Customer support

Answers in minutes in a timezone nobody covers.

Support is a coverage problem before it is a quality problem. An AI support rep answers at 03:00 in a timezone you do not staff, from the answers you have actually written down — and when it does not know, it says so and escalates rather than inventing something plausible. The compounding part is what it does afterwards: the third time a question appears, it writes the doc that ends it.

What you stop doing

The overnight backlog, and the Monday morning spent discovering what broke on Saturday.

PA
Pax
AI Support Rep
On duty
Routines
  • Ticket sweepEvery 15 minutes, around the clock
  • Promise chaseEvery day, 14:00
  • Theme reportFridays, 16:40
  • Shift handoverEvery day, 21:30
Skills
  • answer-from-docs
  • reproduce-and-file-bug
  • link-duplicate-threads
  • write-doc-from-repeat-question
  • weekly-theme-report
Next RunEvery 15 minutes, around the clock
The working day

What Pax does between waking the servers and going quiet.

Every hour below is one Routine you can open, read, and change — including the one where it stopped and asked a human.

One TuesdayPax · AI Support Rep8 Runs · 1 Decision
  1. 02:40

    Answers while the office is dark

    Four tickets arrive from Europe and Asia. Three are answered in under six minutes from the documented answer, with the exact link. The fourth has no documented answer, so it says so and escalates.

    Email
  2. 07:10

    Re-opens what regressed

    Finds the two threads a customer replied to after they were closed, links three duplicates onto the original, and puts the one that got worse overnight at the top of the human queue.

    Email
  3. 09:30

    Writes the doc that ends the question

    The same import error has come up eleven times this month. It opens a documentation pull request with the fix, the cause, and the error string people actually paste into search.

    Repositories
  4. 11:15

    Files a bug an engineer can act on

    Reproduces the checkout failure, captures the steps and the exact log line, and files it in Tasks against the right Project — not a screenshot in a channel with 'is anyone seeing this?'.

    Tasks
  5. 12:40

    Stops at the money

    Stopped for you

    A refund request goes past what the written refund policy covers. It assembles the account history, the contract, and its recommendation into a Decision, and does not touch the money.

    Decisions
  6. 14:00

    Keeps yesterday's promises

    Goes back through every thread where someone said 'I'll get back to you' and either gets back to them or explains why it is still open. Nothing ages silently.

    Email
  7. 16:40

    Tells product what it is hearing

    Groups the week's tickets by theme with counts and quotes, and posts it where the people who can fix it will read it.

    Workspace
  8. 21:30

    Hands over the night

    Writes the shift note: what is open, what is waiting on a customer, and the one thread that will need a human first thing.

    Notes
By the end of the day
6 min
Median first response, overnight
11
Repeat questions turned into one doc
1
Decision that needed you
What it brought to you
  • This refund is bigger than the policy covers. Full, partial, or hold the line?
  • A customer is asking for a contractual commitment we have not made before.

Answering one performs no side effect. The employee writes the question and the options itself, and an ordinary Member can answer it.

What a support rep can do

Capable in the ways that matter, bounded in the ways that count.

Answers from your truth, not its memory

Replies are grounded in the docs, runbooks, and past threads you granted it, with the link attached. A question with no documented answer gets escalated, which is the correct answer to give.

A Decision is a first-class outcome

Not knowing is a result, not a failure. Anything it cannot ground in your documentation, anything the refund policy does not cover, and anything a customer is angry about becomes a written question for a human, with the history attached.

Bugs engineers accept

A filed issue carries a reproduction, the environment, and the exact log line, in the Project that owns it — the difference between a ticket that gets fixed and one that gets asked for more detail.

Every thread is one timeline

Duplicates are linked, re-opens are visible, and the whole history sits on the Contact — so the human who picks it up is not reading the story for the first time.

Setting it up

Four documents, and then you stop being the trigger.

Everything that makes this a support rep rather than any other role is plain, editable text. Change how it thinks by editing a document, the way you would rewrite a job description.

Soul

How it judges

  • Voice, priorities, and the lines Pax will not cross alone.
Skills

What it repeats

  • answer-from-docs
  • reproduce-and-file-bug
  • link-duplicate-threads
  • write-doc-from-repeat-question
  • weekly-theme-report
Routines

When it works

  • Ticket sweep — Every 15 minutes, around the clock
  • Promise chase — Every day, 14:00
  • Theme report — Fridays, 16:40
  • Shift handover — Every day, 21:30
Grants

What it can reach

  • Gmail Connection for the support mailbox
  • The documentation Repository
  • The Projects it may file bugs into
  • Product docs and runbooks in Resources
Questions

The ones people actually ask.

How do I stop it inventing an answer?
It answers from granted Resources and past threads and cites what it used. Its Soul carries the rule that an ungrounded answer must be escalated instead — and you can read any Run to see exactly which sources it had.
Can it issue refunds?
It prepares them and it records them; a person moves the money. Genosyn deliberately ships no tool that lets an AI Employee disburse a refund, so the worst case is a well-argued Decision sitting in your queue.
Does it work my existing inbox?
Yes — Email is a real mail client over your Gmail mailbox with two-way sync, so your team and the AI Employee work the same threads rather than a shadow queue.

Build an autonomous company today.

Install Genosyn, choose an AI Model, write the first role, and put it on a schedule. Tomorrow morning, one job runs without you. The company grows from there.