How I Automate My Business Using Agents

Running multiple client projects alongside a personal brand means a constant stream of operational tasks: checking error dashboards, reviewing uptime, keeping track of what's shipped and what's next. Individually they're small, but collectively they eat hours every week and force context switches that break deep work.

I decided to hand these routines to autonomous agents. Not chatbots that wait for prompts, but scheduled pipelines that run on their own, query real services, and produce actionable summaries ready to review in minutes instead of assembled by hand.

What agents actually are

An agent, in this context, is a Claude Code session that runs autonomously with access to external tools. It can read files, call APIs through MCP servers, and write structured output, all without human intervention. The key difference from a regular chat interaction is that agents are scheduled: a cron trigger fires them at set intervals, and they execute a defined skill from start to finish.

Each skill is a focused instruction set. It tells the agent what to query, how to interpret the results, and where to write the output. The agent handles the execution: making API calls, parsing responses, deciding what's worth flagging.

The four automations

Error monitoring

The monitor-errors skill connects to Sentry via MCP, pulls the top unresolved issues across all projects, and compiles a weekly error digest. It flags new regressions, groups recurring patterns, and writes a summary that takes under two minutes to scan.

Uptime reporting

The monitor-uptime skill queries BetterStack for monitor availability, response times, and any incidents from the past week. The output is a health summary per project: uptime percentage, slowest endpoints, and whether any heartbeats went silent.

Email triage

The triage-email skill processes forwarded emails: client requests, invoices, scheduling threads. It classifies each message, drafts a reply where appropriate, and writes a per-client digest. Instead of a manual inbox scan, the output is a structured summary to review, approve, or edit.

Business health sync

The sync-business-impact skill reads roadmaps, dependency files, and config across all active projects. It detects what changed since the last run and pushes signals to a central project index, a single dashboard that shows the current state of every project without opening each one individually.

The pipeline

The ROI of boring automation

None of these automations are flashy. They don't generate content or make decisions. They collect information, structure it, and put it where it can be found. The value is in the hours not spent context-switching between dashboards and inboxes, and in the problems caught early because the digest surfaces them before they become incidents.

What's next

Two automations are on the roadmap that move from observability into business intelligence:

Revenue & Cash Flow: an agent that reads invoices, tracks payment status, and produces a monthly cash flow summary. The goal is to flag overdue payments and forecast upcoming gaps before they become problems, without manual spreadsheet reconciliation.

Client Delivery Risk: an agent that cross-references project roadmaps with actual commit activity, open tickets, and deadlines. If a project is falling behind or a milestone is at risk, it surfaces the signal early, leaving time to adjust scope or communicate proactively with the client.

Both follow the same pattern: scheduled agent, specialized skill, structured output. The infrastructure is already in place. It's a matter of writing the skills and connecting the data sources.