The archive · Developer & Business Tools · Product decision · 2023–2026
HumanLayer bets AI agents need humans at the riskiest tool calls
YC F24 startup from an ex-Replicated founder turns a SQL-agent scare into an approval SDK; 6.6K GitHub stars by 2025-10-30.
HumanLayer
What the business is
HumanLayer (San Francisco, founded 2023) sells human-in-the-loop infrastructure for AI agents: Python and TypeScript clients plus an API that let software pause and ask a human for feedback, input or approval, across Slack, SMS, email and other channels, with any LLM or agent framework. It has since expanded into CodeLayer, a multiplayer coding-agent IDE and cloud for running parallel Claude sessions across a team.
How it started
Founder Dexter Horthy spent seven years at Replicated helping developer teams launch Kubernetes-native products. Before HumanLayer he was building an autonomous agent to clean SQL data warehouses — an MVP that would coordinate with humans in Slack and drop tables not queried in over 90 days. He was not comfortable letting an AI application run raw SQL unsupervised, so he wired in human approval steps and realized the pattern was general: the most useful functions an agent can perform are also the most dangerous.
What happened
In August 2024 Horthy built an MVP, demoed it to startups across San Francisco and landed his first paying customers, which got HumanLayer into Y Combinator's Fall 2024 batch. The launch positioned human oversight as the unlock for shipping autonomous agents safely, and the repo grew to 6.6K GitHub stars by late October 2025. Through Q1 2025 conversations with dozens of engineering teams showed everyone was rolling their own agent architecture, so Horthy published '12-Factor Agents: Patterns of Reliable LLM Applications' in April 2025, which became a reference for context engineering. By Q4 2025 the team had launched CodeLayer — parallel Claude agent sessions across worktrees and remote cloud workers — arguing that once multiple people ship AI-written code, coordination and review, not generation, are the bottleneck.
No ending yet — it is still running.
Background
HumanLayer (San Francisco, founded 2023) started as human-in-the-loop infrastructure for AI agents: an API and SDK that let software pause and ask a human for feedback, input or approval, delivered through Slack, SMS, email and other channels and usable with any LLM or agent framework.
The founding wager grew out of a SQL-cleanup agent. Founder Dexter Horthy, previously seven years at Replicated, built an MVP that would drop warehouse tables not queried in over 90 days — then refused to let it run raw SQL unsupervised and wired in human approval steps. He concluded that the most useful functions an agent can perform are also the riskiest, so dependable agents need oversight at the tool layer.
An August 2024 MVP brought first paying customers and a place in Y Combinator's Fall 2024 batch. The humanlayer repo reached 6.6K GitHub stars by 2025-10-30 per ROSS Index, with $0.5M raised from YC, Hawk Hill, Multimodal, Pioneer and Preface; founder Horthy's April 2025 '12-Factor Agents' guide became a reference for context engineering.
Through 2025 the company expanded from approvals to coding agents: CodeLayer runs parallel Claude sessions across worktrees and remote workers, on the argument that once teams ship AI-written code, coordination and review become the bottleneck. Guvi reported several large engineering-team pilots since Q4 2025, and YC's company page now describes HumanLayer as a multiplayer control plane for the software factory.
What has to be true
- The founding moment was empirical: an agent with raw SQL access was genuinely useful and genuinely dangerous, so human approval was the unlock, not an afterthought.
- Human review lets teams ship good-enough agents into production and iterate, instead of spending months on evals and prompt tuning before anything runs.
- A framework- and model-agnostic tool-calling integration meant one SDK covered any agent stack, removing the adoption question for prospective customers.
- The pivot to coding agents followed the customers: when teams scaled AI-written code, review and coordination became the bottleneck, and HumanLayer's approval DNA applied directly.
What can be applied
Sell the control layer first: an approval mechanism at the tool layer lets teams ship agents before they are perfect, and that trust becomes the platform for what agents do next.
Aftermath
As of September 5, 2026, HumanLayer is positioned as a multiplayer coding-agent IDE and cloud bringing agent sessions, plan artifacts and code diffs together. Guvi reports that CodeLayer runs parallel Claude sessions across worktrees and remote cloud workers, with several large engineering-team pilots since Q4 2025. The human-in-the-loop API still anchors the origin story, but the stated direction is the control plane for agent-driven software engineering — from guarding risky actions to orchestrating how teams ship AI-written code.
Sources
- Top trending open-source startups in Q4 2025 — ROSS Index
- Launch YC: HumanLayer — Human-in-the-loop for AI Agents and beyond
- HumanLayer: The multiplayer control plane for your software factory
- How Three YC Startups Built Their Companies with Claude Code
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card