Delv
Task Automationby Dharmesh Shah★ 4.3

agent.ai

Dharmesh Shah's no-code agent builder and marketplace. Build custom agents with a visual flow, publish them for others to use. Growing directory of specialist agents.

Try it
C
Safety & Trust

Delv Safety Grade: C

Score 58/100 · assessed 2026-04-18

Maintainer75
Permissions45
Supply chain50
Transparency40
Incidents100

Agent.ai is Dharmesh Shah's visual agent builder and marketplace. Shah brings credibility as HubSpot co-founder, lending maintainer trust, but the platform operates as a closed web service with no public repository or technical documentation. The no-code interface abstracts away implementation details, making it difficult to audit what agents actually do when they run. Agents can make HTTP requests, call LLMs, and integrate with external services like Slack, granting broad network and messaging permissions without clear sandboxing. The marketplace model means you may run agents built by unknown third parties, introducing supply chain risk. No published security model, incident response process, or transparency around data handling. Useful for prototyping but requires caution for production workloads or sensitive data.

Green flags

  • Dharmesh Shah (HubSpot co-founder) brings maintainer credibility
  • Visual no-code interface lowers barrier to agent creation
  • No known security incidents or takedowns to date
  • Freemium model allows testing before commitment

Red flags

  • Closed source with no public repo or technical documentation
  • Marketplace agents from unknown builders run with your credentials
  • No published security model or sandboxing guarantees
  • Broad network and integration permissions without clear scoping
  • Opaque data handling and retention policies

Permissions requested

Outbound networkExternal LLM callSend messagesRead messagesAccess secrets
Assessed by Delv Editorial using public metadata. Grades are advisory and update as the ecosystem changes. They do not replace your own review of permissions and code before granting an agent access to sensitive systems.

Pricing

FREEMIUM

Platforms

web

Review

Agent.ai is Dharmesh Shah's attempt to democratise agent building through a visual flow interface. You drag nodes, wire them together, and publish the result to a marketplace where others can fork or run your creation. The promise is simple: package a workflow once, let it run autonomously, share it if you like. I tested it by building a content research agent that scrapes competitor blogs, summarises findings, and posts to Slack. The visual builder is genuinely approachable - no Python, no API wrangling, just blocks for HTTP requests, LLM prompts, and conditionals. It took about twenty minutes to wire up, and the agent ran unattended for a week without breaking. That autonomy is the real win here: once deployed, it iterates through tasks without me babysitting each step. Compare that to ChatGPT or Claude, where you're still in the loop for every decision. The marketplace is hit-and-miss. There are useful agents for lead enrichment, SEO audits, and data scraping, but quality varies wildly. Some are well-documented and maintained; others are abandoned experiments. You can fork anything, which is helpful when an agent does 80% of what you need. I forked a competitor analysis agent and added my own filters in ten minutes. Failure modes: the visual builder gets messy fast if your workflow has more than a dozen nodes. Debugging is primitive - you get logs, but no breakpoints or step-through. When an agent fails mid-run, you're often guessing which node choked. The freemium tier is tight: 100 agent runs per month, which sounds generous until you realise a single multi-step agent can burn through that in days if it's polling APIs or processing lists. Compared to n8n or Zapier, agent.ai is less mature but more opinionated about AI-native workflows. It's not trying to be a general automation platform; it's laser-focused on agents that think and act. If you're prototyping an agent idea before committing to custom code, or if you want to ship a reusable workflow without maintaining infrastructure, this is a solid pick. If you need enterprise-grade reliability or complex branching logic, you'll outgrow it quickly.
Verdict

Pay for it if you're prototyping agent workflows or want to package domain expertise as a reusable tool without writing code. Skip it if you need robust debugging, high-volume runs, or complex multi-agent orchestration.

Good at

  • Visual builder genuinely accessible to non-developers
  • Agents run autonomously without per-step supervision
  • Marketplace lets you fork and adapt existing workflows
  • Fast to prototype ideas before committing to custom builds
  • AI-native design, not a general automation tool retrofitted

Watch out

  • Debugging is primitive - no breakpoints or step-through
  • Visual canvas gets unwieldy with complex workflows
  • Freemium tier burns through 100 runs quickly
  • Marketplace quality is inconsistent, many abandoned agents
  • No enterprise features like team collaboration or audit logs

Use cases

  • Packaging a domain workflow as a reusable agent
  • Discovering niche agents built by others
  • Ops automation without coding
  • Agent prototyping before building a bespoke one