sumn
Request early access
Learn

The questions people actually ask.

Running an agent without watching it raises questions a chat session never did: what actually ends the loop, and who checks work nobody has read. Each page answers one of them in its first two sentences, then shows the mechanism underneath.

26 Jul 2026guide

What is an agentic loop?

An agentic loop is what an agent actually does: pick an action, run it, feed the result back, go again. What matters is what gets to end it.26 Jul 2026guide

What is graph engineering?

Stages pay where parts run independently, where access changes, where a check has to hold, or where a rejection needs somewhere to go. Otherwise they cost you.26 Jul 2026guide

What is loop engineering?

A fresh agent session has no memory unless something restores it. Loop engineering is what you leave outside the conversation so the next pass can carry on.26 Jul 2026guide

Should agent work run as a loop or a graph?

One test sorts them. If the model gets a decision wrong and no check catches it, that was a decision to settle before the run started.26 Jul 2026guide

What actually changes as you scale from one agent to a thousand?

Every step up in agent count takes a job off the human, and each one only works once something other than you can prove the work is fine.21 Jul 2026guide

How do you run AI agents on a schedule?

Firing an agent on a timer takes an afternoon. The real work is keeping it safe to leave: a budget it can't outrun, and a way to know when it stopped half-done.16 Jul 2026guide

What should an agent have before it can act in production?

Nine questions that show whether an agent's safeguards are real, from what it can reach to what happens when things break. For any agent that acts, including ours.21 Jul 2026guide

What is an agent inbox?

One place a person decides what your agents can't decide alone. Why the queue exists, and how to design approvals people can clear in a minute.21 Jul 2026guide

How do you control what AI agents spend?

Provider caps, token meters and run budgets do different jobs. What each one stops, and why unattended agents need a cap: a report only lands after the money's gone.21 Jul 2026guide

What is multi-model code review?

Multi-model code review runs one change past several models independently. When a panel beats a single reviewer, how to read disagreement, what it costs.21 Jul 2026guide

What are AFK agents?

Five things have to keep working without you before leaving an agent running is safe.12 Jul 2026guide

What are background agents?

Background agents are AI agents that work without a person watching. What that actually requires, why chat sessions can't do it, and what to look for in a runtime.