productivity

Gumloop Review: What We Tested in 5 Workflows (2026)

Name: Gumloop
Availability: InStock
Author: AI Demos

No-code agent builder for structured business workflows and approval gates, but free-tier research is unreliable.

Visit Gumloop

No-codeHuman-in-the-loopFree tier available4/5 tasks passed

TL;DR — our verdictUpdated July 2026 · 10 test artifacts

Strong on structured business agents; weak on free-tier research and exact handoff fidelity.

Where it wins

You want to build AI agents using plain-English instructions without writing code.
You need structured business workflows such as lead qualification, routing, or approval-gated email drafts.
You want a genuine human approval gate before a risky action is finalized.

Main limitation

You need dependable, fast live web research on the free tier.

Pricing (verified plans)

Free $0/monthPro $37/monthEnterprise Custom pricing

Strongest test artifacts

Plain-English Agent Configuration →Structured Classification and Routing →Document-Grounded Answering and Drafting →

Our take

Gumloop passed 4 of 5 anchor tasks in testing: lead qualification, leave-policy retrieval plus drafting, customer routing, and multi-turn approval gating all worked without code. The clear weak spot was company research on the free tier, where web search took minutes, returned no sources, and leaked a system prompt.

Clean screen-recording demo of Gumloop's agent builder and tested workflows.

In-Depth Review

Our detailed analysis of Gumloop — features, performance, and real-world testing.

AI Demos Team

Expert Reviewer

Verified Review

Feature-by-Feature Breakdown

Plain-English Agent Configuration

▾

Test Summary

Feature tested: Plain-English Agent Configuration

Result: Passed

Gumloop lets a business user define agents from plain-English instructions without code or API setup. In testing, it was used to specify lead-qualification, leave-policy, customer-routing, company-research, and email-approval agents conversationally.

text

INPUT: Lead Qualification Agent instructions — classify a lead using Budget, Authority, Need, and Timeline; output Lead Score, Qualification Summary, and Recommended Action.

↓→

image

Output artifact for "Plain-English Agent Configuration" test: Structured lead qualification output produced after plain-English agent setup: the agent returned a qualification assessment, reasoning, recommended action, follow-up email draft, and CRM-style note., image-3.png

Structured lead qualification output produced after plain-English agent setup: the agent returned a qualification assessment, reasoning, recommended action, follow-up email draft, and CRM-style note.

text

INPUT: Email Approval Agent instructions — draft an email, ask for YES/NO approval, and only mark it Ready to Send after explicit confirmation.

↓→

image

Output artifact for "Plain-English Agent Configuration" test: Approval-gated workflow output from a plain-English setup: the agent drafted the email, asked for approval, handled YES and NO turns, and kept the conversation in an approval loop., image-17.png

Approval-gated workflow output from a plain-English setup: the agent drafted the email, asked for approval, handled YES and NO turns, and kept the conversation in an approval loop.

Bottom Line

The core no-code setup works well: Gumloop can produce different business agents from plain English instructions without requiring code.

Structured Classification and Routing

▾

Test Summary

Feature tested: Structured Classification and Routing

Result: Partial

Gumloop can turn conversational inputs into reusable business decisions, such as a BANT-style lead verdict for a sales lead and a priority-based routing decision for a billing complaint. The same workflow pattern also supports structured classification outputs for different inbound cases.

text

INPUT: Customer routing test — a customer says they were charged twice, want a refund, and have waited 5 days with angry escalation language.

↓→

image

Output artifact for "Structured Classification and Routing" test: The agent correctly categorized the complaint as a billing issue, set high priority, flagged escalation, and drafted a response., image-10.png

The agent correctly categorized the complaint as a billing issue, set high priority, flagged escalation, and drafted a response.

text

INPUT: Same customer routing test, checked for exact downstream handoff fidelity against the instructed routing map.

↓→

image

Output artifact for "Structured Classification and Routing" test: Annotated failure screenshot showing the routing-map drift: the case was assigned to Billing Support Escalations Team instead of the instructed Billing Team., image-10.png

Annotated failure screenshot showing the routing-map drift: the case was assigned to Billing Support Escalations Team instead of the instructed Billing Team.

text

INPUT: Same customer routing test, checked for production-style case metadata in the structured output.

↓→

image

Output artifact for "Structured Classification and Routing" test: Annotated failure screenshot showing missing operational metadata: no ticket ID, SLA timer, or sentiment score were present in the structured routing output., image-13.png

Annotated failure screenshot showing missing operational metadata: no ticket ID, SLA timer, or sentiment score were present in the structured routing output.

Bottom Line

Structured classification is a strength, but the routing output drifted from the exact map and omitted standard support metadata like ticket ID, SLA timer, and sentiment score.

Document-Grounded Answering and Drafting

▾

Test Summary

Feature tested: Document-Grounded Answering and Drafting

Result: Partial

Gumloop can read an uploaded PDF, answer questions from it, and use that information to draft a workflow output. In the leave-policy test it extracted entitlement and approval rules from the policy document and produced a leave request draft for Arjun Desai.

pdf

FutureSmart-AI-Leave-Policy.pdf

↓→

image

Output artifact for "Document-Grounded Answering and Drafting" test: The agent read the uploaded leave policy PDF, answered the policy question, asked whether the request was full day or half day, and drafted a leave request with employee, dates, reason, and manager fields filled in., image-6.png

The agent read the uploaded leave policy PDF, answered the policy question, asked whether the request was full day or half day, and drafted a leave request with employee, dates, reason, and manager fields filled in.

Bottom Line

Good at policy retrieval plus drafting, but it is not a full HR workflow on the free tier because the PDF is not persistent and the response did not check remaining leave balance.

Human Approval Gates and Revision Control

▾

Test Summary

Feature tested: Human Approval Gates and Revision Control

Result: Partial

Gumloop supports approval checkpoints that pause a workflow before a risky action is finalized. In the email workflow it drafted an email, asked for YES/NO approval, produced a revised version after NO, and remembered prior approval state within the session.

text

INPUT: Email follow-up request for Vikram Singh at TechNova Solutions; draft an email, wait for approval, only mark Ready to Send after YES.

↓→

image

Output artifact for "Human Approval Gates and Revision Control" test: The agent paused for approval, asked for explicit YES/NO confirmation, and only moved the email to a ready-to-send state after approval., image-17.png

The agent paused for approval, asked for explicit YES/NO confirmation, and only moved the email to a ready-to-send state after approval.

text

INPUT: Follow-up session after approval — user replies NO and asks for changes in the same thread.

↓→

image

Output artifact for "Human Approval Gates and Revision Control" test: The approval state persisted within the session, but the screenshot shows no stored audit trail or session history anywhere in the interface., Email approval output.Gumloop 2.png

The approval state persisted within the session, but the screenshot shows no stored audit trail or session history anywhere in the interface.

text

INPUT: Revision comparison request after NO in the same approval session.

↓→

image

Output artifact for "Human Approval Gates and Revision Control" test: The agent produced a new draft after the revision request, but the original draft disappeared and there was no side-by-side comparison., image-20.png

The agent produced a new draft after the revision request, but the original draft disappeared and there was no side-by-side comparison.

Bottom Line

The approval gate itself is a standout strength, but the free tier keeps the workflow session-local with no audit trail or version comparison.

Web Research Automation

▾

Test Summary

Feature tested: Web Research Automation

Result: Failed

Gumloop can launch live web research for a company-research workflow. In the Lenskart test it searched for several minutes and attempted to assemble a research brief from web sources.

text

INPUT: Company research request — company name Lenskart, website lenskart.com.

↓→

image

Output artifact for "Web Research Automation" test: The research flow triggered web search automatically, but it took roughly 6–7 minutes and produced no usable sources., image-14.png

The research flow triggered web search automatically, but it took roughly 6–7 minutes and produced no usable sources.

Bottom Line

This is the clearest failure in the review: the free-tier research flow did not produce any of the required company fields and exposed internal instructions instead.

Pricing & Access

Credit-based plans; the free tier was enough to test the core builder, but the free plan lacked direct integrations and persistent document storage in testing.

TESTED

Free

$0/month

5,000 credits/month, 1 seat, 1 active trigger, 2 concurrent runs, 5 concurrent agent interactions, unlimited agents and flows, forum support — sufficient for testing and prototyping.

Pro

$37/month (20% off with annual billing)

20,000+ credits/month, unlimited seats and teams, 5 concurrent runs, 25 concurrent agent interactions, agent reflections, unified billing, connector policies and guardrails, 1 MCP server hosting.

Enterprise

Custom pricing

Everything in Pro, plus role-based access control, SCIM/SAML, admin dashboard, audit logs, data exports, incognito mode, virtual private cloud, 5 MCP server hosting.

Pricing checked June 2026; verify current plans on the official site.

✓ Use This If

●You want to build AI agents using plain-English instructions without writing code.

●You need structured business workflows such as lead qualification, routing, or approval-gated email drafts.

●You want a genuine human approval gate before a risky action is finalized.

●You are okay re-uploading reference documents at the start of each new conversation on the free tier.

✕ Skip This If

●You need dependable, fast live web research on the free tier.

●You need a persistent knowledge base that survives across conversations.

●You need out-of-the-box CRM, email, HRMS, or ticketing integrations on the free plan.

●You need exact routing labels and strict downstream handoff fidelity.

●You need numerical lead scoring or weighted classification breakdowns.

productivityagent-platformstext

No. In this review, the tested agents were configured with plain-English instructions rather than code or API setup.

Lead qualification, leave-policy retrieval plus drafting, customer routing, and multi-turn approval gating all worked well enough to pass their anchor tasks.

Yes. It correctly read the FutureSmart AI leave policy PDF, answered from it, and used it to draft a leave request.

Not on the tested free tier. The leave-policy PDF had to be attached again at the start of each new conversation.

Yes. The email workflow asked for explicit YES or NO approval, and only moved to a ready state after approval.

Not on the tested free tier. The Lenskart research run took about 6 to 7 minutes, returned no usable sources, and exposed internal instruction text.

In the free-tier tests, no connected CRM, HRMS, email, or ticketing integrations were available, and the outputs stayed as chat text that had to be copied manually.

Mostly, but not always. In the billing routing test it correctly classified and prioritized the issue, but it invented a different team name instead of using the exact routing label provided.

The reviewed free plan was listed at $0/month and included 5,000 credits/month, 1 seat, 1 active trigger, 2 concurrent runs, 5 concurrent agent interactions, unlimited agents and flows, and forum support.