Workflow
AI System Prompt Design Workflow
Design a system prompt that holds up in production — define the role precisely, engineer the behavior and guardrails on top of it, then check it reads clearly before you ship.
Build an AI Support Agent with AI
The full path to a support agent you can put in front of customers — write its instructions, ground it in your docs, route and handle tickets, then evaluate and cost-control it before it goes live.
Overview
Most 'build an AI support agent' guides stop at a clever system prompt — which is exactly the part that demos well and fails in production. A support agent customers actually rely on is a system, not a prompt: it needs instructions it follows under pressure, grounding so it answers from your real documentation instead of inventing policy, a way to classify and route what comes in, the support-handling behavior itself, a loop that learns from feedback, an evaluation harness that proves it behaves before launch, and cost control so a busy day doesn't blow the budget. This project walks that whole journey — each stage connecting to a NewPrompt workflow you can run on its own — from a blank system prompt to an agent you can put in front of real customers with your eyes open. You bring the product and policy judgment; the project makes sure no stage of a real agent build gets skipped.
The journey
Each stage runs a NewPrompt workflow, with a supporting resource and tool. Work them in order — the output of each stage feeds the next.
See the execution map →Design
Design the right solution before building.
-
Write the agent's system prompt
Start with the standing instruction that defines the agent — its role, tone, boundaries, and the rules it operates under on every reply. This is the agent's constitution; everything later builds on it.
Outcome A system prompt that defines the agent's role and rules. -
Instruct it to act across steps
A support agent doesn't just reply — it looks things up, decides, and escalates. Give it an explicit plan and guardrails so it follows a reliable sequence instead of improvising when a conversation gets complicated.
Outcome Agent instructions and a step plan it executes reliably.
Build & Refine
Build, test, secure, and make it production-ready.
-
Ground it in your knowledge base
An agent that answers from memory invents policy. Prepare your help docs for retrieval — sized and grounded — so the agent answers from your real documentation and admits when the docs don't cover a question.
Outcome Source docs prepared so the agent answers grounded, not guessed. -
Classify and route what comes in
Not every ticket is the agent's to answer. Classify incoming messages by intent so the routine ones get handled, the sensitive ones escalate, and nothing lands in the wrong place.
Outcome Incoming tickets sorted and routed by intent. -
Handle the support conversations
Shape the actual support behavior — on-brand replies, the right level of detail, and a clean handoff to a human when the agent should step back. This is the work the customer experiences.
Outcome On-brand support replies with a clear escalation path. -
Learn from the feedback
Every resolved and escalated ticket is signal. Pull the themes out of real interactions so you know where the agent helps, where it fails, and what to fix next — instead of guessing.
Outcome Themes from real interactions that point to what to improve. -
Evaluate it before it ships
Don't launch on a good demo. Build test scenarios with expected outputs, catch the failures and hallucinations, and set up a regression check — so you ship on evidence the agent behaves and stays behaving.
Outcome Tested behavior, hallucinations caught, and a regression guard. -
Run a security review
An AI support agent touches customer data, escalation paths, and the integrations it runs on — review its instructions and those paths the way an attacker would, and back the findings with tests, before it's live.
Outcome The support-agent instructions, customer-data handling, escalation paths, and integration boundaries reviewed for vulnerabilities before release. -
Keep the token costs in check
A support agent's bill scales with every conversation. Measure the cost per interaction, trim the waste in prompts and context, and re-measure — so the agent stays affordable at real volume.
Outcome Cost per conversation measured and trimmed before scale.
Ship & Validate
Ship with confidence and validate results.
-
Validate the product with real evidence
The agent is live enough to face customers — now measure whether it actually helps. Weigh deflection, resolution rate, and CSAT against the goal you set for it, render a hit/partial/miss verdict, and decide whether to iterate, pivot, or scale before widening its remit.
Outcome The shipped outcome measured against its success signal, with an iterate, pivot, or scale decision documented.
Expected outcome
An AI support agent that's grounded, evaluated, and cost-controlled — instructed to behave, answering from your real docs, routing what it can't handle, improving from feedback, and proven against test scenarios before launch — a support agent you can ship to customers, not a demo that breaks on the first hard ticket.
Best for
- Putting an AI agent in front of real customers, not a toy demo
- Grounding the agent in your own docs so it stops inventing answers
- Teams that need routing, handling, and evaluation, not just a prompt
Not for
- A general-purpose chatbot with no support domain
- Replacing your whole human support team in one step
FAQ
Will this replace my support team?
No — it builds an agent that handles the repetitive front line and escalates the rest, freeing humans for the hard cases. The journey includes routing and evaluation precisely so you keep a human in the loop where it matters.
How does it stop the agent from making things up?
Grounding. A core stage connects the agent to your knowledge base so its answers come from your real docs, not the model's guesses — and an evaluation stage checks it before it ships.
Do I need my own model or infrastructure?
You bring the model and where it runs; this project gives you the instructions, grounding, routing, and evaluation that turn a raw model into a support agent. It is the operating design, not the hosting.
How do I build an AI support agent?
You build it as a system, not a single clever prompt: write its system prompt, instruct it to act across steps, ground it in your real docs, classify and route tickets, handle conversations, learn from feedback, then evaluate and cost-control it before launch. This blueprint sequences all ten stages so no part of a real agent build gets skipped.
What documents do I need before building an AI support agent?
Your real help documentation, before anything else. The grounding stage prepares your existing help docs and knowledge base for retrieval — sized and grounded — so the agent answers from your policy instead of inventing it. You also bring product and support judgment; the blueprint supplies the workflows, prompts, and tools to turn those docs into a grounded agent.
Does this blueprint connect to my helpdesk automatically?
No. NewPrompt does not wire the agent into Zendesk, Intercom, or your helpdesk for you. The blueprint designs the classification, routing, and escalation logic — how tickets get sorted and when they hand off — and you connect that logic to your own helpdesk and own the integration and its security review.
Workflows in this project
Workflow
AI Agent Instruction Workflow
Instruct an AI agent that runs on its own without it wandering off — anchor it to a role, write the agent system prompt, then lay out the multi-step plan it works through.
Workflow
AI RAG Context Workflow
Prepare documents for a RAG system so retrieved answers stay accurate — budget the chunk size to the model, ground the sources against drift, and split them on clean boundaries for retrieval.
Workflow
AI Classification Workflow
Build a text classification step you can automate on — pull out the unit to classify, assign a label from a fixed set, and validate the label is one you actually allow.
Workflow
AI Customer Support Workflow
Run inbound support the same way every time — triage and route the ticket, pull the details that matter, draft a reply in a consistent voice, and log the resolution for the record.
Workflow
AI Customer Feedback Analysis Workflow
Turn a pile of reviews, surveys, or support comments into themes and priorities — extract the real signal, classify it by theme and sentiment, then summarize what's worth acting on.
Workflow
AI Agent Evaluation Workflow
Find out whether an AI agent behaves before users do — define what correct means, build test scenarios with expected outputs, catch failures and hallucinations, then regression-test each version.
Workflow
AI Security Review Workflow
Review code for what an attacker would do, not just what tests catch — anchor the model as a security engineer, run a threat-focused review, then back the findings with auth and input tests.
Workflow
AI Cost Optimization Workflow
Cut what an AI feature costs without dumbing it down — price the prompt as it runs today, see where the tokens go, trim the waste, and re-measure to prove the saving holds at scale.
Workflow
AI Product Validation Workflow
Find out whether the thing you shipped actually worked — define the success metric, plan the measurement, classify the real evidence, then render a verdict and an iterate / pivot / scale decision.
Resources used in this project
Resource
AI Agent System Prompt
Build a system prompt for an autonomous, tool-using AI agent — plan before acting, one tool at a time, act on real results, stop before irreversible steps.
Resource
AI Workflow Prompts: From Goal to Sequence
An AI workflow prompt set turns one goal into an executable sequence — objective, prompt, and expected output per step, ready to run top to bottom.
Resource
Context Window Planning for RAG — Budget the Retrieved Docs
RAG context is a budget with line items: retrieved documents, the question, and the answer all share one window. Plan how many chunks actually fit.
Resource
Classify Support Tickets with AI
Billing, Technical, Account, How-To, Feature Request — ticket triage with definitions that decide the borderline cases for the model.
Resource
Customer Support Agent
Configure AI to answer support questions within your actual policy boundaries — not generic best-guess answers.
Resource
Categorize Customer Feedback with AI
Praise, Complaint, Feature Request, Bug Report, Question — multi-label feedback categorization where one message can carry three signals.
Resource
Validation Test Prompt
Required fields one at a time, invalid formats, business rules at their exact boundaries — validation tested the way users break it.
Resource
Security Code Review Prompt
Twelve security checks — injection, auth, secrets, SSRF, privilege escalation — reviewed the way an attacker would read the code.
Resource
Estimate Cost per 1,000 Calls
A classification call costs a fraction of a cent — until you run a million of them. This prices a small repeated prompt at the scale that actually bills.
Resource
Agent Test Scenario Prompt
Build the test set an agent has to pass — scenarios across the happy path, edges, and adversarial inputs, each paired with the expected behavior to grade against.
Resource
Agent Evaluation Scorecard Prompt
Grade agent output the same way every time — a rubric scoring correctness, grounding, safety, tone, and completeness, with a pass threshold instead of a gut call.
Resource
Hallucination Detection Prompt
Catch the confident invention — check an AI output's claims against its source and flag every statement that isn't supported, with the unsupported span quoted.
Resource
Product Validation Report Prompt
Consolidate the whole validation into one shareable report — the success metric, the evidence, the hit / partial / miss verdict, and the iterate / pivot / scale decision, in a form the team and stakeholders can act on.
Resource
Product Feedback Summary Prompt
Key Findings, Recommendations, Open Risks — collected feedback compressed for the roadmap discussion, with recommendations the source actually supports.
Tools used in this project
Tool
System Prompt Generator
Build structured system prompts from role, tone, constraints, and model target.
Tool
Multi-Step Prompt Builder
Break a big goal into a sequential prompt workflow — each step with an objective, prompt, and expected output.
Tool
Context Window Estimator
Will this fit the model's context window? Token budget planning, range-honest fit verdicts, and model comparison.
Tool
Data Classification Prompt
Build classification prompts that assign labels from a closed set — with label definitions and edge-case rules.
Tool
Extraction Prompt Generator
Build prompts that extract defined fields from unstructured text — emails, invoices, tickets, résumés.
Tool
Test Case Prompt Generator
Build test generation prompts — unit, integration, or E2E — with framework modes and edge-case coverage rules.
Tool
Code Review Prompt Generator
Generate code review prompts for any scope — snippet, file, diff, or full pull request — with focused review rules.
Tool
Token Counter
Estimate how many tokens a prompt is and what it costs — honest ranges across GPT, Claude, and Gemini, with per-call and per-1,000-call pricing.
Tool
Markdown Output Builder
Build prompts that produce documents in a fixed structure — headings, sections, and tables.
Guides for this project
Guide
How to Write Test Scenarios for an AI Agent
Turn an agent's instructions, allowed tools and business rules into scenarios you can actually run — a fixed situation, an expected behavior you can observe, and a pass or fail that two people would agree on.
Recommended next project
Project
Build a Customer Support System with AI
The full path to a support operation, not just a bot — stand up the knowledge base, route the tickets, add the AI agent, integrate your stack, close the feedback loop, evaluate, and deploy.
Related projects
Project
Build an API Backend with AI
The full path to a backend you can put clients on — define the requirements, design the architecture, API contract, data model, and access control, then build it reviewed, tested, secured, and shipped.
Project
Build a RAG System with AI
The full path to a retrieval system that returns grounded answers — understand the corpus, chunk and ground it, extract and classify the metadata, then evaluate that retrieval actually works.