Introducing the Agent library

Today we are opening the Agent library: a catalog of agents we have built, deployed, and run for partners and ventures, ready to deploy into your operation. There are twelve at launch across support, sales, marketing, operations, engineering, finance, and research.

These are not chatbots. Each one does a job end to end inside the tools you already use, comes back for approval when it should, and is measured against the queue it replaced.

What is in the library

• Support triage. Reads inbound tickets, tags and routes them, drafts replies for approval, and escalates the ones that need a person.
• SDR outreach. Researches accounts, scores contacts, drafts personalised sequences in the rep's voice, and books meetings on the rep's calendar.
• Lead qualifier. Works inbound forms and chat, asks the qualifying questions, and hands warm leads to a person with a summary.
• Content scheduler. Turns a content calendar into drafted posts, routes them for approval, and publishes on schedule.
• Review responder. Watches review platforms, drafts responses in your voice, and flags reviews that need escalation.
• Ops reconciler. Matches orders, shipments, and payments across systems and lists every mismatch each morning.
• Invoice chaser. Tracks receivables and sends polite, escalating reminders on a schedule you set.
• Code reviewer. Reviews every pull request for correctness, security, and style against your repo's conventions.
• Bug triager. Reproduces incoming bug reports, files structured tickets, and assigns them.
• Reporting analyst. Pulls from your analytics, ads, and CRM into a weekly report with the three things that changed.
• Market monitor. Tracks competitors, pricing, and news in your category and delivers a daily brief.
• Onboarding guide. Walks new customers through setup, answers questions from your docs, and flags stuck accounts.

How deployment works

An agent from the library deploys in one sprint. Week one we connect it to your systems, load your conventions and voice, and run it in shadow mode against real work so we can compare its output to what your team actually did. Week two it goes live in approval mode. A person reviews every action until the numbers earn it more autonomy.

Every agent ships with three documents: a boundary that says what it may do, must ask about, and may never do; an evaluation set built from your real cases; and a runbook for the person who owns it.

{  "agent": "support-triage",  "may": ["read tickets", "draft replies", "tag and route"],  "ask_first": ["issue refunds under policy", "close tickets"],  "never": ["change account credentials", "contact customers off-platform"]}

What it costs

Library agents are priced as a sprint to deploy and a monthly retainer to run, which covers monitoring, evaluation reruns on every model or prompt change, and the human review time. Details are on the pricing page. Custom agents, for jobs not in the library, are scoped as regular Build work.

Why a library

Because the same twelve jobs come up in nearly every company we talk to, and because an agent that has run in production for months is safer than one built from scratch on Monday. Agents enter the library only after they have run for a partner or a venture. The library is where they go once they have earned it.