A Re-evaluation of Workflow-based AI Agent Development Tools

Open original

This report reevaluates no-code/low-code platforms for building enterprise-grade AI agents, focusing on agent authentication, sandboxed code execution, secrets management, lineage tracking, and evaluation features. It scores vendors on their native support for these security and operational capabilities, highlighting gaps in sandboxing, guardrails, and LLM hallucination checks.

  • Only Google, Langflow, Workato, CrewAI, Sim.ai, and Gumloop support full credential-passing for agent-to-third-party auth, and only about half the market offers any sandboxing for LLM-generated code (often outsourced to third parties like E2B).
  • Lineage tracking and secrets management are nearly absent industry-wide—only Google, Workato, and Gumloop score on lineage, and Google, Sim.ai, and Gumloop lead on secrets handling.
  • No vendor excels at both running human-written scripts and safely sandboxing LLM-generated code, despite most platforms marketing themselves to "citizen developers."
  • Whether a tool started as AI-native or pivoted from workflow automation no longer predicts its security posture—actual feature completeness matters more than origin.

Enterprises building AI agents still lean heavily on no-code workflow tools that mix predefined actions with LLM-driven steps. But these platforms rarely deliver truly self-governing agents—they require users to map out flows in advance and plug in code where needed. In this second annual report, Andrew Green zeroes in on what makes an “enterprise-grade agent” secure and auditable. He separates authentication for human users (SSO, MFA via the company’s identity provider) from agent-level auth (API keys, JWTs, mTLS). Only Google, Langflow, Workato, CrewAI, Sim.ai and Gumloop provide full credential-passing when agents call third-party systems. Fewer still—about half—offer any sandbox for running untrusted, LLM-generated code, and most of those outsource sandboxing to third-party services like E2B. Lineage—tracing every agent action back to a real person—and robust secrets management barely exist across the market. Only Google, Workato and Gumloop earned any lineage scores; Google, Sim.ai and Gumloop scored highest on secrets handling. A handful of vendors (notably Google and Gumloop) bundle proxy-based filtering, policy definitions, attribute-based tool access, auth, lineage and secrets management into a coherent package. Others tack on LLM-based “evaluations” to flag PII or factual errors, then route flagged content through secondary agents or regex-style handlers. Those approaches blur the line between rule-driven security and probabilistic LLM checks. On integration and extensibility, most platforms implement both host/client MCP (multi-agent communication protocol) roles, though Google’s proprietary agent-to-agent protocol shows up only at Google, CrewAI, Retool and Sim.ai. Citizen-developer pitch aside, real users write code. Yet no single vendor excels at both executing human-written scripts and safely running LLM-generated code. This year’s additions—filesystem-access metrics and an explicit sandbox requirement—highlight emerging gaps. Finally, the report notes that whether a tool was born AI-native or pivoted from workflow automation no longer dictates its security posture; feature sets matter more than pedigree.

Ask about this article