Operating notes

Build a calmer, more observable company.

Field notes on running a company from one page: failed payments, agent cost, the seats nobody uses, and the alerts people actually read.

Revenue7 min read

Failed payments: why they happen, and how to hear about them before your customer does

Most failed payments are ordinary. The expensive ones are the cluster you did not notice for a day. Here is how to tell the two apart.

Operating systems6 min read

Every tool reported green while checkout was broken for six hours. Uptime was never the question.

Operating systems7 min read

The category is built for on-call rotations you do not have. What a small team needs is smaller, and cheaper, and mostly about noticing.

AI agents8 min read

Tracing tells you what the model said. Monitoring tells you what the agent cost, whether it finished, and whether anyone noticed.

AI agents6 min read

Per-token pricing is public. Per-run cost is what you pay. The gap between them is where budgets go missing.

Cost6 min read

Enterprise spend tools start at a price that would pay for the seats you are trying to reclaim. Here is the version that fits.

Operating systems5 min read

Every alerting tool can post to Slack. Most channels get muted within a month. The difference is who decided what deserves a ping.

Operating systems6 min read

External status pages explain downtime to customers. Internal status pages explain what the business needs from you right now.

Incident response5 min read

The most useful alert is not that a metric crossed a line. It is that a promise to a customer or the business is at risk.

AI operations7 min read

An agent completing a task is not the same as an agent producing the intended business outcome safely and economically.