Every IT company suddenly "does AI." We've grown with these tools from the days of pasting questions into a chatbot, through AI coding assistants, to the teams of AI agents we run today. Along the way we've done it all: the most capable AI services in the cloud, AI on your own hardware so sensitive data never has to leave the company, and a clear sense of which to use where - and for what. Now we use that experience to help you find the tools that fit your business.
Setting up a new client site used to be a 2–3 hour job across four dashboards. Now it's one command: it plans the whole launch - domain, hosting, security certificate, a safe home for the site's code, publishing the site, email - shows us exactly what it's about to do, and waits for approval before it touches anything. About five minutes, start to finish.
The tools that keep our 100+ client websites updated have existed for years - a handwritten update script and a lot of practice. With AI we rebuilt them into something complete: updates that check their own work, and safety checks before anything touches a live site.
We use cloud AI services and AI models on our own hardware, and we choose between them job by job. When a task involves sensitive documents, we can run it in-house, so that data never has to leave the company. When it needs more horsepower than our hardware has, it goes to the cloud.
One of our agents has a dedicated machine: it browses, clicks, and works a full desktop, physically walled off from passwords and logins it doesn't need. It hunts missing receipts during reconciliation and specs out and shops for hardware when we need it.
We were organized before. Now we cheat a little: an agent reads our project threads and client requests, tracks every decision and to-do, and sends a daily digest - the tasks stuck waiting on a reply, the deadlines quietly sneaking up, the follow-ups that would otherwise slip. The humans still do the work. The agent just never forgets what the work is.
AI automation takes on the work that fixed rules stall on. Most small businesses already run some plain automation: a nightly backup job, or a reminder that goes out on the first of the month. Those tools repeat the same steps and stop as soon as the input looks different.
Your staff spend their hours on the input that looks different. A client email asks three questions and mentions a deadline in passing. A supplier sends a quote as a scanned PDF. An AI model can read input like that, work out what it means and suggest the next step.
Some jobs need no AI. If the steps never change and the input arrives clean, a plain rule costs less and is easier to audit. We'll say so when that's the case.
An AI agent is a model with a goal and access to tools. It can open files, use software and work through several steps on its own, then hand the finished work back for review.
For years, a handwritten update script and a lot of practice kept our 100+ client websites updated. We rebuilt those tools with AI, so each update checks its own work and safety checks run before anything touches a live site. Setting up a new client site used to take two to three hours across four dashboards. Now one command plans the domain, hosting, security certificate, a safe home for the site's code, publishing the site and email, and the setup takes about five minutes.
We run agents on office work too. One reads our project threads and client requests, then sends a daily digest of the tasks waiting on a reply and the deadlines coming up. Another hunts missing receipts, and specs out and shops for hardware. Our people still do the work, and the digest shows them what's waiting.
The work that suits agents has a common shape:
Some work suits agents poorly. One-off decisions and work built on relationships belong with your people, and so does anything where one mistake costs more than a year of saved hours. Most AI for business pays off on a dull, recurring task that eats hours each week, so start there.
An agent that is allowed to change things can make a wrong change to a live website or to the systems your office runs on. We build three controls into client AI work.
An agent gets the access its task needs and no more. Our office admin agent has a machine of its own, kept apart from the passwords and logins it doesn't need. If you gave an agent your website updates, it would need to reach the sites it maintains. It would have no reason to hold the login for your bank or your payroll system. Give each agent its own account, and you can switch one off without touching anything else.
An audit trail is a record of what an agent did and who approved it. If a setting on a website looks wrong months later, someone can trace it back to the step that changed it. We build that record in from the start. Before you trust any AI tool with your systems, ask the vendor to show you that record for a single change.
The agent pauses before a consequential step, and a person approves it. If a check fails or the agent can't decide, it stops and leaves the item for a person to handle. Our own site setup tool has a gate like this: it lays out what it plans to create for a new client site and waits for our approval before it touches anything.
Most AI tools send your data to a provider's servers for processing. For a lot of tasks that's acceptable. For patient charts or legal files, a professional obligation or a client contract may rule it out. In Alberta, the Health Information Act sets rules for health information held by custodians such as clinics, and the Personal Information Protection Act (PIPA) sets rules for personal information held by private-sector organizations. Depending on the data and your contracts, keeping it in-house can be the simplest way to meet them.
For data like that, we can run the model on a machine in your office or server room. Your staff reach it over your own network, so those documents don't have to go to an outside AI provider.
A model small enough for one office machine gives up some capability for that privacy. It can pull invoice numbers and dates from a stack of scanned forms, or sort incoming documents into the right folders. Summarizing a long contract or reasoning across a year of correspondence still goes better on a large cloud model, where the data allows it.
Our AI consulting starts by asking where your team's hours go. We ask which tasks repeat every week and which ones stall when the one person who knows them is away.
Bring real examples to that first conversation, such as the checklist someone works through every month or the email thread where a request got lost. Real samples show us where the input is messy and where a fixed rule would do the job.
A good first project for AI workflow automation comes with a clear test for a right answer. For a website update, the test is whether the site still works afterward: the pages load, the contact form sends and the store still takes orders. For document sorting, the test is whether each file lands in the right place. For a follow-up digest, you check whether a promised task ever drops off the list.
Brown Walrus is an independent Edmonton IT company, founded in 2011, with a team that has worked in IT since 2001. The same company handles managed IT, web development and cybersecurity, and a person picks up when you call during support hours.
Automation follows fixed rules, while AI can interpret messy input and decide what to do with it. A rule gives the same output for the same input and stops when something unexpected arrives. Most useful business systems combine the two. A mail rule can file every message from a client into the right folder, and an AI model can read that message, spot the deadline mentioned in passing and add it to the follow-up list.
AI helps a small business most on recurring work that eats staff hours, such as keeping software up to date, sorting documents and tracking follow-ups. Brown Walrus, an Edmonton IT company, runs AI agents on that kind of work in its own operations, from updates on more than 100 client websites to a daily digest of stuck tasks. A good first project is one repeated task where the time saved is easy to measure and the result is easy to check.
To automate a task with AI, pick one recurring workflow and write down the rules a person follows to do it. Decide which steps need human approval, give an AI agent only the access that workflow needs, and keep a log of what it does. Check its output before trusting it with more work. Brown Walrus starts client projects with a single workflow and widens the scope only after the output checks out.
AI agents for business are AI models given a goal and access to software, so they can carry out multi-step work instead of only answering questions. A chatbot can explain how to set up a website. An agent can plan the web address, hosting, security certificate and email, show a person the plan and carry it out once approved. Brown Walrus sets up new client sites this way, and builds client workflows with limits on what each agent can touch, a record of everything it does, and a person signing off on the important steps.
A small business can try consumer chatbots without an AI consultant. Outside help earns its cost once AI touches client data or the systems you run on, because someone has to limit what the agent can access and decide where a person signs off. Brown Walrus does that work as an Edmonton IT company that also runs managed IT, websites and cybersecurity, and it will tell you where AI won't save you time.
Yes. A business can run AI on a local model hosted on a machine in its own office or server room, so sensitive documents don't have to leave the business. Brown Walrus can set up local models for clients whose health records, legal files or client financials need to stay on their own network. Models that run on a single on-site server are usually smaller than the leading cloud models, so they suit focused tasks like sorting documents and pulling fields from forms.
we identify the workflow where AI genuinely saves you hours, and tell you honestly where it wouldn't.
limits on what each agent can touch, a record of everything it does, and a person signing off on the important steps, designed in from day one. Nothing runs unsupervised.
setting up AI models on dedicated hardware when your data needs to stay in-house.
start with one workflow, check the results, expand what works.
