About all-agents
What we build, who builds it, and why we publish what we learn about automating recurring work.
What all-agents does
all-agents takes a recurring business process — the kind someone on your team performs the same way every week — and turns it into something the computer runs on its own.
It starts by doing the work with the person who owns it. The first runs are handled by an agent that reads the inputs, makes the judgment calls and gets the job done while the operator watches. Anything irreversible stops and asks for approval first. Every accepted run is captured as a concrete example, and the written procedure is sharpened run by run.
When the operator decides the process is well enough understood, the system walks through it with them step by step and turns what it has learned into deterministic code. Steps that genuinely need judgment stay with the model. Steps where a person must stay accountable keep a permanent approval gate. See how runs are recorded and approved for what that looks like in practice.
Who it is for
The person we build for is the operator who owns a repeatable process and is drowning in it: reconciling invoices, chasing approvals, triaging tickets, enriching leads, updating three systems for every new customer.
They do not want to manage infrastructure, and they should not have to know what an API or a webhook is to get value. They know their own process, and that is the only expertise the product asks for.
Who builds it
all-agents is built and operated by Uli Prantz, who also writes the guides published here.
Why we publish guides
Most writing about automating business work is either vendor copy or generic listicles. Neither helps the person who actually has to decide which steps of their invoice process can safely run without them.
The library on this site is our attempt at the missing version: job playbooks, decision frameworks, downloadable procedures and working tools, built around the questions operators actually ask. Every page names its author, shows its sources, and says where its advice stops applying.
How we do that is written down in our editorial policy and research methodology. If we get something wrong, tell us and we will correct it.