Agentic AI Fast Track - Building Agents w/ Tools @Pinkfish

Unlock Agentic AI to access the expertise of a thousand people, giving you superpowers at work with easy access to vast knowledge and capabilities never before possible.

Agentic AI is an LLM operating in a loop with tools. The model selects, uses, and reevaluates tools like email, drive, tickets, and cases to fulfill user goals.

Explore the agentic ai loop where an agent uses tools to execute workflows and Pink Fish generates production-ready tools from an English bullet list for custom ai workflows.

Get oriented with the pink fish UI, from the universal agent and workflows to integrations and orchestration. Learn to manage resources, runs, and account settings, including billing and vault features.

Launch your first hello world workflow by prompting a coding agent to generate and run deterministic code. Iterate with prompts to refine the output and optionally add emojis.

Discover slash commands in pink fish that unlock AI tools and llms for classification, browser tasks, document processing, and data visualization, plus image and audio generation workflows.

Connect to apps through integrations, authenticate with Zendesk, and pull data into a workflow using slash commands, built-in services, and the Zendesk skill.

Master integration troubleshooting by re-specifying the connection in each step and using the slash command to reference the integration, enabling reliable ticket creation with Zendesk.

Master a Zendesk-like data workflow by loading a sample JSON data set, parsing and enriching it with a triage team rule, and building a simple HTML dashboard for secure sharing.

Generate a visual overview of your workflow with a summary and diagram, then dive into each step for details and debugging, and share the flow with others.

Explore how inputs drive dynamic workflows by testing with inputs like greeting. Reference variables using inputs.greeting and prepare your workflow for publishing.

Publish releases for public APIs teaches you to version and lock a workflow, create API triggers and keys, and test inputs end-to-end with curl and orchestration insights.

Learn to configure triggers: API endpoints for external calls, email triggers with approved senders or domains, and schedule triggers for timed workflow runs, plus testing inputs.

Upload csv and pdf input files to your workflow, access their data in subsequent steps, and avoid processing data in the same step to prevent errors.

Learn to use automatic and custom logging to debug steps in agentic ai workflows, requesting detailed logs to inspect processing, filtering, and value calculations.

Reference data from previous steps using insert path into chat to fetch exact values or entire objects, and rerun steps after reordering to ensure correct results.

Learn to distribute your workflow by turning it into a runnable app with a landing page, hero image, and shareable templates for internal or public use.

Explore workflow app options for the overview page, including markdown descriptions, AI-generated text, lead image selection, poster images, video URLs, tags, diagrams, and inputs.

Configure inputs by selecting types, labels, placeholders, and constraints; upload a concept file or image to a collection, return its URL, and use a slider to set billboards for workflow.

Lock the latest release and publish your built workflow as a public template, generating links to the workflow, author page, organization, and tag groups while cloning only the workflow code.

Explore pink fish's built-in data features, including three resources—databases (a key-value store), files (a file repository), and knowledge bases with indexed documents and semantic search.

Create and explore a key-value database that functions as a document store, using items with keys, sort fields, and metadata, where the value stores the content, with a 10,000-item limit.

Create and populate a Mydb database using a new workflow to generate demo orders for home repair products, then refine to store ten separate items for easier querying.

Learn to query an orders dataset by customer and price, use sort keys to narrow results, handle schema changes, and build data-driven workflows from retrieved data.

Explore file storage as a bucket-like system that holds, uploads, and retrieves files, uses tags for sorting, and toggles public or private access within the mystuff workspace.

Retrieve a file from file storage in a workflow, then manipulate the data in steps, add a GDP column, and save it as a new version for version history.

List files in a storage collection, create a new file, and append content to build a continuous log across workflows.

Explore knowledge bases and the rag pipeline, which embeds and chunks documents for semantic retrieval in a vector store, then match queries to relevant chunks with an llm.

Match questions to knowledge base chunks and view confidence scores. Click a chunk to reveal exact matching text and see how chunk relevance feeds the rag pipeline toward llm processing.

Demonstrate the rag pipeline by retrieving the matching chunks from a knowledge base, vectorizing and chunking them, then asking the llm to answer the user's query in workflows and agents.

Query knowledge bases from workflows and agents using slash commands to search across collections, retrieve chunks, and rely on ai search results formatted in markdown.

Differentiate workflows from agents: workflows run deterministically, while agents use the LM to think non-deterministically, and can call deterministic workflows for onboarding.

Use the universal agent, an LLM, to analyze documents, manage files, and transcribe images, leveraging integrations like Zendesk and Google Drive to complete one-time tasks.

Create custom agents by crafting instructions and wiring workflows, databases, file stores, and knowledge bases, then use previews to demonstrate a pirate agent connected to Google Drive.

Build an onboarding agent that collects user details—name, role, monitor preference, city, and favorite food—via an interactive survey, guided by prompt engineering to stay on track.

Store data in the data store by granting read/write access, defining onboarding as the key, and writing entries with a first name and last name sort field.

Onboard an agent to collect and save data, define what to do when it's done, and use perplexity to fetch restaurants, then offer a gift certificate to delight users.

Build an RFP answer agent that can populate a database with answers to RFP questions and pull data from any data source to tailor responses.

Populate the RFP dataset by uploading a local file, generating sample questions and answers with an llm (Gemini 2.5 flash), and storing them in the RFP info data store.

Build an RFP agent that reads from a data store to answer questions, control access with read permissions, and ground responses in data to prevent hallucinations.

Demonstrate answering RFP questions from a document in batch by feeding a document to an agent, searching the data store for each question, and compiling accurate answers without fabrication.

Demonstrate using an AI agent to fill out a Google Sheet with 13 RFP questions and answers in two columns, then export them to a ready-to-use RFP document template.

Turn workflows into tools by coupling a nondeterministic agent with a deterministic price validation workflow. The agent captures new prices and the workflow flags what passes or fails against rules.

Build and validate a price-checking workflow for an agent, using current and discount price inputs to ensure discounts stay within 20% of the current price, with test via trigger.

Assign the price validation workflow to the agent, configure required inputs like discount price and current price with a trigger and API key, then publish and test using sample prices.

Craft prompts to guide a price validation workflow for CSV inputs, return validations as a new column, and empower agents to work with deterministic workflows.

Learn to run batch jobs at scale with the pink fish batch and queue service, processing thousands of items sequentially or in parallel for invoice processing and contract review.

Develop an expense violations dashboard by processing thousands of expense items through a pipeline that uses the LLM as the intelligence engine, with options to view violations or all expenses.

Build a one-at-a-time expense workflow that processes an expense csv against a travel policy using an llm, flags violations with reasons, and emails the report.

Explore a batch service workflow with producer, consumer, and aggregator: load csv rows, process each row with the lm, store results, and generate a summary report.

Build consumer workflow that ingests expenses, loads travel policy, analyzes them with Claude sonnet four, flags violations with reasoning, stores results under expense review, and exposes an API endpoint consumer.

Learn to run a producer manually by creating an expense queue, tuning concurrency, retries, and priority, uploading a comma-separated values file, and testing batch processing.

Aggregate expenses from the data store and build an HTML dashboard of expense violations. Expose a batch aggregator API to automate the producer, consumer, and aggregator cycle.

Resolve the dashboard bug by updating the agent to fix the violations click issue, then publish the dashboard to file storage and host it at a URL for easy access.

Use a workflow as producer to load csv data from a step, url, or manual paste, and run a batch job with the batcher to process items.

Chain batches in a producer–consumer–aggregator workflow to process data sequentially, using trigger calls to move between producers, with a call center quality management example illustrating progress tracking and data flow.

Build a batch-driven transcription pipeline from S3 files to quality analysis, culminating in a call center dashboard, using producer, consumer, and aggregator steps with batch IDs.

Learn how to set up and manage organizations and sub organizations using enterprise or teams accounts, invite admins and members, switch between organizations, and control permissions.

Learn how to create, own, and securely share workflows and agents within an organization, set permissions (view or manage), invite members, and manage cross-organizational copies.

Duplicate a workflow to share a copy with another organization without collaborative editing; you must have access to the target organization and see runs only in that organization's orchestration.

Share integrations separately from workflows to control access to sensitive data. Ensure recipients run workflows only with their required integrations, while development and production access are managed independently.

Visualize how a user shares a workflow and its Zendesk integration with a teammate, granting access, refreshing permissions, and enabling the other person to run the full workflow.

Share resources such as file storage and knowledge bases across accounts. Ensure access for any resource used by a workflow; grant permission or connect your own resource to continue.

Share agents with individuals or organizations and grant fine-grained per-resource permissions. Set read, write, and delete access for files, databases, workflows, and knowledge bases via Slack or Teams.

Earn certification by completing a capstone project: automate a time-consuming process with a workflow and video, or build graphics or a useful agent, then post on Discord for LinkedIn badge.