Phi Consulting
Talk to Phi

AI-Native Internal Tool Build

GTM teams use AI agents to spec, build, and iterate internal applications that close workflow gaps. Works for routing, qualification, reporting, and enrichment tools without traditional engineering.

OutcomeCustom internal tools that automate GTM workflow gaps ship in days instead of quarters, removing bottlenecks that slow revenue capture.

AdvancedTime to value: 1 to 2 weeks5 tools

How it flows

How it flows.

Source
GTM Problem Definition
Process
Workflow Gap
Process
Gap Specification
Process
Input & Output Spec
Process
App Specification
Process
Features & Flow
Process
Roles & Permissions
Process
UI & Data Logic
Process
Errors & Validation
Store
Reference Material
Source
Figma Designs
Source
Screenshots
Source
Documents
Agent
Prompt Generation
Agent
MVP Generation
Store
GitHub Repository
Process
Local Environment
Agent
AI-Powered Iteration
Output
Internal GTM App
Drag to pan · scroll to zoom

    Build it, step by step

    Build it, step by step.

    1. 01

      Map the workflow gap

      Identify the exact GTM problem the app needs to solve and who will use it. Name the data inputs, expected outputs, and the gap in your current process.

      • Name the GTM problem: routing, qualification, reporting, enrichment, or another gap
      • Specify the workflow gap this app will close
      • Define all data inputs the app will consume
      • Define expected outputs and their format
      • Identify user roles and their specific use cases

    2. 02

      Write the full app spec

      Document every feature, workflow, role, permission, validation rule, and UI expectation before you build anything. Completeness here determines quality later.

      • List all features and their workflows
      • Define user roles and permission levels
      • Specify UI behavior and data flow for each screen
      • Document validation rules and edge case handling
      • Set theme, layout, and UX expectations

    3. 03

      Gather reference material

      Collect visual and functional context that shows what you want. Screenshots, Figma files, and written notes all reduce ambiguity.

      • Add Figma designs if available
      • Capture screenshots of similar tools or flows
      • Include written notes or process docs
      • Organize all files for easy upload

    4. 04

      Generate the build prompt

      Use Claude to convert your spec and reference material into one complete, structured prompt that includes pages, data models, logic, integrations, and UI instructions.

      • Share all app requirements with Claude
      • Upload reference material and context
      • Ask Claude to generate a single structured prompt
      • Prompt must include pages and components
      • Prompt must define data models and schemas
      • Prompt must specify logic rules and integrations
      • Prompt must include UI and layout instructions

    5. 05

      Validate the prompt

      Review the generated prompt for completeness. Every feature should be described. No instruction should ask the tool to assume or decide on its own.

      • Confirm every feature is clearly described
      • Check for missing flows or logic gaps
      • Remove any instructions that ask the tool to decide or assume
      • Ensure all edge cases are covered

    6. 06

      Generate the MVP

      Paste the validated prompt into Lovable and upload all reference files. Each request consumes credits, so aim for a clean first pass.

      • Paste the validated prompt into Lovable
      • Upload all reference files and images
      • Attach Figma designs if available
      • Avoid unnecessary iterations to conserve credits
      • Review the generated MVP for accuracy

    7. 07

      Set up local environment

      Connect the Lovable project to GitHub, clone the repository, and open it in VS Code or Cursor so you can iterate locally.

      • Connect Lovable project to GitHub
      • Generate repository from Lovable
      • Copy the Git repository URL
      • Clone the repository locally using Git
      • Open project folder in VS Code or Cursor

    8. 08

      Iterate using AI agents

      Use Cursor or Claude Code to request logic changes, feature additions, UI improvements, and refactoring. Work in your local environment and push changes to GitHub.

      • Open Cursor and use built-in chat for changes
      • Or use Claude Code in VS Code terminal
      • Request logic changes and bug fixes
      • Add new features or modify existing ones
      • Improve UI and refactor code as needed
      • Push changes to GitHub after each iteration

    9. 09

      Test and finalize

      Test core workflows, confirm data accuracy, and ensure the app is usable by the intended teams. Ship it when it solves the problem.

      • Test all core workflows end to end
      • Confirm data inputs and outputs are accurate
      • Validate edge case handling
      • Ensure the app is usable by internal teams
      • Deploy and document usage instructions

    The stack

    The stack.

    Related workflows

    Related workflows.

    Revenue WorkflowVisitor De-AnonymizationCaptures anonymous website traffic, unmasks company and contact data, scores fit, and triggers tiered outbound motion based on intent strength.
    Revenue WorkflowTarget Account Build and ScorePull closed deal data from your CRM, backtest to define your real ICP, build and score a broad TAM, then orchestrate outbound and signal tracking that pushes qualified accounts back into pipeline.
    Revenue WorkflowSignal ActivationThis workflow ingests intent signals from CRM, product, review sites, partner networks, technographic feeds, and social channels. It cleans, scores, and routes them into your CRM and activation tools so reps act on qualified opportunities fast.

    Talk to us: Want this workflow run for you?

    We wire it into your stack, then run it.

    Talk to Phi