withHuman
Product Hunt Chris Battarbee
withHuman sits between AI agents and the tools they want to use. Simple rules handle the easy calls; risky ones get sent to a person.
Based on reporting by Product Hunt, Chris Battarbee — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
withHuman is trying to solve a problem that gets bigger the moment you let agents do real work: they can move fast, and they can also do the wrong thing fast. The product puts an approval layer in front of those tool calls, intercepting them before anything runs.
The idea is blunt and practical. Routine requests can be approved or denied automatically by rules, while the calls that look risky are routed to the right person on the team. That means a company can set one policy and apply it across the agents it actually uses, instead of managing a separate mess for each system.
The pitch leans hard on breadth. withHuman says it works across Claude Code, Codex, and agents built with LangGraph and Pydantic AI, among others, through an open protocol called AAP. It’s open source, and there’s also a hosted version for teams that don’t want to run everything themselves.
That mix tells you exactly where the product is aimed: not at making agents more clever, but at making them safe enough to trust. There’s even a launch offer — code PRODUCTHUNT gets three months of Scale free, valid until Friday. Nice little reminder that governance tools now need launch campaigns too.
My take — AI-written commentary, not fact-checked reporting
This is the boring part of AI that actually matters, which is usually how you know it’s the right part. Agents without approval layers are just confident interns with API keys. The industry keeps chasing autonomy; the smart money keeps building brakes and paperwork.
Read more about this at: Product Hunt