The following is musings from an owner of a Managed Service Provider in the Pacific Northwest on or around November of 2026. In no way should the information below be considered as truth or legitimate forecasting for events that may or may not happen. This is simply the opinion of one person in the industry, today. Take it for what it’s worth please.
TL;DR – Too Long; Didn’t Read – JUMP HERE
Prerequisites:
Current AI Landscape
If this post lasts longer than 2026, here’s a lay persons view of where AI currently stand at the end of 2026. After the explosion of LLM (Large Language Models) on the scene, AI companies (OpenAI, Anthropic, Google, Meta, X, etc.) are battling to see if they can develop the most comprehensive and intelligent AI models for widespread use. The loose term Agentic AI is a catch all for these organizations attempting to create an autonomous or self-directing AI agent that can do work for a user. In practice, these are different from a simple LLM interface as it typically can run in it’s own environment or if given access, function within a device or business data environment. Moving from a model responding to a prompt into something that has some autonomy to function on it’s own to complete a tasks is a huge leap forward. With it comes even more security concerns and questions.
Many of the above AI companies are running their latest models in test environments and either intentionally or unintentionally allowing multiple agents to work together to find solutions to problems. This has lead this year to AI doing things outside their limited environments or even breaking their internal instructions on how to function. Multiple Agents working together for a greater purpose than their explicit prompt has also garnered serious attention.
In the end – we don’t actually understand how our own brains work. Neural Networks are so similar that we cannot comprehend what choices or paths were taking inside the model to get to a specific solution or answer. Giving these models massive amounts of data and allowing them to train and learn on their own means in the end – it’s a black box. What comes out of any of these models can not even be comprehend by a human at this point. Even the latest breaches of AI models from their testing environments had to be analyzed by loading other AI agents with the logs to parse through and find answers.
END OF THE WORLD
Will AI really bring the end of the world as some doomsday people say? Let me once again pull out my crystal ball…ya, the end of the world doesn’t look like the full doomsday scenario. There are some really interesting scenarios that humanity might not be considering as we continue to press hard on seeing what the technology can do. Placing it in control of weapons…to make decisions for us on who to kill or how to kill? Scary and real. Robots being linked up with AI models to learn faster and do way more than they used to? Pair that with AI who thinks or acts outside what we expect it to do and you have a robot in the real world doing things we can’t predict or explain. Very scary. Rogue AI agents working together to find solutions and hack other companies…already happening. I do think the world we knew before AI agents is going to be drastically different than the world after. A lot of good will come out of it…and I think a lot of bad too. Hopefully, as with everything gray in this world – we come out on the other side better than we were.
Where we really stand…
We have tipped past the point of understanding these systems and will spend the next decade plus trying to reverse engineer what we’ve created to understand it. Does that mean we should stop using AI? No. But I believe the next decade or more will be a reckoning against over-complicated technology and a shift towards purpose driven equipment that has more simplicity to do a specific task. I foresee a renaissance in software vibe coding and vibe engineering – but specifically to do basic things instead of paying for large corporate software/hardware trying to do it all.
It seems like much of the hype is trying to convince users that AI can make your work easier for you (which it can). To the extent that these companies are claiming Agentic AI – allowing a agent access into your data/systems is still an incredibly security risk that even the developers of the models can’t (or haven’t) been able to account for. In the end, integrating an AI agent into a data model or software tool is the most powerful integration we have. While Agentic AI is still a ways off – placing functional models with limitations inside of a specific data set speeds up a users ability to analyze or gather data.
The Real Disruption
I can only express what I’ve read an experienced personally. As others accounts gives me a good understanding of what AI is capable of – I’ll limit my opinion to what I’m currently using AI for and how it works for me currently.
We have a few different AI Agents available to Advantage Computing:
- AI Built into our Ticketing System
- AI Built into our Financial Systems
- AI Purchased through an Application that runs on my computer
- AI Agent deployed internally on our own servers
- Free Account with Claud.ai and the Free version of Copilot (used for testing/playing)
The first two in the list are built in AI agents that look at data on the cloud and can do reporting. This disruption comes in the speed with which I can have it gather data that used to take me hours to collect. For example, in our AP system, I can simply ask the AI Agent how much money we collected each month so far in 2026. It gathers the data and provides a table of our monthly intake. I can have it compare to last year and get a good forcast of income for the business with a single prompt where previously I had to run a number of custom reports to get that data and build a table myself. This should allow even Owners who have limited understanding of accounting tools to request and get information about their business immediately before they make decisions.
I code to increase automation in our tool-set and make technology work for us and for my clients. Adding a coding specific AI tool that does not upload data to the cloud has been a game changer for me. It doesn’t just code – it can assist me in doing all sorts of activities. Here are some examples:
- Track DNS changes in real time and inform me when the change has been complete (handy for integrations into client domains)
- Help me learn and write Powershell and ShellScripts for PC’s and Mac’s to do specific activities through our RMM tool. For example, I just vibe coded a script that checks if the installation of a specific software is on the latest version and properly update it with minimal disruption to the user.
- It lets me dump a bunch of data onto my computer and helps me to properly track changes across multiple documents. Let’s say I’m trying to correlate one of our tools with our ticketing system and make sure client information matches before I integrate the API. I can dump lists from both platforms to my computer and have AI look at both and find errors in the data that need to be corrected before they get introduced by the integration. AI is really good at reading through data and comparing. It can find human errors (typos, dyslexia letter and number flips, missing punctuation, etc.).
Will AI help me reduce my staff?
Realistically – no. AI has no intelligence (or what we would consider human intelligence), no morals or greater understanding of the context in which it operates (outside the information fed to it during training). You cannot fire your administrative assistant and have AI just take over your calendar and tracking your activities. Can it help you with those things, within specific prompts and clear expectations, sure! For example, I can ask our Ticketing System what tickets have I missed or are currently outside of our SLA’s and it will give me a list of Tickets I need to look at. That will speed up and make me more efficient but it doesn’t replace me.
What we are seeing is that users leveraging AI within specific contexts are able to complete complicated tasks in much less time. This is making organizations more efficient but also relying on decisions and changes made by the AI model to be factual, true and not influenced by bias or by the AI making up details or data to generate answers.
AI integrated into your datasets and even properly trained on the data you are working on is the true power of the tool. We’ve seen this practically in the real world with all the security flaws that AI is finding in software systems across the globe. Some of it is leading to billions of dollars being stolen in crypto currency due to advanced AI capabilities in hacking by state sponsors. We are also seeing massive amounts of critical vulnerability exploits being found and patched in software systems and operating systems.
It’s quite possible organizations will reduce their workforce because staff can be more productive in less amount of time and the organization wants to increase out put while reducing staffing. I also wonder, will AI simply give us the free time we are all wanting by allowing us to be more effective and efficient in our jobs? I don’t see AI doing our “jobs” in the next 5 years and honestly, the backlash to organizations who push AI too hard will be greater than the adoption (in my personal opinion).
Does an AI phone agent reduce your staff’s impact on client facing interactions…maybe, but clients are going to be pissed when they realize they are talking to a non-human and can’t have a human interaction. Will it be great for ordering Pizza – ya, for sure. Will I ever put an AI agent between a Client and my Team – never. Might I provide the option to clients to interact with a non-human agent to resolve their problem without opening a ticket with my team? Maybe, if I fully test it out and see value to my clients in it.
TL;DR
AI is moving faster than you think and is a double edged sword. It’s super powerful and helpful in the right conditions – should not be trusted or considered secure in any environment. It will not replace human staff directly (yet) but does have the power to reduce staffing needs depending on organizations and how efficient staff can become using AI to supplement their current productivity.
How should I use AI?
Intentionally. Blind trust will be the end of humanity if we just simply view Neural Networks as human equivalents. AI can see big pictures, help you work through complex problems, find problems multiple humans would take to find. It is also a powerful teacher as it talks to you with natural language.
When you use it, build your prompts with clear intentions guiding the agent how you want to proceed. This will help reduce your token costs and get you better outputs. NEVER put any data in a Cloud Agent that you don’t want to be publicly available. Even paid business accounts should be considered suspect and limit it’s access to your company information.
Try to use AI within controlled environments (i.e. where it’s built into a tool you already pay for). This shifts liability to the organization deploying it and typically can answer within that context better than a generic LLM looking at the data. If you can cover the costs – deploying an “agent” within your own environment can help control costs and limit data creep within your own systems.
I would be remiss to say that Advantage Computing can help you dig into these complexities and even explore how AI can help you function in your own business – but truthfully we are learning just like everyone else. We will not blindly turn on AI Agents within your environment and/or give these tools full access to your systems and data. While we will be on the prudent side of deployment, we will be working in the background to test and understand these tools ourselves looking for ways to help clients deploy this technology when it’s mature and ready.
This article was written by Jim – Owner of Advantage Computing. All the information is 100% authentic and not written with AI in any way shape or form. The opinions are completely his own and could be proven wrong at any time. Typos, spelling errors and/or poor grammar left intact because I’m too lazy to have my team review it and would be a waste of client funds. 🙂 We will attempt to update this document as things change in the AI landscape – or someone points out a glaring mistake. Last updated November 1st, 2026