaIndex workflows.
An English n8n tutorial uses Google Vision OCR for PDF and image invoices, an LLM for structured fields, and Google Sheets for storage, making the OCR-to-extraction sequence explicit.
A public Mastra Discord thread shares a working custom execute wrapper that calls an agent with a Zod structured-output schema and returns the validated object; agent-step visualization remains the author's open question.
Google's tutorial builds a conversational analytics agent that discovers BigQuery metadata and generates SQL against the public thelook_ecommerce dataset, using a maintained toolset rather than custom database plumbing.
Zack Proser's review describes assigning repository-and-branch-specific maintenance jobs to Cursor Agents, following up in each task's own context, and opening the changes directly in the editor for hands-on refinement.
Make's Kevin Cookie Company tutorial builds a fictional support assistant using a Llama model, a Google Docs FAQ tool, a separate email tool, and a Tally trigger, illustrating reusable versus scenario-specific tools.
Google's multimodal tutorial uses a LangGraph orchestrator to send files to specialist Gemini agents, gather their modality-specific findings, and synthesize whether a requested object is present.
A Show HN author shares a model-agnostic generator used to evaluate document extraction, with configurable schemas and locales plus malformed totals, missing fields, and merchant-name typos.
Wordware's maker description outlines a content agent that researches a topic with web-enabled models and loops over a table of contents, using the IDE's logic and function-calling features.
Skyvern's Product Hunt listing names automated invoice retrieval as a deployed use case and exposes hosted browsers, a prompt-based workflow UI, an SDK, and MCP as implementation paths.