HumanLayer made its name as a human-in-the-loop API: an SDK that let AI agents require human approval before sensitive tool calls and contact humans for answers over Slack and email. In 2026 the product changed direction: humanlayer.dev now sells an AI coding IDE built around a phased development workflow, and the approvals API is no longer on the site. If you came to HumanLayer for human-in-the-loop and not for an IDE, you need a tool that still does that job. For coding agents, that is exactly what Orako does: your agent asks a question, the teammates who own that domain get it where they already work, the first to claim answers, and the reply lands back in the agent.
Key takeaways
- HumanLayer today is an AI coding IDE ($100/user/month, free up to 3 members). Its site no longer offers the human-in-the-loop approvals API it was known for.
- Orako covers the ask-a-human job for coding agents: dispatch by domain, first-to-claim ownership, answers in Slack, Teams, Discord or the dashboard, returned to the agent over MCP.
- Every answer is saved to a team knowledge base with a confidence score, so the same question never has to be asked twice.
- Orako is $12/seat/month with a 14-day free trial. You pay only for the people who answer, not for agents or questions.
What changed at HumanLayer
The original HumanLayer (the one most “human-in-the-loop API” articles still describe) offered approval decorators around tool calls, routing to individuals or teams, escalations and timeouts, and human contact across Slack and email; you brought your own LLM and framework. The current humanlayer.dev instead sells a team IDE where engineers and agents collaborate through six phases (Questions, Research, Design, Structure, Plan, Implement), with pricing at $100/user/month and a free tier capped at 3 members and 200 sessions/month. That is a real product, but it is a different product.
HumanLayer’s original features, mapped to Orako
HumanLayer v1 human-in-the-loop features mapped to Orako equivalents
| Job to be done | HumanLayer (original API) | Orako |
|---|---|---|
| Agent asks a human | contact-human-as-a-tool via SDK (Python/TypeScript) | ask-a-human tool over MCP; works with Claude Code, Cursor, Codex, and any MCP client, no SDK code |
| Who gets asked | Routing to a named individual or team you configure | Dispatch to everyone who owns the domain (auth, payments, infra…); the first to claim owns the 1:1 |
| Where humans answer | Slack and email | Slack, Microsoft Teams, Discord, or the dashboard inbox; reply in the thread, no new tool |
| Escalation | Configurable escalations and timeouts | Automatic: nudge → back to the pool excluding the silent owner → team-channel alert |
| Reusing answers | Learning / auto-approvals from prior decisions | Searchable knowledge base with confidence scores; contested answers are flagged and re-confirmed |
| Pricing | Discontinued (current product: IDE at $100/user/month) | $12/seat/month, 14-day free trial, no credit card; self-hosted Enterprise |
If what you actually want in 2026 is an AI coding IDE with a structured spec-first workflow, today’s HumanLayer is that product and worth evaluating on its own terms. This page is for teams that needed the human-in-the-loop layer it used to be.
Why domain pools beat named-person routing
HumanLayer’s API routed each approval to a person or channel you wired up front. That works until the person is on holiday, in a meeting, or gone. Orako doesn’t guess availability and doesn’t track presence: the question goes to every owner of the domain at once, and whoever is actually there proves it by claiming. Silence triggers escalation on its own, so a question can’t fall through. That is the failure mode that hurts most when an agent is blocked mid-task.
Fair point
When HumanLayer is the right callIf your team wants an AI coding IDE with a structured, spec-first workflow rather than a way to reach humans from any agent, today’s HumanLayer is built for exactly that. Evaluate it on those terms. Orako isn’t an IDE, and it isn’t trying to be one.