We Need to Talk About Agents

The word agent is everywhere right now.

That’s a problem.

For a lot of small businesses and nonprofits, the word makes AI sound bigger, riskier, and more technical than it needs to be. When you’re running a small team in Calgary or managing a non-profit in St. Albert, you don’t have time for more technical debt. You need things that work.

Leaders hear “agent” and picture some semi-autonomous digital worker roaming across systems, making decisions, taking actions, and creating new headaches. Sometimes that’s fair, but often, it isn’t.

That matters because many of the most useful agents for small organisations are already here. They just don’t look like the flashy ones getting all the attention. If we spend all our time worrying about the futuristic versions, we miss the tools that are right in front of us that can save five hours this week.

Most small organisations are being shown the wrong kind of agent

Have you noticed how much of the AI conversation feels like it’s written for Silicon Valley developers rather than Alberta small business owners? Some agent tools are built for technical users who are comfortable with setup, permissions, maintenance, and risk.

Take OpenClaw, for example, which is getting SO much attention. It sits much closer to the DIY agent-platform end of the spectrum. For most micro businesses and nonprofits, that’s a poor starting point. It can quickly become one more thing to manage, one more security concern, and one more system nobody really has time to supervise.

That doesn’t mean agents are hype. It means most small organisations should stop staring at the hardest version first. We don’t need to build a digital employee from scratch when we can use a tool that’s already been built to handle a specific task.

The useful agents are already here

What small organisations should care about are not the most autonomous agents. They should care about the most useful ones. We’re looking for the ones that save time, reduce friction, fit the tools you already use, and don’t create a governance mess.

Let’s break down the five types of agents that actually matter for your day-to-day operations.

1. Custom specialists

These are tools you shape for one job. Think of them as your dedicated subject matter experts that never get tired and never forget the brief.

Gemini Gems let you create custom versions of Gemini for specific tasks. Google describes them as personalised AI experts. ChatGPT GPTs do something similar. OpenAI says GPTs can be tailored with instructions, knowledge, capabilities, apps, and actions.

For a small business, that could mean: • a proposal-writing assistant • a quote follow-up helper • a customer FAQ specialist

For a nonprofit, it could mean: • a grant-writing assistant • a donor communications helper • a board briefing assistant

These aren’t fully autonomous agents. That’s part of their value. They’re focused, reusable, and much easier to trust. You aren't giving them the keys to the building: you're giving them a specific desk and a specific file.

I use a Gemini Gem for my business. ‘His’ name is Kai and ‘he’s’ my co-CEO. I’ve trained him with all my key business context. Kai remembers more about my Business Plan and my custom research than I do. Kai acts as a sounding board and advisor - I’ve trained ‘him’ to respond as a truth-teller not a smoke blower. I want a reality check not apple polishing. 

2. Contextual coworkers

These tools keep work together over time. One of the biggest drains on a "Chief Everything Officer" is context switching. These agents help by maintaining the memory of a project so you don't have to start from zero every time.

ChatGPT Projects are built as persistent workspaces with files, instructions, chats, and project memory. Claude Cowork is Anthropic’s move toward more agentic, multi-step work, aimed at helping Claude handle ongoing tasks, not just one prompt at a time.

That’s useful for: • grant campaigns • board packages • policy development • vendor research • recurring client work

Think of these less as robots and more as AI coworkers with memory. They understand the "why" behind the work because they've seen the previous drafts and the source material.

3. Research coworkers

Some tools act less like autonomous agents and more like AI research partners. They don't go out and "do" the work in the world, but they help you make sense of the information you already have.

NotebookLM is a strong example. Google positions it as an AI-powered research assistant that works from your own source material: documents, websites, videos, audio, and more. It can summarise, connect ideas, answer questions across your sources, and generate audio overviews.

For a small business, that could mean: • organising market research • comparing vendors • pulling insights from meeting notes and sales calls • building internal knowledge from scattered files

For a nonprofit, it could mean: • synthesising policy material • preparing board briefings • organising grant research • comparing funders, programmes, or service models

NotebookLM isn’t a roaming autonomous agent. That’s part of its value. It helps teams think with their own information, in one place, without needing a technical build. It respects the boundaries of your data.

NotebookLM is one of those tools that I can’t believe more people aren’t using. It’s one of my ‘go to’ applications because it’s a Private Vault not a Public Library and it’s focused on sources I trust. That constraint makes it more, not less, powerful.

4. Focused work agents

These do one job well and save real time. They are the "point solutions" of the agent world. They don't try to be everything: they just try to be the best at a single, repetitive task.

Granola is a good example. It records and summarises conversations, then turns them into usable notes and follow-up material. These could be team meetings, customer conversations, or focused internal discussions about your FAQs, policies, or Standard Operating Procedures (SOPs). Turning unstructured conversations into useful data has real value.

Deep Research (ChatGPT, Gemini, Claude, etc.) is another. These are agents that can find, analyse, and synthesise information into a structured report.

For a small business, that means: • better meeting notes • faster customer follow-up • competitor research • pricing analysis

For a nonprofit, it means: • donor meeting summaries • funder research • policy scans • board briefing notes

These are agents that help real work move faster. They take the "admin" out of the administration so you can get back to the mission of your organisation.

5. Recurring watch agents

Some of the most useful agents for small organisations are the simplest ones. These are the digital equivalent of a "weather watch" or a news alert, but far more sophisticated.

Tasks in ChatGPT let you set up automated prompts that run later or on a schedule. ChatGPT Tasks can recur at specific times and send results by push notification or email.

For a small business, that could mean: • a weekly scan of competitor promotions • a check on pricing changes • a roundup of local market news • a summary of customer review themes

For a nonprofit, it could mean: • a weekly scan of grant announcements • a funder environment update • a summary of policy or sector news • a scan of donor or community trends

That’s an agent too: not because it sounds futuristic, but because it quietly handles repeatable work for you. It’s about the outcome, not the autonomy.

I have three Tasks that run weekly. Every Thursday and Friday morning I get an email with a link to the latest scan I’ve scheduled. This helps me to stay on top of not just breaking news but also bigger trends shaping the small business and nonprofit adoption scene

The wrong question

When we talk about this at augmentus inc., we often find that people are asking the wrong question. They ask: "Is this a real agent?"

The better question is: "Is this useful for a small team without creating more risk than value?"

That’s the standard that matters. We shouldn't be chasing "agentic" behaviour for its own sake. We should be chasing efficiency and clarity. In the Alberta business environment, we value pragmatism over hype. We want tools that respect our time and our existing AI strategy.

What small organisations should look for

If you're looking to bring an agent into your workflow, it should meet a few simple criteria.

A good small-organisation agent should: • save real time • work inside tools you already use • be easy to explain • keep a human in the loop • avoid creating technical headaches

That’s why tools like Gems, GPTs, Projects, Claude Cowork, NotebookLM, Granola, Deep Research, and Tasks matter more to most small businesses and nonprofits than high-friction agent platforms. They allow for "Step 5: Verify" in the augmentus protocol. You can see what they're doing, you can check the output, and you remain the pilot - autonomy is great for humans but you want to limit it for agents.

The agent era is here. But for most small organisations, the best agents aren’t the most autonomous. They’re the ones that quietly help a small team do better work with less friction, less waste, and less drama.

Are you spending your time managing your tools, or are your tools helping you manage your business? That’s the kind of agent worth paying attention to.

The question isn't whether AI can act on its own. The question is: can it help you act more effectively?

Jeff Uhlich, augmentus inc. 

ps - for the past month I've been testing an agentic platform that provides me several AI 'employees' that 'plan' and execute work on my behalf. I'm testing it so you don't have to and I will report on it here with a few more weeks of experience. I'm not recommending it yet. It requires a fair bit of set up and then the agents need to be 'managed'. They're far from perfect but in some domains they've been real time-savers. Stay tuned!