Process Intake Framework — Brothers Automate
Brothers Automate × Gumloop

Process Intake
Framework

Every time you get a process to automate, run it through this. Ten sections. One complete build spec.

10
Sections
3
Skill Blocks Per Agent
5
I/O Dimensions
1
Exportable JSON Spec
How It Works

Work Through It In Order

Don't skip ahead. The workflow map has to come before agent specs. Agent specs have to come before skills. Each section builds on the last.

01–03

Understand the Process

Name it, define the goal, map inputs and outputs, identify how it starts.

04

Map the Workflow

Define the shape, every node, every connection, where agents sit. Do this before touching agents.

05

Spec Every Agent

I/O, tools, confirmation rules, and a full skill set per agent. One agent at a time.

06–09

Complete the Stack

Data sources, connectors, human gates, error cases. Don't skip these — they're where automations break.

10

QA Checklist

Run the checklist before you hand off to a builder or open Gumloop. Catch the gaps now.

The Framework

Ten Sections.
Zero Gaps.

Click any section to expand it. Each one covers what to capture and why it matters.

01
Process Overview
Name it. Define the goal. Set the context.
Start Here

Three fields. Simple but essential — every other section references back to this. If you can't write a clear goal in two sentences, the process isn't ready to automate yet.

Process Name
What is this called? e.g. "Lead Qualification Pipeline", "AP Invoice Processing", "Employee Onboarding"
Goal / Purpose
What does this process achieve when it works perfectly? One or two sentences. Be specific — "save time" is not a goal.
Notes & Context
Constraints, existing tools in use, team involved, edge cases you already know about, anything that shapes how this gets built.
02
Input / Output
Five dimensions. Both sides. Full handoff spec.
5 Dimensions

Map all five dimensions for both the input and the output. Matching them forces you to fully spec what comes in and what leaves — which is exactly where most automations break. If the output isn't defined, the automation has nowhere to go.

Dimension Input — What to Define Output — What to Define
Location Where does the data come from? Where does it land when done?
Method How does it arrive? (email, webhook, form, API) How does it get delivered? (API write, Slack message, email)
Timing When does it arrive? (real-time, scheduled, triggered) When is it expected? (immediately, after approval, daily)
Content What information is in it? What information comes out?
Format What structure? (JSON, CSV, plain text, form fields) What structure? (JSON, Slack message, .csv, record update)
Rule: These same five dimensions apply to every agent's input/output in Section 05. Fill the process-level ones first, then mirror the pattern per agent.
03
Triggers
How does the process start? Be specific.
Entry Points

Select all that apply. Then for each one, add the specifics. "Webhook" is not enough — you need to know which system, what event, and what the payload looks like.

Webhook which system?
Schedule / Cron what frequency?
Manual Run who triggers it?
Form Submission which form?
Email which inbox / rule?
New CRM Record which object / stage?
File Upload which folder / type?
Agent Handoff from which agent?
Rule: A process can have multiple entry points. Define each one separately — different triggers may carry different payloads and need different handling at the start of the workflow.
04
Workflow Map
Shape → nodes → connections → agent placement.
Do This First

Map the workflow before you spec any agents. You need to see the shape of the process, where every node sits, and where agents are placed before you can write a meaningful agent spec. Skipping this creates agents that don't know what they're receiving or passing on.

4A — Process Shape

Linear

Steps run one after another. Every node feeds the next. No branches.

Branching

Decision points split the path. Different conditions lead to different next steps.

Parallel

Multiple tracks run at the same time, then reconverge. Faster than sequential.

Loop

Steps repeat until a condition is met. Common in polling or retry patterns.

4B — Node Types
triggerWhat starts the workflow — a webhook, schedule, form submit, or agent handoff
agentAn AI agent performing a task — always link to the named agent in Section 05
humanA human action or approval checkpoint — define in Section 08
decisionA branch point — if/else logic that splits the path. Label the outgoing edges.
actionA direct tool action — send email, update record, write to Sheet
endThe workflow completes — one per branch, not just one per workflow
4C — Workflow Nodes Format
// List every node. Link agent nodes to a named agent. Node 1: [trigger] — HubSpot stage change: "Ready to Quote" Node 2: [agent] — Deal Validator → Agent: Deal Validator Node 3: [decision] — Validation passed? Node 4: [agent] — Quote Builder → Agent: Quote Builder Node 5: [human] — Approval Gate → Gate: Manager Approval Node 6: [agent] — Quote Sender → Agent: Quote Sender Node 7: [end] — Quote sent, HubSpot updated
4D — Connections Format
Node 1 → Node 2 Node 2 → Node 3 Node 3 → Node 4 (if PASS) Node 3 → Node 7 (if FAIL — alert rep, stop) Node 4 → Node 5 Node 5 → Node 6 (if approved) Node 5 → Node 7 (if rejected) Node 6 → Node 7
4E — Agent Placement
AgentPositionReceives fromPasses to
Deal Validator trigger HubSpot webhook Quote Builder (on PASS)
Quote Builder middle Deal Validator output Approval Gate
Quote Sender final Human approval signal Gmail + HubSpot
05
Agents
One block per agent. Repeat for every agent in the workflow.
Core Section
Each agent has four areas to spec: Overview, Input/Output, Tools, and Skills. Use the tabs below as your template. Repeat the full block for every agent. Never combine two jobs into one agent.
Agent Name
Kebab-style is fine: lead-enrichment-agent, quote-builder, approval-router. The name tells you exactly what it does.
Purpose / Goal
Single sentence. What does this agent do and why does it exist? If you can't write this in one sentence, the scope is too broad — split it into two agents.
Position in Workflow
triggerFirst agent in the chain — receives the raw trigger payload directly
middleReceives output from another agent — transforms, enriches, routes, or decides
finalLast step — performs the terminal action (send, write, publish, archive)

Each agent has its own input and output — not just the process-level ones. An agent in the middle of a workflow receives data from the previous agent, not from the original trigger. Define both sides with the same five dimensions.

DimensionAgent InputAgent Output
LocationWhere does this agent receive data from?Where does it send results?
MethodHow does data arrive? (prior agent output, webhook, poll)How does it deliver? (write to Sheet, call next agent, Slack message)
TimingWhen is it triggered? (immediately, on event, on schedule)When does it complete? (synchronous, async, within N seconds)
ContentWhat specific fields does it receive?What fields does it produce?
FormatJSON object? CSV row? Webhook payload?Structured JSON? Record update? Plain text?

List every integration with specific actions — not just tool names. Explicitly disable actions the agent doesn't need. Fewer available actions = more predictable, safer behaviour.

Example — Quote Builder Agent
ENABLEGoogle Drive → read_filereads quote template
ENABLEGoogle Drive → create_filesaves generated PDF
DISABLEGoogle Drive → delete_filenot needed, prevent accidents
ENABLEGoogle Sheets → read_rowreads pricing table
DISABLEGmail → send_emailthis agent builds only — never sends

⚠ Confirmation Rules

  • Never invent pricing — halt if line item not in pricing table
  • Never modify the template structure — populate only
  • Never send the PDF — build and save only
  • Stop immediately if required deal field is missing
Schedule
Event-driven (called by prior agent) / Scheduled (cron time) / Manual / Agent handoff. Specify exactly when this agent activates.

Skills are how an agent knows how to do something specific. One skill = one job. Instructions live inside the skill. Data lives externally. The agent sees skill names and descriptions first — only loads the full skill when it decides it's relevant. Vague descriptions = the skill never gets found.

Skill Structure — 3 Blocks
1

Instructions

  • Step-by-step process
  • Decision logic (IF/THEN)
  • Rules & constraints
2

External Refs

  • Tool / Platform
  • Location (page, table, folder)
  • Variable / ID for access
  • How operator updates it
3

Examples / Resources / Scripts

  • Sample input → output pairs
  • Reference docs & playbooks
  • Reusable code snippets
Block 1 — Instructions Format
// Step-by-step process 1. Check if lead has company email domain 2. Search enrichment tool for company data 3. Score lead 1–10 based on ICP criteria (read from Notion) 4. IF score ≥ 7 → mark as qualified, continue to routing 5. IF score < 7 → log to nurture list, stop flow // Decision logic IF company email missing → stop, flag for human IF company size > 500 → add "enterprise" tag IF score is ambiguous → default to human review // Rules & constraints ⚠ Never fabricate missing company data ⚠ Always log score reasoning alongside the score ⚠ Stop immediately if required field is null
Block 2 — External Refs Format
FieldWhat to Define
Toole.g. Notion, Airtable, Google Drive, HubSpot
Locatione.g. Notion: /Sales/ICP Criteria, Airtable: Leads table, Drive: /Templates/Quotes/
Variable / IDe.g. {{notion_page_id}}, {{airtable_base_id}} — how the agent accesses it at runtime
Update methode.g. Edit Notion page directly — agent reads live on every run. No code required.
Rule: If content needs to be edited by a non-developer, it lives externally. The skill contains a pointer, not the data itself. This keeps your knowledge base editable without touching the agent.
Block 3 — Examples / Resources / Scripts Format
// Examples — sample input → output pairs Input: { name: "Jane Smith", email: "jane@acme.com", company: "Acme Corp" } Output: { score: 8, qualified: true, tags: ["mid-market"], reason: "200 employees, ICP match" } Input: { name: "Bob", email: "bob@gmail.com", company: "" } Output: { score: 0, qualified: false, flag: "missing_company" } // Resources - ICP criteria doc → Notion: /Sales/ICP Definition - Scoring rubric → Airtable: Lead Scoring table - Email templates → Drive: /Templates/Outreach/ // Scripts — reusable logic const score = (companySize > 200 ? 4 : 2) + (hasWorkEmail ? 2 : 0) + (industryMatch ? 3 : 0) + (budgetSignal ? 1 : 0)
06
Data Sources
Map every source the workflow reads from or writes to.
Skills Point Here

Data lives here, not inside agents. Skills contain pointers to this section. Every time an agent needs information, it should come from a named, external data source that a human can update without touching the agent.

Name
What is this source called? e.g. Lead Database, Pricing Table, Vendor Master List
Tool
Where does it live? e.g. Airtable, Notion, Google Sheets, HubSpot, Supabase
Reads
What tables, pages, fields, or records does the agent read? Be specific — not just "customer data".
Writes
What does the agent create, update, or append? Which specific fields? Which tab or table?
Rule: Repeat one block per data source. If a source appears in multiple agents' skills, it only needs one entry here — skills reference the same source.
07
Tools & Connectors
Global integrations. Enable only what's needed.
Disable Explicitly

Global tools used across the workflow — distinct from per-agent tools in Section 05. For each integration, specify which actions are enabled and which are explicitly disabled. Agents that can do too many things are unpredictable agents.

ENABLEGmail → create_draftused by Outreach Agent
ENABLEGmail → send_emailonly after human approval
DISABLEGmail → delete_threadnot needed — prevent accidents
ENABLEHubSpot → read_contactall agents read contact data
ENABLEHubSpot → update_contactconfirm before bulk updates
DISABLEHubSpot → delete_contactpermanent — never auto-delete
Rule: Always explicitly disable destructive actions even if you don't think the agent would use them. The constraint is the point.
08
Human Gates
Intentional approval checkpoints. Not error handlers.
Designed Pauses

These are intentional pauses in the workflow where a human must act before anything continues. They exist by design — not because something went wrong. Define each one specifically, including what happens if the human doesn't respond.

Example Gate — Quote Approval

What triggers it
Quote Builder Agent has generated and saved a PDF. Approval Router sends a Slack interactive card to the deal owner's manager.
What human does
Review the PDF via Drive link → click Approve or Reject in Slack. On Reject: add a reason — the thread is opened to the rep.
Fallback
No response in 4 hrs → reminder sent. No response in 8 hrs → escalate to RevOps manager. Reject without reason → Slack prompts for one before logging.
Rule: Define the fallback. A gate without a fallback is a workflow that silently stalls. The fallback is what makes the gate safe to rely on.
09
Error Handling
Map failure scenarios before they happen.
Pre-Mortem

Think through what breaks before you build. For every failure scenario: what the agent does first, what it does if that fails, and when a human takes over. Agents without error handling create silent failures — the worst kind.

Enrichment API timeout
Route on form data only. Flag "Enrichment unavailable" in Slack card. Rep verifies manually — flow not blocked.
⚠ No escalation — graceful degradation
Pricing table lookup fails
Hard stop. Never guess pricing. Alert RevOps with deal ID and error details. Wait for manual intervention.
⚠ Escalate immediately to RevOps
Required field is null
Stop flow. Write error to tracker. Slack alert to workflow owner with field name and record ID. Do not continue with defaults.
⚠ Escalate — human must resolve
Duplicate detected
Hard stop. Alert AP manager with both records. Do not process either until human confirms which is valid. Log both to Duplicates tab.
⚠ Escalate immediately
Rule: Every error case needs a handler and an escalation decision. If in doubt, stop and alert — a paused workflow is recoverable. A wrong action may not be.
Section 10

Export Checklist

Run this before you hand off to a builder or open Gumloop. Every unchecked box is a gap that will cost you time mid-build.

Process

Process name and goal are clear and specific
Input/output fully mapped — all 5 dimensions, both sides
Entry point(s) defined with specifics, not just type

Workflow

Shape identified (linear / branching / parallel / loop)
Every node listed with type and label
All connections mapped, branch labels added
Agent placement table complete

Agents

Every agent has a clear single purpose (one sentence)
Every agent has I/O fully mapped with 5 dimensions
Tools listed with specific enabled / disabled actions
Confirmation rules defined for each agent

Skills

Every skill has all 3 blocks filled
Skill descriptions are specific enough to trigger reliably
External refs point to real, accessible locations
Examples cover both success and failure paths

Data & Tools

Every data source mapped — reads and writes
All reference data lives externally (not inside skills)
Global connectors listed with enable / disable actions

Gates & Errors

Every human gate defined with fallback
Main failure scenarios mapped with handlers
Escalation path defined for each error case
Destructive actions explicitly disabled everywhere
Free Resource

AI Automation: The Business Owner's Field Guide

10 key insights, core concepts, real workflow examples, and the right tools for automating your service business. Written for operators, not engineers.

  • What to automate first (and what not to)
  • How lead funnels actually work under the hood
  • The exact tool stack we use for clients
  • Mindset shifts that save you from overbuilding

No spam. We send useful stuff only.

Field Guide

AI Automation
for Business Operators

The technology to build a digital assembly line for your business already exists. This guide explains what it is, how it works, and what you actually need to know to use it.

The core idea: Define your inputs and outputs clearly. Let the machine handle everything in between. You don't need to understand every technical detail -- you need to understand your own operations.

What Business Owners Need to Know

Tap each to expand

The real value isn't saving clicks. It's offloading the mental load of evaluating options, routing information, and following up consistently. Every time you manually run a process, your brain loads every possible path before choosing one. That energy compounds into exhaustion. Automation does the evaluation for you -- because you already did the thinking when you built the system.
Automation doesn't fix a broken or undefined workflow. If you can't explain the steps manually, a system can't run them for you. Start by mapping what you already do. If you can walk through it step by step, with clear branches and decisions, it can be built and offloaded.
You don't need to understand what happens in between -- that's the machine's job. But you need to be specific: What data enters the system? What result do you want on the other end? Don't ask for 30 reports you won't read. AI can process everything; the constraint is knowing what you actually need.
A weekly email summarizing new leads in your CRM. A form submission that automatically adds a contact and sends a personalized follow-up. These aren't flashy, but they run every day without you. Small systems compound into large amounts of reclaimed time and mental energy over a year.
You can collect a few answers from a prospect, have AI research them, and automatically send a response tailored to their specific situation. What used to require a dedicated person can now run on its own. The result feels personal to the recipient -- because it is, based on what they told you.
If you're an expert in your field, you can turn that knowledge into an automated funnel. Prospects answer a few questions, AI matches their answers to your best content or recommendations, and you capture their information in the process. You're using AI to automate the selection -- not replace your expertise.
If something always happens the same way, use a workflow. If it requires interpreting context or choosing between options -- like triaging a new lead or responding to a varied inquiry -- that's where an AI agent adds value. Knowing which tool fits which task saves you from building the wrong thing.
CRMs, email platforms, forms, databases, research tools, image generators -- almost anything can be connected to anything else today. The tools exist. The hard part is knowing what you want connected, why, and being specific enough about it that a system can be built to do it reliably.
Build the system, find the gaps, fix them. The goal is a machine that runs cleanly -- not a perfect machine on day one. Every iteration makes it more reliable. Error handling is part of the build, not a sign that something went wrong. Expect to refine it.
Even when a task only takes one path, your brain loads every possible option before ruling them out. A 100-branch process might only ever use one branch -- but you consider 50 before choosing. Multiply that cognitive load across a full work day and it's significant. Automation doesn't just save time. It preserves focus for things that actually need your judgment.

Core Concepts

The building blocks, in plain language

Data Layer

API

A precise, predefined connection between two software systems. You specify exactly what call you're making -- get this data, post this record. Because they're explicit, they're reliable and predictable.

Think of it as: a specific form you fill out to make a specific request. Same form every time, same result every time.

Intelligence Layer

MCP

Model Context Protocol -- what AI agents use to interact with connected tools natively. Instead of one specific call, it opens a range of possible actions. The agent decides which action fits the situation.

Think of it as: giving an employee full access to a system and trusting them to figure out the right action, rather than scripting every click.

Trigger Layer

Webhook

A push notification between platforms -- when something happens somewhere, data is immediately sent somewhere else as a JSON payload. The entry point for most automations.

Think of it as: a form submission that automatically fires a signal to your systems the moment someone hits submit -- no manual checking required.

Process Layer

Workflow

A defined, repeatable sequence. Trigger, then Action, then Action, then Output. Same path every time. Best for structured, predictable processes that don't require interpretation.

Think of it as: a checklist that runs itself. Every step is predetermined. No judgment needed.

Intelligence Layer

AI Agent

An LLM with access to tools and the ability to make decisions. It can interpret varied inputs, choose the right action from its available options, and execute across connected platforms.

Think of it as: a smart employee who has access to all your systems and can figure out what to do based on what they're given -- without needing step-by-step instructions every time.

Language Layer

LLM

Large Language Model -- the AI brain (like Claude, GPT). Exceptional at processing, interpreting, formatting, and generating text. The reasoning engine behind agents and many workflow steps.

Think of it as: the smartest intern you've ever had -- can process any information, draft anything, research anything, but needs direction on what matters to you.

How It Actually Works

A real example: form submission to personalized outreach

01
Someone fills out your form

A prospect submits a contact or inquiry form on your site. This is the trigger -- the event that starts the whole chain.

02
Webhook fires to your automation platform

The form submission immediately sends a data payload -- name, email, answers -- to a tool like Gumloop or Make. This is your entry point.

JSON payload received: {name: "Sarah Chen", email: "sarah@...", interest: "accounting automation"}
03
Data is parsed and routes split

The platform extracts the relevant fields. From here, you can run parallel tracks -- one route adds them to your CRM, another begins the outreach flow.

04
Option A: Simple personalized email

Name and email go to an email tool (Resend, Gmail). A template pulls in their first name and the specific interest they mentioned. Sent within seconds of their submission.

"Hi Sarah, thanks for your interest in accounting automation. Here's what we do for firms like yours..."
05
Option B: AI-researched, fully tailored outreach

Name, email, and company get passed to an AI agent. Using tools like Perplexity or Exa via MCP, it researches them, then generates a response specific to their situation before sending.

Agent finds Sarah's firm handles 40+ clients, specializes in e-commerce. Email references this specifically.
06
You receive a summary, not the work

A simple report lands in your inbox. New lead added. Outreach sent. Anything that needs your judgment is flagged. Everything else ran without you.

The Tool Stack

What connects to what

Workflow BuilderGumloop

Visual workflow builder and agent platform. Good for connecting systems without deep coding knowledge.

Database / CRMAirtable

Flexible database that works as a CRM. Easy to connect to automations via API.

Email SendingResend

Programmatic email sending via API. Clean, reliable for automated outreach and notifications.

Research ToolPerplexity / Exa

AI-powered search and research. Agents use these via MCP to research leads or gather market data.

Web ScrapingFirecrawl

Scrapes websites at scale. Useful for competitive research, content gap analysis, SEO data.

AI BuilderClaude Code

LLM-powered coding tool for building custom internal software. Good for one-off tools tailored to your exact process.

Landing PagesFramer

Fast, design-quality landing page builder. Quick to spin up funnels and lead capture pages.

Image GenerationGoogle ImageFX

AI image generation for ad creatives, landing page visuals, and content assets.

WorkspaceNotion

Documentation and knowledge base. Can serve as a lightweight internal tool or client-facing resource.

The Knowledge Funnel

Turning expertise into qualified leads -- click each stage

You have expertise. Prospects want specific information they can't easily find elsewhere. The knowledge funnel connects these two things -- and captures what you need to convert them in the process.

Why they do it: They're getting something specific in return. Not a generic newsletter -- information tailored to their answers. The specificity of the promise is what gets them to fill it out.
You've already done the hard work: building the knowledge base from your expertise, defining what good answers look like. The agent just does the matching -- fast and at scale. It's not replacing your expertise. It's automating the selection.
The personalization isn't superficial. It's based on what they actually told you. People know when they're getting something generic. When the response reflects their specific situation, they notice -- and they're more likely to take the next step.
Their answers tell you what matters to them, what stage they're at, and how to position your offer. Your follow-up can reference this directly. Instead of a cold pitch, you're continuing a conversation they already started.

The Right Mindset

How to think about this before building anything

"Ford took every process of manufacturing a car and systematized it so it ran on its own. He couldn't do that with his accounting. Now you can -- digitally, for the back end of your entire business."
Define your assembly line before you build it. Know every step of your process. The clearer your manual process, the better your automated one will be. Vague in, vague out.
Complexity is fine. Ambiguity is not. Your process can have 100 branches. That's okay. What isn't okay is not knowing which branches exist. A complex but clearly defined process can be automated. An undefined one can't.
Start with what you already do manually. Don't try to automate something you haven't done yet. Pick one process you run regularly, map it out, and build that. Get one system running cleanly before adding another.
Build in error handling from the start. Assume things will break. Add notifications when they do. An automation that fails silently is worse than no automation. Know when your system needs your attention.
The goal is to stop thinking about things that should think for themselves. Every time you save a future version of yourself from having to load a process into working memory, you've created real leverage. That's what this is for.

Looking to offer automation to your clients?

Take the 2-min quiz →