Make.com AI Agents: Build Autonomous AI Workflows Visually

AI Agents in Make.com

An AI agent in Make.com is a Scenario that:

1. Receives a goal or task (via trigger)
2. Gathers information autonomously (search, API calls)
3. Reasons about what to do (Claude/OpenAI)
4. Takes action based on reasoning (update CRM, send email, create document)
5. Optionally iterates (loops back to gather more info or refine output)

Make.com enables this through chained HTTP modules, Routers for decision branching, Iterators for processing lists, and Data Stores for memory.

Research Agent Pattern

Use case: For each new sales opportunity, automatically research the company and prepare a brief.

Scenario:

1. Trigger: HubSpot deal created
2. HubSpot: Get full contact and company details
3. HTTP (web search): Search for recent company news

1. Iterator: Process each search result
2. HTTP: Fetch content of top 3 results
3. Text Aggregator: Combine all fetched content
4. HTTP (Claude): Synthesise research into a structured brief

1. HubSpot: Save brief as internal note on deal
2. Slack: "Research brief ready for {{company}}: {{HubSpot deal link}}"

Classification and Routing Agent

Use case: Route inbound support tickets to the right team automatically.

Scenario:

1. Webhook: Receive support request
2. HTTP (Claude): Multi-step classification

1. Tools → Parse JSON: Extract all fields
2. Router: Branch on can_auto_resolve

- True path: send suggested_response via email/chat; close ticket - False path: continue to team routing

1. Router (second): Branch on escalate_to + urgency

- critical + engineering → PagerDuty alert + Slack #incidents - billing + high → Billing team Slack + Zendesk P1 - etc.

Document Analysis Agent

Use case: Process incoming contracts, flag issues, extract key terms.

Scenario:

1. Trigger: Gmail new email with PDF attachment
2. Gmail: Download attachment
3. HTTP (PDF parser): Extract text from PDF
4. HTTP (Claude): Analyse contract

1. Parse JSON
2. Router: Based on recommended_action

- accept → email notification with summary - review → create task in Asana with full analysis - reject → email legal team with high-risk clauses flagged

1. Airtable: Log contract to contracts database

Multi-Step Research Agent

Use case: Research a competitor thoroughly — products, pricing, recent updates, customer reviews.

Scenario (with loops):

1. Trigger: Airtable new row (competitor name added)
2. HTTP (Claude): "What are the 5 best sources to research {{competitor}}? Return a JSON array of search queries."
3. Iterator: For each search query (5 iterations)
4. HTTP (SerpAPI): Execute search
5. Iterator: For each result URL
6. HTTP: Fetch page content (filter to text)
7. Text Aggregator: Collect all research
8. HTTP (Claude): "Synthesise this research about {{competitor}} into: product comparison, pricing, strengths vs us, weaknesses vs us, recent announcements, customer sentiment."
9. Notion: Create competitive analysis page
10. Slack: Notify product team

Memory Across Runs

Make.com's Data Stores provide agent memory:

Conversation history for AI chatbot:

1. Webhook: new user message
2. Data Store → Get Record: fetch conversation history by user ID
3. HTTP (Claude): send full history + new message
4. Data Store → Update Record: save updated conversation
5. Return response to user

Decision log for audit: After any AI decision: Data Store → Add Record: log decision, reasoning, timestamp, inputs, outputs.

This creates an audit trail of all AI-made decisions.

Recommended Tools

• Make.com — Visual agent workflow builder
• Claude API — Best reasoning for agent decisions
• OpenAI API — Alternative with native Make module
• SerpAPI — Web search for research agents
• Airtable — Agent memory and result storage
• HubSpot — CRM integration for sales agents