Company
Humans decide. Machines operate.
Building and operating a business should not require spending your life operating software: humans provide intent, judgment and authorization, agents handle reasoning and technical operation, and Stackap handles execution.
The position
Humans should decide. Machines should operate. Building and operating a business should not require spending your life operating software. People should provide intent, judgment and authorization. Agents should handle reasoning, coordination and technical operation. Stackap should handle execution.
That means we do not send people hunting through settings when a machine can make the change. We do not use people as couriers between systems. We do not ask people to diagnose information an authorized agent can inspect itself.
When a person is needed
When something genuinely requires human judgment, whether money, identity, legal consent, security, ownership or risk, we bring that exact decision to the human. Not a general permission to act, and not a screen full of settings: the one decision, in front of the person who owns it.
The goal is not to remove humans from the business. It is to remove the friction between deciding what you want and getting it done.
Three roles
The human provides intent, judgment and authorization. The agent provides reasoning, planning and operation. Stackap provides execution: the infrastructure and the business capabilities the agent needs, with the guardrails that make it safe to let an agent use them. Each role does what it is best at, and none is asked to do another's job.
Four principles
- Don't send the human to a setting.
- If an operation can be a command, it is a command. A step that only exists as a switch in a web page is a step an agent cannot take.
- Don't use the human as a courier between machines.
- Copying a key from one dashboard into another is not judgment. Stackap keeps the pieces in one place so nothing has to be carried.
- Don't make the human diagnose something the agent can inspect.
- Build logs, runtime logs, health and metrics are readable by the same tools that deploy, so a failure can be understood without a person describing a screen.
- If the agent has to guess, the documentation has a bug.
- Documentation is part of the product. Gaps in it are defects, filed and fixed like any other.
Where people stay in the loop
Deploying to production, pointing a domain, rolling back and anything destructive remain a person's call. The aim is for those to be the only moments that need a person at all. Everything else is built to be done for you, and to be done the same way every time.
This is built into the platform, not left to good intentions. An agent holding an approval-mode token that tries a risky action, today deleting a project, is stopped with an approval_required answer. A link goes to the account owner by a separate channel, never through the API. The owner sees exactly that one action and approves or denies it. An approval covers only that exact request, works once, and expires after an hour, so an agent can neither reuse it nor nag for it.
Stackap is in early access. Tell us what you run and we will reply with a straight answer about whether it fits.
Ask for an invitationLast updated .