---
title: "AI Agents and Automation · 316Software"
description: "Reading invoices, sorting requests, copying data from one system to another: the rule-based work that eats your team’s week. I build agents that do it the same way every time and show how they got each answer."
canonical: "https://316software.com/services/ai-agents-automation"
last_updated: "2026-10-21"
---
# AI Agents and Automation

Reading invoices, sorting requests, copying data from one system to another: the rule-based work that eats your team’s week. I build agents that do it the same way every time and show how they got each answer.

![A mechanical arm lifting a paper invoice from a full in-tray toward a laptop, with the finished papers filed in a binder on the other side.](https://316software.com/img/services/ai-agents-automation-1536.webp)

The tray on the left is what someone works through by hand every week. The agent handles the 90th item the same way as the first.

## What you get

- **An agent with one clear job**: Narrow, testable, and easy to describe in one sentence. Agents with a broad, vague job are where these projects usually go wrong.
- **Work you can check**: Every output links back to the input it came from. No mystery numbers.
- **A person signs off where it matters**: Anything involving money, liability or a signature gets drafted by the agent and approved by a person.
- **A test set from your own work**: Real cases from your business, so any model change gets tested against them before it goes live.

 

Each value in the record points back to where the agent read it on the invoice, so you can check one against the other. The total is money, so a person approves the record.

## Pricing

Fixed prices for the common jobs, and a quote after discovery for anything bigger. Prices in CAD, plus GST.

- 01 Fits most businesses

  **One agent**

  One job you can describe in a sentence, done the same way every time.

  $25,000 fixed · 3 to 4 weeks

  One narrow task: read from one source, produce one output into one destination, with a human check step, an evaluation set of 50 of your own real cases, deployment, and 30 days of tuning.

  - One source, one output, one destination
  - A human check where money or a signature is involved
  - An evaluation set of 50 of your own real cases
  - Deployed, then 30 days of tuning

  [Start with One agent](https://316software.com/contact?service=ai-agents-automation)

- 02

  **A workflow**

  Several steps, several systems, approvals in between.

  from $35,000 after discovery

  Several steps across 2 or more systems, with approvals, and a dashboard of every run. Scoped in discovery.

  - Several steps across 2 or more systems
  - Approvals where a person has to decide
  - A dashboard of every run and its source
  - Paid by phase

  [Talk about A workflow](https://316software.com/contact?service=ai-agents-automation)

- **Run plan · from $3,000 a year**: The infrastructure it runs on, kept up, backed up and watched. The server it runs on. Storage, and backups you can be restored from. Dependency and security updates. Monitoring, and someone who notices. Cancel at any renewal. You own the software either way.

- That’s the starting price. A busier or more complex system costs more to run, and we agree on the number in writing along with the build, so it never comes as a surprise.
- Not included: what you pay other companies. AI model and API keys, third-party licences and app subscriptions stay in your own accounts, at cost. I don’t mark them up.
- The plan doesn’t include hours for changes. Each change gets its own price, agreed before I build it.
- I’ll never take a percentage of what you sell: no royalties, no cut of transactions, no per-user fees for software you own.

Priced per agent. I won’t quote a broad, do-everything agent: those are the ones that fail.

15%

**One narrow job, a person in charge, 15% more done.**

In a Fortune 500 company’s customer support, 5,172 agents were given an AI assistant that suggested replies they could use, edit or ignore. Issues resolved per hour rose 15% on average, and 30% among the less skilled and less experienced; the most experienced gained little. The fixed tier has the same shape: one narrow job, a person signing off where money is involved, and 50 of your own real cases to check it against before it goes live.

[Brynjolfsson, Li & Raymond, Generative AI at Work, Quarterly Journal of Economics, 2025. 5,172 support agents.](https://academic.oup.com/qje/article/140/2/889/7990658)

## When it’s the right choice

- The task happens often, follows rules a person could write down, and mostly involves reading documents or moving data.
- It’s done well today, just slowly, so there’s a right answer to measure against.
- Someone on your team will own it after handover.

### When it isn’t

Skip it if an occasional mistake would be expensive and nobody’s going to check the output. Someone on your side has to keep an eye on what it produces.

## Get in touch

[Talk about AI agents & automation](https://316software.com/contact?service=ai-agents-automation)

I’ll read your message myself and reply within 2 business days.

## Related services

[All 8 services](https://316software.com/services)

- ### [AI Consulting](https://316software.com/services/ai-consulting)

  When you’re not sure yet which task is worth automating.

- ### [Integrations](https://316software.com/services/integrations)

  When the rules are fixed and a plain connection between systems will do.

- ### [Software Development](https://316software.com/services/software-development)

  When the agent needs a system around it to work in.

---

Page: https://316software.com/services/ai-agents-automation
The whole site for agents: https://316software.com/llms.txt