Use case
AI automation discovery
Automation usually stalls on the first question: what is even worth automating? Point an agent at the exhaust of your working week and its only job is to notice the rituals you stopped noticing, then apply for the ones worth taking over.
Can AI find what to automate?
Yes. Give an agent a feed of your work exhaust, such as shell history, git activity, a calendar export, or forwarded email, and it can watch for sequences that repeat and write up the ones that look like a job. On Qoren the agent only proposes. A person approves before any proposal becomes a scheduled task.
The problem
Most automation dies at discovery, not at build. Nobody has a list of what is repetitive, because repetition is invisible from the inside: the four-step deploy check you assemble every Friday feels like thinking rather than a procedure, and the invoice email you write from the same skeleton every month feels like writing. Workshops only surface what people remember to mention, which is usually the loud annoyances rather than the quiet, expensive rituals. So the tasks that would pay back fastest are the ones nobody ever names.
How an agent handles it
- Read the work exhaust you choose to send it: shell history through the Qoren CLI, git hook output, a weekly calendar export, forwarded email.
- Watch across weeks instead of one session, so a monthly ritual is still visible.
- Describe a repeated sequence back in plain language, with the dates it happened.
- Write the skill definition for the task it wants, and show past runs it would have handled.
- Hold every proposal at a human approval gate rather than acting on what it finds.
- Turn an approved proposal into a scheduled task in the same environment.
Why it sells
- A shortlist of automation candidates drawn from real work rather than a workshop.
- Each candidate arrives pre-scoped, with the skill definition already written and evidence attached.
- The office grows one approved job at a time, at a pace a person sets.
How it runs, step by step
Connect the exhaust
You pick what the agent is allowed to see, and it only ever sees what you send. The Qoren CLI is the shortest path from a terminal: a nightly line such as qoren agent message discovery "$(history | tail -n 200)" hands over the day's commands. A post-commit git hook can send the same way, and calendar exports or forwarded email cover the work that never touches a shell.
Set the observation window
The agent runs always-on in its own environment, so it accumulates rather than sampling. You set how long it watches before it is allowed to propose anything, typically a few weeks, because a ritual that happens every Friday needs several Fridays before it looks like a pattern rather than a coincidence.
Look for ritual, not for work
Its standing instruction is narrow: find sequences that recur with the same shape and a stable cadence. Four commands in the same order every Friday for six weeks is a candidate. A hundred one-off commands are not, however busy they look. Volume alone is never the signal; repetition with a rhythm is.
The job application
When it finds one it does not act on it. It writes an application: what the ritual is, how often it fired, a draft skill definition for the task it wants, the schedule it would run on, and three past occurrences it would have handled, so you can check its reading against work you remember doing.
The approval gate
Nothing takes effect until a person says yes. You approve, edit the scope first, or reject it, and a rejection is useful signal too because it tells the agent which patterns are not jobs. This gate is the whole design: an agent that could deploy its own proposals would be a very fast way to grow automation nobody asked for.
Approved, then scheduled
An approved proposal becomes a scheduled task in the same environment, starting at whatever autonomy you chose, usually draft-only until it has earned more. It runs alongside the agents already there, under the same spend caps, and its output lands in whatever recurring oversight review you run across the office, so a job approved in March is still on the record in June.
When this is not the fit
This finds rituals, so it needs work with a rhythm and a trace. Genuinely different weeks contain nothing to find, and work that leaves no exhaust the agent can read stays invisible to it no matter how repetitive it is. It is also not an audit of an entire organization: it watches the feeds one person or one team chooses to send, not every system in the company. And a team that already keeps an honest list of what it wants automated should just build from that list.
Templates to deploy for this
Start from a ready-made agent and tailor it to the client. Each one runs on a managed environment, online on schedules and triggers.
Kpi Reporter
Friday afternoon: what happened to your numbers, why, and what to do — anomalies flagged mid-week
Founder Os
The operating cadence good companies run on — quarterly goals, a Monday plan of three bets, a daily pulse that only speaks when something slips, a Friday review with your real numbers, decisions graded against what actually happened.
Frequently asked questions
Through whatever feed suits the work. The Qoren CLI lets you pipe from a shell, so a scheduled one-liner can hand over the day's command history, and a git hook can post commit activity the same way. Calendar exports and forwarded email cover work that never touches a terminal.
Related use cases
Deploy this for your clients.
Pick a template, configure it per client, and go live in minutes. Plans from $39/mo.
Get started