Skip to content
Splice Work

Agents

Meet the agents. Find your next handoff.

AI agents for small business, each focused on one job: calls, email, invoices, listing graphics or website management. See what each one does and how it fits your systems.

What they touch

AI agents for small business, inside the systems you use

Nico

  • Twilio
  • Slack or Microsoft Teams
  • Salesforce
  • Google Calendar

Answers the phone after hours as your receptionist, holds a real conversation, decides whether it is urgent, and writes the call up for the morning.

How Nico works

Beep Beep

  • Slack
  • Salesforce
  • ZoomInfo and Apollo
  • Firecrawl
  • NeverBounce
  • Gmail

Asks in Slack whether you want a campaign, builds the list from your CRM, researches each company, writes every email individually, follows up three times, then reads the replies and hands you the ones worth your time.

How Beep Beep works

The finance agent

  • Salesforce
  • QuickBooks Online
  • BILL
  • Stripe
  • Slack

Watches for a closed deal, creates the purchase order and invoice after approval, confirms the downstream copy matches, and prepares the payment link and email. It checks payment status before drafting follow-ups. A person approves every send.

How the finance agent manages receivables

Gable

  • Google Forms
  • Google Sheets
  • Google Slides
  • Slack

Fills the brokerage’s listing templates with supplied details, fits one uploaded property photo and returns an editable flyer in Slack for human approval.

How Gable works

Milo

  • Next.js
  • Vercel
  • GitHub
  • Slack
  • Google Search Console

Brings scheduled website checks and small improvements into a Slack workflow, with a preview for your review before publication. Initial scope covers an existing, supported Next.js site on Vercel with code in GitHub.

How Milo works

The one you actually need

We can build one for the job you name.

Most conversations start with a job that is not on this page. Someone is re-keying orders out of a supplier portal, or reconciling two systems that disagree every Monday, or reading the same category of email forty times a week and filing it. That is the work an agent is for, and it is almost never identical between two companies.

The build looks the same regardless of the job: we watch the work, scope one agent, write down what it will never do, build it against your accounts, and run it. What changes is the job itself. Tell us the one that eats your week and we will tell you whether it is a fit, including when the answer is no.

How they're built

Four rules every Splice Work agent follows

One job each

An agent that does one thing well is easier to trust, test and switch off than a platform that does everything badly.

Finishes inside your systems

The call becomes a card in Slack. The closed deal becomes an invoice in QuickBooks. Nothing lands in a dashboard you have to go and check.

A person approves

Customer-facing and financial actions wait for a click until you say otherwise. Every action is logged with what the agent saw and why it acted.

We run them

Hosting, monitoring, model updates, and a person on our side reading the logs. A flat monthly fee, cancel any time.

Questions we get asked

What is an AI agent, in plain terms?

Software that uses a language model to read something that arrives (a call, an email, a document, a record), decides what to do, and does it in your systems. A chatbot answers questions. An agent finishes a job.

Are these AI agents for small business, or enterprise tools scaled down?

They are built around the systems small teams already use: Salesforce, QuickBooks, BILL, Slack and Google Workspace. Gable, for example, handles listing graphics in a live brokerage pilot.

Can I buy one agent, or is it a platform?

One agent. Start with the job that hurts most. Add another when the first one has earned it.

What happens when an agent gets something wrong?

It's caught at the approval step, and the log shows exactly what it read and why it acted. We fix the rule, add a test, and the fix ships to every customer running that agent.

Can you build an agent for a job that isn't on this page?

That is the usual case. These agents are examples of the pattern, not a menu. If the work arrives somewhere predictable, the rules mostly fit on a page, and a mistake is cheap to catch with an approval step, it is probably a fit. Describe the job and we will tell you either way.

Who runs the agent once it is built?

We do. Hosting, monitoring, model updates when providers change them, and a person on our side reading the logs. It is a flat monthly fee you can cancel. You can also take the code and run it yourself, but most people would rather not own that.

How long before an agent is doing real work?

It depends almost entirely on how clean the access to your systems is. A documented API with credentials ready is a very different job from an on-premise system with none. We scope it after looking at your actual setup rather than quoting a number first.

Do the agents talk to each other?

Only where it helps. The email agent writes unsubscribes back to the CRM that the finance agent also reads, so nobody gets chased after asking to be left alone. Beyond that they stay separate on purpose — one job each is what makes them testable.

Want one of these on your systems?

Tell us which job. We'll tell you what it would take to run it for you.