Blog
Running AI agents in production
Guides to running AI agents in production: observability, approvals, testing, cost and scheduling, and how we run our own company on AgentOS.
Case study · 27 September 2026
How we run AgentOS on AgentOS
Our support, outreach, onboarding and social posts run on our own agents. What that looks like, and four things it taught us.
Case study · 27 September 2026
Every post on our LinkedIn page is drafted by an agent and approved by a person
How our company page is run by two agents and one approval, what went wrong, and what we changed.
Guide · 27 September 2026
What shipped: late September 2026
Approvals that read like the thing being approved, a plain-language summary on every run, safer errors, and a new look across the product.
Guide · 26 September 2026
How to add human approval to AI agents
Which agent actions need a person's sign-off, how an approval gate should work, and how to keep the queue from turning into a bottleneck.
Guide · 26 September 2026
A weekly review for your AI agents
A short routine, in the same order every week, that catches the slow problems no alert fires for. With examples from our own fleet.
Guide · 25 September 2026
What to log for every AI agent run
The record you need to answer "what did the agent do, why, and what did it cost", and the alerts that tell you when to look.
Guide · 20 September 2026
Template: Inbound Support Triage
A ready-made agent that sorts your support inbox, answers how-to questions from your own product notes, and flags the rest for a person.
Guide · 13 September 2026
Who owns an AI agent? Roles, audit and accountability
As agents take on real work, someone has to be accountable for each one. How to decide who can build, change and approve, and how to keep a record of it.
Guide · 6 September 2026
Scheduling AI agents: what should run on its own
Which agents should wait to be asked and which should just show up, how to pick a cadence, and how to notice when a scheduled agent goes quiet.
Guide · 30 August 2026
How to build a support inbox agent that knows when to stop
A worked example of a support agent that answers the routine questions, leaves the sensitive ones to a person, and shows its work.
Guide · 23 August 2026
How to know when an AI agent breaks
Agents fail quietly. The signals worth watching, the alerts that catch them, and how to get from a failure to a fix without reading every run.
Guide · 16 August 2026
Human-in-the-loop without the bottleneck
The difference between approving an action and answering a question, and how to bring a person into an agent's run only when it matters.
Case study · 9 August 2026
What we learned from letting agents do our cold outreach
We built a pipeline of agents to find prospects and write to them. The writing worked. The finding did not. What we learned, and what we do now.
Guide · 2 August 2026
How to trace AI agents you already run
Three ways to get a full trace of an agent you have already built, from a few lines of SDK to pointing your OpenTelemetry exporter somewhere new.
Guide · 26 July 2026
What an AI agent really costs, and where it hides
Why the monthly bill is the wrong place to look, the usual sources of waste, and how per-agent budgets catch a runaway agent early.
Guide · 19 July 2026
How to test an AI agent before it goes live
Write down what good looks like as test cases, check what the agent did and not only what it said, and run them after every change.