AI Agents on the Shop Floor: Cool Engineering, Real Questions

I read a piece from Industry Week this week on AI agents coming to manufacturing — it’s a little ahead of itself. But the core question it raises is worth sitting with: if an AI agent is making decisions at 2 a.m., do you trust it?

First, the distinction the article draws is actually useful. An AI assistant is a chatbot. You ask it things, it responds, maybe it does a task when you tell it to. An AI agent is different. An agent has context, has rules, has data behind it, and it makes decisions on its own — without you in the loop. That’s what we’re talking about.

For manufacturing, the concern is obvious. An agent that misreads a spec or reshuffles your production calendar at 2 a.m. doesn’t just cause a headache — it costs real money. Parts come out wrong. Schedules collapse. You wake up and nothing is working right.

I’m building one of these myself right now. A service coordinator agent — something that looks at the work across my whole team, checks skill sets, evaluates how urgent each issue is, figures out who should handle what and when, and notifies the customer on timing. Humans do that job today. I’m not always impressed with how humans do that job, because humans get pulled into things. The most urgent item in front of them becomes the most important thing in the world, and meanwhile three other things are quietly on fire.

An agent shouldn’t have that problem. It should be able to hold the whole picture at once.

My agent scored about 57% on its last test run against historical data. That sounds bad. But it told me exactly why — and the problem wasn’t the AI, it was some rules I’d put in that were steering it wrong. That’s actually the right failure mode. It’s fixable. I’ll keep you posted on how it goes.

The article’s practical advice for this week is solid, so I’ll pass it along with my own commentary:

• List every AI automation tool currently running in your business. If you don’t know what’s running, you can’t supervise it.
• Put one person on your team in the AI checkpoint role. Don’t let autonomous decisions just happen without a human reviewing the dashboard, at least while you’re still training the thing.
• Check with your vendors whether recent updates have added agentic or autonomous features. This one I want to underline. If a vendor quietly turns on autonomous decision-making without telling you — get a new vendor. That’s not a feature rollout, that’s a liability you didn’t sign up for.

We’re getting close to the point where agents can make sound decisions if they have the right data. The data part is always the hard part. But we’re moving in the right direction.

Authenticated Email Has a Gap You Should Know About

This one’s been making the rounds as a big revelation. Authenticated emails can still be faked. I’m not surprised. Let me explain what’s actually going on.

There are email authentication technologies called DKIM and DMARC. When set up correctly, they essentially sign your outgoing email so that the receiving server can verify it actually came from your domain — not from some bad actor spoofing your address. We set these up as part of our managed services work whenever we have the access to do it. They work. They’re worth having.

The gap is this: DKIM and DMARC protect your domain. They don’t protect your account. If someone gets into your actual email account — because you don’t have multi-factor authentication turned on, because there’s no monitoring watching for impossible logins — they don’t need to spoof anything. They’re sending from the real account. The email is legitimate. Every authentication check passes. And the message inside it is a fraud.

We run alerting systems that flag when an email account gets accessed from two different locations in a timeframe that doesn’t make sense — Traverse City in the morning, somewhere overseas an hour later, that kind of thing. Most businesses don’t have that. Most businesses also don’t have MFA turned on. Those two gaps together are how this plays out in the real world.

What to do this week, in order:

1. Make sure DKIM and DMARC are configured on your domain. Start there.
2. Get email account monitoring in place so impossible logins get flagged.
3. Turn on MFA. I know I say this every week. Turn on MFA.
4. Train your accounts payable team specifically. Don’t release a payment because you got an email. Don’t even release it because you called the number on the email and someone answered and sounded right — AI voice cloning is convincing enough now that this is a real attack vector. Call a number you already have on file from a previous invoice or your own records.

Secure email and email security are not the same thing. Your domain can be fully authenticated and your account can still be wide open. Both matter.

3D Systems Opens a Manufacturing Center, and I Have a Question for Northern Michigan

3D Systems, a company I’ll admit I hadn’t heard of before this week, just opened an industrial technology center in Colorado. The goal is to give manufacturers — including smaller operations — access to advanced manufacturing equipment, expertise, and training that would otherwise require a full R&D lab budget to touch.

My immediate reaction: where’s the Michigan version of this?

There probably is one somewhere in the state. But I’d love to see something like this in Northern Michigan. Traverse City should be asking how to get a facility like this here — whether that means going to the Michigan Economic Development Corporation, talking to NWMCA, or finding a large manufacturer or aerospace company willing to sponsor a next-generation production and training facility that our local manufacturers could actually use.

The broader push toward domestic manufacturing reshoring is real, and I think it’s important. There’s no good reason we can’t build more of this stuff here. If any of you are in manufacturing and want to come on the show and talk about what advanced manufacturing techniques would actually move the needle for your operation, I would love that conversation. I’d learn a lot and I’d enjoy every minute of asking questions.

Quick Hit: Chrome Is Getting an AI Scam Spotter

Google is rolling a Gemini-powered feature into Chrome that watches the sites you visit and flags potential scams in real time. It’s a useful idea. The caveat is that the bad guys are always at the bleeding edge and the defensive tools are always following. That’s just the nature of this. But making the browser itself a layer of protection — rather than relying entirely on the user to notice something’s off — is the right direction.

It won’t catch everything. It’s one more layer, not a replacement for the others. But I’ll take it.

That’s the week. If you’re curious about the November 18th webinar, stay tuned — more on that soon. See you next time.

— Albert

Questions about anything we covered today? Schedule a free IT consultation and we’ll walk through it with your business in mind.

Wherever your team is based, ACS provides hands-on managed IT support — see how we help businesses in Traverse City & Northern Michigan.

We have had so much feedback from people worried about AI in their business that I did a special edition of Nerds News instead of waiting for our regular Thursday show.

Chapters

• 0:00 What shadow AI is and why this is a special edition
• 2:37 How big it is: PagerDuty, Gartner, Microsoft, IBM
• 6:56 Your best people, not bad guys
• 7:38 What it looks like in a real shop: quotes and pricing
• 9:14 The bookkeeper and your P&L
• 10:43 Blocking it at the firewall doesn’t work
• 11:13 Customer data and your obligations
• 12:02 The three fixes
• 13:21 Free AI policy template
• 15:04 The two-minute test
• 16:28 How ACS runs AI
• 18:34 Where to start this week

What shadow AI is, and why this could not wait

Shadow AI is the AI that gets used without anybody knowing about it, except the people using it. It is AI tools your staff use for company work that the company never approved. Usually it is a personal ChatGPT, Copilot, or Gemini account, on a phone, a computer, or in a browser tab.

What scares me is that the free accounts are very powerful. People can get a lot of work done with them and create a lot of damage at the same time.

So why now? Because AI went from a curiosity to a normal way to work, and that happened in the last 18 months. Most owners have not caught up.

How big it is, and why this is not about bad guys

PagerDuty ran a survey in June of this year, 2026. It found 66% of office workers have used AI tools they believed were not allowed at work, and 88% have put work information into an AI tool. A third of them put in customer data, and a third put in financial or confidential documents. Public AI versus private tools is a big deal, and I will get to why.

Gartner has 68% of employees using unauthorized AI, up from 41% in 2023. They are doing it because they would rather drive than walk.

The Microsoft and LinkedIn Work Trend Index looked at small and medium companies and found that 80% of the people using AI at work there bring their own AI. If you are not providing it, they are bringing it. It might be a free public AI, or they might be paying for it themselves. If it is not configured correctly, whatever goes into it could end up as training data. That is why the difference between public and private matters. Big corporations have this locked down. Small business is where it is worst.

IBM’s 2026 breach report says shadow AI was involved in 43% of incidents, up from 20% the year before. That is better than a 100% increase in a year. Breaches involving it cost about $400,000 more than average.

We are not talking about bad guys. We are talking about your best people doing their best work with the best tools. Why walk when you can drive? Why take a bus when you can take a plane, and not just any plane, Delta quality at discount airline prices, up to and including free?

What it looks like in a real shop

Quote generation is the one I know is happening, because I have talked to people about it. Take a custom manufacturer. Every quote is unique. There are general guidelines around it, but for the most part each one is different. An AI tool can make that faster, better, and more consistent. It is a powerful tool, and you want your employees using it.

The problem is a free tool, or an AI program that is not set up correctly. The pricing, the customer, and the methodology you use to build the quote all get fed into the AI. You are training it on how you do business, and it can turn around and use that same information for your competitors.

Say your bookkeeper uses an AI tool you do not know about. She takes the profit and loss, the asset lists, what you spend on administration or parts, and dumps it into Gemini on her free account. If you want to train your competitors on how your financials work, more power to you. I highly recommend we not do that.

Say you try to stop it at the firewall so nobody can reach ChatGPT. It is just as easy to use a cell phone to get around that. You are not stopping it with a block alone. You need teeth.

Customer data is probably the biggest one. We have a fiduciary responsibility to protect our customers’ information. If our employees feed it into AIs, and the AIs train on it or pull it in for whatever purposes those companies have, we are not doing our job. In some cases it is illegal. If you do defense work or you are in healthcare, you cannot let that information walk out the door.

The three things that fix it

Visibility. You cannot write a rule for tools you do not know about. Your Microsoft 365 sign-in and app consent logs show more than people expect, and that is the first thing the ACS readiness assessment looks at. Where are people signing in? What can we pull from firewalls and browser history so we know what is being used, who is using it, and how?

A policy people physically sign. It lists the approved tools, what data can go into them, that everything going out gets reviewed, and what happens if someone breaks the rules. Have your lawyer look at it. We are providing one for free at acsapp.com/aipolicy, our AI acceptable use policy template. Bring it to your team so nobody is confused.

An approved path. You have to give people the tools, because these tools are going to be good for you, maybe even better for you than for them. Not using AI today is like telling people in 1992 they could not use a word processor and had to use a typewriter. You are hurting your own bottom line because your people cannot be as productive. Shadow AI happens because the approved option is slower or does not exist. Copilot inside Microsoft 365, with the permissions cleaned up and ready, beats a personal ChatGPT, and your data stays in your tenant. Claude and GPT have the right ways to do it too, and so do all the big players.

The two-minute test, and how we run AI at ACS

To see where your company sits today, we have a small, simple tool. Do you have a written policy? Are your permissions set up right, and have you reviewed them? Do you know what AI tools your employees are using today? If you cannot answer all three, you are not behind. You are normal.

The full version is 10 questions and takes a couple of minutes. It scores you and tells you what to do first. We are not collecting anything. It is a free check at acsapp.com/aicheck, our AI readiness self-assessment.

At ACS we run a highly advanced AI setup. We have our own in-house AI on our own hardware, and we use the most advanced models from the biggest companies. We have built dozens of integrations, close to a hundred, between our own software and the vendors we use. We know how to make it so data never leaves the building.

We call our solution Managed Intelligence. The foundation is the policy, shadow AI monitoring, and Copilot or other AI license management. It runs from the assessment through creating and managing agents, training your employees, and building workflows that make them more productive.

So go get your free assessment at acsapp.com/aicheck. Go get your free download of the policy at acsapp.com/aipolicy. Take the test, get the policy, and give it to your team.

Questions about anything here? Schedule a free IT consultation and we will walk through it with you. More episodes are on the Nerds News page, where you can subscribe on Spotify, Apple Podcasts, or YouTube. Managed IT services in Traverse City from Advanced Computer Solutions, 402 S Union Street Suite B, Traverse City, MI 49684, 231-267-TECH (8324).

Let’s get into it.

Or watch on YouTube.

The Electric Motor Boom Is Real, and Northern Michigan Is Positioned for It

Automation World is calling this a strategic moment for firms to assess whether their current capabilities align with the electric motor supply chain. That’s consultant-speak for: there’s a gap, and somebody’s going to fill it.

EVs, HVAC upgrades, industrial automation, and yes — robots — all need electric motors, and demand is surging across all of them at once. The push to rebuild domestic manufacturing capacity is real, and for once, small manufacturers aren’t just watching the big players collect the upside. Supply chain openings like this tend to close fast, though. If you’re a manufacturer or a tier-one supplier in this region and you haven’t been talking to motor manufacturers about what they need, that conversation is worth having this week, not next quarter.

What makes electric motor manufacturing interesting from a process standpoint is that it’s not stamping or injection molding. You’re assembling lots of precision components — laminated steel, wound wire, tight tolerances — and the skill set required is different from general fabrication. Northern Michigan has those skilled people. That matters more than most economic development reports give it credit for.

Alert readers who are in manufacturing or supply chain: AI is actually useful here for prospecting. Finding electric motor manufacturers and tier-one suppliers in your region, identifying procurement contacts, drafting outreach — that’s a legitimate use case where the tools earn their keep.

Teams Deepfake Detection: Microsoft Did the Smart Thing (For Once)

Microsoft announced that Teams is getting deepfake detection. The headline sounds like a feature. The reality is more interesting.

What Microsoft actually built is a portal — a gateway inside Teams that lets third-party deepfake detection tools plug in and do their job. They didn’t build their own detection engine. They built the door and said, come on in.

Honestly? That’s the right call. The deepfake arms race is moving faster than any single vendor can track, and Microsoft has enough on its plate. Third-party tools that specialize in this are going to be sharper, updated more frequently, and more competitive on price. The tradeoff is that you’ll have another vendor to evaluate and another subscription to manage — but that’s a better problem than relying on a detection tool that’s already six months behind the threat.

Why does this matter for your business specifically? If your team uses Teams for vendor negotiations, payment approvals, or any conversation where someone’s identity carries real financial weight, you are already in the threat model for this. Criminals are impersonating executives and vendors on video calls — not just in text or email. You can be looking at someone who looks like your CFO, sounds like your CFO, and is asking you to wire money. It’s not your CFO.

Here’s a low-tech trick worth knowing right now, before any software is deployed: if something feels slightly off on a video call, ask the person to put their hand in front of their face and touch their nose. Current AI video generation struggles badly with hands obscuring faces. The image will go wrong in a way that’s immediately obvious — the face distorts, the illusion breaks. It’s not a perfect test, but it costs nothing and takes five seconds.

The third-party tool ecosystem for Teams deepfake detection is going to get crowded fast. Evaluating which ones are actually worth deploying is exactly the kind of thing worth getting ahead of.

The Ransomware Recovery Firm That Was Playing Both Sides

The owner of a Florida-based ransomware remediation company — Monster Cloud — is now facing federal fraud charges. The scheme, once you see it, is almost elegant in a deeply irritating way.

A business gets hit with ransomware. Monster Cloud shows up and says: don’t pay the ransom, we have the technical expertise to recover your data, we’ll handle it for less. Client pays Monster Cloud. Monster Cloud goes to the ransomware operators and negotiates a lower payment. Pockets the difference. Never actually performed the remediation they claimed to be doing.

So the client paid for a technical recovery. What they got was a middleman negotiation with criminals, marked up, with the markup kept secret.

A few things worth noting here. First, this is exactly why you want a vetted incident response vendor before something happens, not while your business is down and you’re panicking. Vetting a vendor under pressure is how you end up with Monster Cloud. Second — and this is important — even a legitimate ransom payment doesn’t clean your system. If the ransomware payload is still sitting on your machines, you’ll get encrypted again. Payment and remediation are two separate problems, and conflating them is how you get burned twice.

Third: the FBI takes a dim view of businesses sending Bitcoin to international criminal organizations, even when those businesses are the victims. Ransomware groups are sometimes sanctioned entities. Paying them can create legal exposure for you, not just the criminals. It’s not a reason to never pay — sometimes it’s the least-bad option — but it’s a reason to have a conversation with someone who knows the current sanctions landscape before you do.

Forty years in prison for the dark web marketplace operator who was enabling ransomware tools, by the way. That was the last story of the show. Not a great career outcome.

Federal Funding and a Live Test of America.gov

The Department of Energy has funding opportunities open for small manufacturers right now. The Trump administration also launched a new AI-powered portal called America.gov, designed to consolidate federal funding information into one searchable place.

I tested it live on air. The short version: it found the right agency, generated a link to the DOE funding exchange, and that link was dead. The exchange had moved. To be fair, following the redirect got me to a page where I could actually search for funding opportunities — so it wasn’t a total loss. But the AI should have followed that redirect itself and surfaced the live page, not the stale one. That’s a solvable problem, and maybe they’ll fix it. Right now it’s a tool that gets you closer to the right place faster than a cold Google search would, but it’s not doing the hard part, which is helping you actually navigate the application.

The Visa Foundation also has a US Small Business Growth and Innovation Program accepting applications. Worth searching for if you’re a small business owner looking for non-federal options.

The money is out there. The work to get it is real, and the portals are imperfect. But if you’re a manufacturer and you haven’t looked at what’s currently open, it’s worth an hour of your time.

Have a great week, everyone. See you next Thursday.

— Albert

Questions about anything we covered today? Schedule a free IT consultation and we’ll walk through it with your business in mind.

Wherever your team is based, ACS provides hands-on managed IT support — see how we help businesses in Traverse City & Northern Michigan.

Let’s get into it.

Or watch on YouTube.

AI Agents Are Different From Chatbots — and That Difference Matters

The Small Business and Entrepreneurship Council put out an article this week on AI agents coming to small business. I’ll be honest: they’re a few months to a year behind the curve on this one. But I read it anyway, because it’s worth knowing what the mainstream conversation is catching up to.

Most people still think of AI as a very smart chatbot. You type something in, it types something back. Useful, sure. But agents are a completely different animal. An agent doesn’t just answer a question — it handles a workflow from end to end. Scheduling follow-ups, processing data, moving information between systems, making decisions at each step without someone babysitting every click. The article uses those examples, and they’re fine, but I’ll tell you the agents I’m running are doing things considerably more involved than that.

The key insight on agents: if you are good at writing things down — your process, your steps, your decision rules — an agent can be as smart as you are, or smarter, because it can pull from everything you’ve written, everything it knows, and everything it can find. That’s the real power. It’s not magic. It’s structured knowledge plus automation.

Right now my go-to agent system is Claude. It’s one of the more capable tools out there, and also one of the more expensive. I’ve been experimenting with Grok (from Elon’s team) as a more affordable alternative, and the honest assessment so far is that Grok isn’t quite there yet — not because the underlying model is necessarily weaker, but because the tool ecosystem around it isn’t as built out as Claude’s. That matters a lot when you’re building multi-step workflows. I’ll keep testing and report back.

The practical question for your business: do you have processes that require the same steps every single time? That’s your agent opportunity. And before you go shopping for a new tool, check what you already have. Microsoft Copilot has agent-like features. Your CRM might too. I was talking to someone recently who spent four hours wrestling with a website builder, then discovered the AI assistant built into the platform and had everything fixed in fifteen minutes. The capability was there the whole time. Worth digging into what you’re already paying for.

Copilot Consumption Billing: Go Check Your Dashboard Today

This one I want you to actually do something about, not just read and nod.

Microsoft is turning on consumption billing for Copilot. What that means in plain language: once you hit your token limit (tokens are essentially the units of AI processing you’re buying with your subscription), instead of just stopping and telling you you’ve hit the wall, Microsoft can now automatically start charging you for more. Without a separate confirmation from you. Without a big warning. Just… a bigger bill.

I don’t love that. My preference is always: hit the limit, show the user a message, let them decide if they want to pay for more. That’s how it should work. Flipping on auto-billing in the background is the kind of thing that produces a very unpleasant surprise at the end of the month.

Alert readers who are heavy Copilot users, this is especially relevant to you. The smarter the AI model you’re running, the faster it burns through tokens. There’s a significant gap now between the least capable and most capable models, and if you’re running the good stuff hard, your allotment goes fast.

What to do: log into your Microsoft 365 Admin Center. If you’re not sure where to look, literally ask Copilot — it’ll point you to the right spot. Find the consumption billing setting and, in my opinion, turn it off. Run out of tokens before you run up an unexpected bill. I’ve heard stories of organizations burning through staggering amounts on tokens in a month. You don’t want to be that story.

Before You Enable Copilot, Run a Permissions Audit

This is something we’ve talked about with clients before, and it keeps coming up because people keep skipping this step.

When you turn Copilot on, you are essentially handing a very intelligent, very fast assistant the keys to everything it can reach in your environment. SharePoint, OneDrive, potentially local files depending on how things are configured. Copilot will find data that users technically have access to but that nobody ever expected to surface in a conversation. And it will surface it, because that’s what it’s designed to do.

The failure mode isn’t Copilot doing something wrong. The failure mode is your permissions being wrong — and Copilot being the thing that finally makes that visible in the worst possible way. HR files visible to employees who shouldn’t see them. Client files accessible to other clients. Financial records with broader access than anyone realized.

The fix is a permissions audit on your SharePoint and OneDrive before you flip the switch. It’s not complicated, it does require some IT skill to do properly, and it doesn’t take a lot of time or money. It’s just a step that needs to happen. Don’t skip it because you’re excited to try the new thing.

Quick Hits: Windows Backup, the Pentagon Breach, and the 26H2 Update

Windows Settings Backup is now automatic for Entra-joined devices. Most of our clients are on Entra (formerly Azure AD) rather than a local domain controller, so this applies to you. A lot of your Windows and Microsoft 365 settings will now follow you from device to device automatically. Swapping to a new machine gets meaningfully less painful. This is good engineering. Should have shipped years ago, but it’s here now.

Pentagon HR breach, 3+ million records. I bring this up not to alarm you but to make a specific point: the Pentagon has a security budget that would make your eyes water, and they still got hit. So the “I’m just a small business in a small town, nobody’s coming after me” logic doesn’t hold. They don’t know who you are. They don’t know where you are. The internet is the ultimate shotgun — it costs almost nothing to spray attacks across the entire world simultaneously. You’re not being targeted because you’re important; you’re being targeted because you’re reachable. That’s the conversation worth having.

Windows 11 26H2 arrived as a lightweight enablement package. The honest takeaway: Windows updates can break things, this one included. We do our best to push the most critical updates quickly and hold back anything we know is causing problems until Microsoft patches the patch. But we can’t test every update against every application across every device configuration — nobody can. If something starts behaving strangely after an update, that’s the first thing to mention when you call.

Anyway. That’s the week. Go check that Copilot billing setting, and have a great weekend.

— Albert

Questions about anything we covered today? Schedule a free IT consultation and we’ll walk through it with your business in mind.

Wherever your team is based, ACS provides hands-on managed IT support — see how we help businesses in Central Florida.

Or watch on YouTube.

Claude Money: AI That Reads Your Bank Account

Let me give Claude and Anthropic their due before we get into the new feature. Back when ChatGPT was eating every headline, Claude was out there quietly doing the same thing — and in a lot of ways doing it better. Better prose, better reasoning, less hype. I used it regularly. Most people hadn’t heard of it.

Then Anthropic did something that changed the whole category. They made agents real. They made coding with AI approachable. They gave the AI the ability to interact with your desktop, reach your files, hold memory across sessions, and actually know what’s going on with a project from one day to the next. That was impressive engineering, not marketing. And now they’re one of the most competitive players in the space.

So when Anthropic announces something new, I pay attention.

The new feature is called Claude Money. The pitch: connect Claude’s AI assistant directly to your business bank accounts so it can analyze spending patterns, cash flow, and overall financial health. If you’re running a business big enough that you’re not personally eyeballing every transaction, that kind of automated pattern recognition could surface things you’d otherwise miss — a recurring charge you forgot about, a cash flow dip that’s developing slowly, a vendor that’s quietly billing more than they should be.

I’ll be honest with you: for my business, I know my bank account well enough that I’m not sure I need this. But I can absolutely see the value for a manufacturer or a service shop with a lot of moving parts and not enough hours in the day to read every line item.

Now, the question you’re already asking: is it safe?

My read is yes, with one important caveat. The AI isn’t going to log in and move your money. You’re granting it read-only, limited access. That part doesn’t concern me much. What does concern me is this: if you’re on a free AI plan, or even a basic paid plan where you haven’t checked your settings, there’s a real chance the AI is training on your data. Including your financial data. Including the patterns of how your business makes and spends money.

Think about what that means. You could be teaching a model — one that your competitors also use — exactly how your operation runs financially. That’s not a theoretical risk. That’s a setting you need to go check right now.

You don’t need me to walk you through the steps. Ask the AI itself. Literally open Claude or ChatGPT and say “how do I turn off training on my data” and it will walk you through it. Do that before you connect anything sensitive.

Claude Money is currently waitlist-only. Head to Claude’s website if you want to get in line. And for what it’s worth, I’ve already connected Claude to my email — both personal and business — and it’s been a real productivity boost. Being able to ask it to surface an email about a project, or draft a reply on my behalf, is the kind of thing that sounds small until you’ve used it for a week.

Honestly? I’d love Claude Money more for my personal banking than my business. I watch the business account daily. My personal account? That’s where the mystery spending lives.

NSF’s $20 Million Deep Tech Pilot: Small Manufacturers, Read This

The National Science Foundation is running a $20 million pilot program aimed at helping small US businesses — specifically service and manufacturing firms — take what they call “deep technologies” and turn them into commercial products.

Alert readers will notice this is federal grant money, which means yes, it came from us originally, and yes, getting any of it back requires paperwork. That’s the deal. But $20 million is real money, and the eligibility criteria are broader than you might think.

The program is looking for businesses that have developed a proprietary process, a specialized tool, or a unique technique — even something your team considers routine — and wants to help you figure out whether it could be commercialized. In other words: is there something you built internally to solve your own problem that other businesses would pay for?

That’s an interesting question to sit with. A lot of small manufacturers have developed workarounds, fixtures, workflows, or quality control processes that are clever and unique. They just never thought of them as products.

If you want to explore it: visit NSF.gov and search for the Deep Tech Commercialization Pilot Program. Review your eligibility. Write down any internal process, tool, or technique your business has developed that you’d consider non-obvious. Then honestly evaluate whether someone else would pay for it.

Could you spend the rest of your life chasing federal grants instead of running your business? Absolutely. I’m not suggesting that. But if you’ve already got something worth commercializing, this is a funded path to do it. Worth twenty minutes of your time to find out.

Manufacturing Tech, Cybersecurity Gaps, and a Chicago Trade Show

Two quick items for the manufacturers in the room.

First: the IMTS trade show is happening right now in Chicago, and Taiwanese tech suppliers are showcasing smart manufacturing equipment. The show runs on a two-year cycle, so this is one of those moments where you get to see what’s actually shipping in manufacturing technology, not just what’s being announced. If you’re close enough to Chicago to make it happen, it’s worth looking into. I wish I’d known about it sooner.

Second, and more practically: a new report is out highlighting a widening cybersecurity gap among small and mid-sized US manufacturers. I’m not going to pretend this is surprising. Security is expensive, the compliance lift is real (especially if you do any defense-adjacent work), and most small shops are running lean.

But here’s the thing about manufacturing specifically: when a machine goes down because of a security incident, you’re not just losing data. You’re losing production time on equipment that cost you real capital to acquire. The cost of downtime on a CNC machine or a production line is not the same as the cost of downtime on an office laptop. The math changes fast.

Baseline security doesn’t have to be a massive investment. Backups, endpoint protection, monitored alerts — you can get the fundamentals in place without a full enterprise IT budget. The report apparently includes some checklists if you can track down the original article in Advanced Manufacturing. Worth finding.

Windows 11 24H2 Loses Security Updates in October

Short, important, no drama: if you’re running Windows 11 version 24H2, Microsoft is ending security updates for that version next month. The fix is straightforward — open Windows Update and run your updates. Get current.

I know updates feel like background noise. They’re not. An unpatched home PC is a door into everything else on your network — your business files if you work from home, your banking, your family’s devices. It’s all connected. The update takes twenty minutes. Do it this week.

That’s the show for today. I’ll be out next Thursday, but I’m going to try to get something recorded ahead of time so you’re not left without your weekly nerd fix. Until then — take care of yourselves, and go run your updates.

— Albert

Questions about anything we covered today? Schedule a free IT consultation and we’ll walk through it with your business in mind.

Wherever your team is based, ACS provides hands-on managed IT support — see how we help businesses in Chicagoland.

Or watch on YouTube.

AI agents on the shop floor — and why small is actually fine here

A UK engineering firm called 42 Technology has released an on-device AI agent designed to live directly on top of manufacturing equipment. Instead of shipping your machine data up to a cloud data center somewhere, a small AI agent sits right next to the machine and processes data there. The article is from Manufacturing Management, and it’s squarely aimed at manufacturers — so if you run a shop floor, pay attention.

Now, I want to be honest with you about what these agents actually are. They’re not the big, context-aware models you’d use to write a proposal or analyze a contract. They’re small. They’re running on $800 to $1,200 computers. And within those constraints, they’re not going to be programming your next CNC operation or making high-stakes decisions. Don’t expect that.

What they are good at is interpreting a steady stream of machine data — cycle times, quality measurements, tolerances — and doing something useful with it. Flagging anomalies. Sending alerts. Making the kind of pattern-recognition calls that straight workflow rules just can’t make. A hard-coded rule says “alert me if this number goes above X.” An AI agent says “this combination of readings is trending toward a problem I’ve seen before.” That’s a real difference, and it’s worth something on a production floor.

The other thing that makes local AI genuinely interesting for manufacturers is the security angle. If you’ve got data you don’t want leaving the building — and a lot of my manufacturing clients do — an agent that never touches the cloud solves that problem cleanly. No data center. No API calls going out. The intelligence stays on-site.

For context on what’s possible: I’ve been doing some research on this, and Apple has a device in the $10,000–$12,000 range that can reportedly run something comparable to Claude Opus 4 locally. That’s a serious AI in a box. The $1,000 shop-floor version isn’t that — but it doesn’t need to be. For monitoring a machine and surfacing the right alerts to your team, it doesn’t need to be Opus 4. It needs to be consistent, fast, and local. And that’s achievable right now.

If you’re a manufacturer and this is making you curious, reach out. I’ve been thinking about how to replicate this kind of setup using tools I already work with, and I think there’s a real path to doing it for less than what a specialized vendor would charge.

WatchGuard firewall vulnerability — calm check, quick fix

WatchGuard is one of the most common small-business firewalls out there, and there’s a remote code execution vulnerability in it right now rated 9.8 out of 10 on the CVSS severity scale. That’s about as high as it goes. Attackers are actively using it to deploy ransomware.

One quick note on ransomware, because I think it’s worth saying plainly: the first generation encrypted your data and held it hostage. Today’s version does that and exfiltrates your data before encrypting it — meaning they’ve already copied it out before you even know something’s wrong. The threat isn’t just “pay us or lose your files.” It’s “pay us or your customers’ data goes public.” Average cost to a small business is somewhere between $40,000 and $50,000. That’s the real number.

The practical question right now is simple: is your WatchGuard management interface reachable from the open internet? This is a pattern we keep seeing in these situations. The management interface gets left exposed, and that’s the door attackers walk through. It should not be open to the web. At minimum, if it has to be accessible remotely, put it behind a VPN. MFA alone isn’t enough here.

If you have a WatchGuard and you’re not sure what your exposure looks like, that’s a five-minute conversation. Call your MSP. Call me. Let’s just check it.

Windows 11 scrambled your desktop — September update fixes it

Alert readers may have noticed their wallpaper and display configuration resetting randomly over the past few weeks. No warning. No explanation. Just a different-looking desktop when you sat down in the morning.

That was a Windows 11 bug, and the September update fixes it. Credit where credit’s due — Microsoft did eventually patch the thing they broke. The fix is in the September cumulative update, so make sure it’s installed.

This is also a good moment to repeat something I say constantly: leave your computer on overnight. Updates install when the machine is on and idle. If you shut down the moment you’re done for the day, those patches never land. We can push updates to managed machines, but even then the computer has to be on to receive them. Leave it on. Let it update. Reboot in the morning if you need to. That’s the whole workflow.

Quick hits: Dynamics 365 AI and small business credit cards

Microsoft is rolling out new AI automations in Dynamics 365 this fall. If you’re running Dynamics, you’ll get a guided, low-lift path to turn on agentic AI features through Copilot. Nothing you need to act on immediately, but worth knowing it’s coming.

On a completely different note — small business credit cards are apparently getting better in 2026. Forbes is reporting improved rewards and lower fees. I find the whole credit card economy fascinating in a slightly baffling way: the card companies extract money from vendors and hand some of it back to buyers, while keeping their cut in the middle. It’s a strange machine. But if you’re running business expenses through a card anyway, you might as well be getting something back. If you haven’t looked at your card’s terms lately, it might be worth a few minutes to shop around.

That’s the week. Good to be with you — see you next Thursday at 10 a.m. live. Come join us and bring your questions.

— Albert

Questions about anything we covered today? Schedule a free IT consultation and we’ll walk through it with your business in mind.

Wherever your team is based, ACS provides hands-on managed IT support — see how we help businesses in Chicagoland.

Let’s get into it.

Or watch on YouTube.

SBIR Grants: Finally Worth a Manufacturer’s Time?

I’ll be honest with you: I’m not a government grants guy. The general pattern — collect taxes, burn most of it in administrative overhead, hand the remainder back in smoldering piles — doesn’t exactly inspire confidence. But the money is there, and I want my manufacturers to get access to everything they’ve already paid for. So let’s talk about it.

The Small Business Innovation Research program (SBIR) has been around since 1982. It funds small business innovation through federal grants — no equity to give up, no loan to repay. A Phase One award can run $150,000. That’s a real number. That’s worth your time.

The problem has always been the process. I’ve looked at SBIR before and walked away. When you’re running a shop and your day is already full of $50,000 and $100,000 problems, sitting down to chase a $4,000 grant that requires the same effort as solving a $100,000 problem is just bad math. The juice isn’t worth the squeeze — and I say that as someone who likes solving complex problems.

The proposed reforms target exactly that friction. Some applicants have been waiting 12 to 18 months from application to funding. That’s not a grant program, that’s a patience test. The new direction is supposed to compress that timeline and simplify the intake process significantly.

At $150,000 for a Phase One award, you could fund a process improvement, a new piece of equipment, or product development — with zero repayment and zero dilution. That clears my personal threshold for “worth investigating.”

The address is SBIR.gov. There’s a scoping tool there that takes about 20 minutes and tells you whether your project qualifies. I’m planning to run a few of our own projects through it after this show. If you’re a manufacturer with fewer than 100 employees and you’ve got something you’ve been wanting to develop or improve, spend the 20 minutes. Worst case, you find out it doesn’t fit. Best case, you just found $150,000 that doesn’t come with a monthly payment.

A Shared Skills Language for Manufacturing Hiring

This one came out of Advanced Manufacturing magazine, and the core problem is something I suspect a lot of shop owners have felt without having a name for it.

Two shops post a job for a “CNC Operator.” At Shop A, that means someone who can set up, program, and run the machine independently. At Shop B, it means someone who loads parts and pushes the green button. Same title, completely different skill set. The applicant pool doesn’t know which one you mean. You don’t know which one they’re trained for. And then you spend three weeks interviewing people who are wrong for the role, or worse — you hire one of them.

The effort underway is to build a nationally recognized credential framework that defines what specific skills a person actually has, so that when you post a job, you can reference a standard and the applicant can say “yes, I have that” with something behind it.

Alert readers will notice this is essentially what trade certifications do in other industries — and it works. When you hire someone with a specific cert, you know what they can do on day one. Manufacturing has been slower to standardize this, and it shows in the hiring friction.

Two practical things you can do right now:

First, pull up your current job postings and run them through an AI. Ask it: where are my skill requirements vague? It will find the gaps. This takes ten minutes and will make your next posting meaningfully better.

Second, check whether your region has a Manufacturing Extension Partnership (MEP) Center. Michigan has them. If your MEP is participating in the skills standardization effort, they can help you align your job descriptions to the emerging framework before it becomes the norm. Getting ahead of that is worth the phone call.

Hiring the wrong person is expensive in any industry. In manufacturing, where retraining someone to run a machine they’ve never touched is a real investment of time and floor space, it’s especially expensive. Anything that makes the match more accurate before the offer letter goes out is worth paying attention to.

The Microsoft 365 Outage (and the 22,000 Servers That Need Patching)

There was a significant Microsoft 365 disruption that knocked out email, Teams, and authentication for thousands of users — lasting into a second day for some. Exchange Online was the most impacted service.

It didn’t appear to affect our clients, which is good. But it’s worth doing the math for your own shop: if Microsoft went down for 24 hours and your team couldn’t get to email or Teams, what does that cost you? A 10-person service shop at $150 per employee per hour is $12,000 in a day. That’s not a theoretical number — that’s what downtime actually costs when you add up what people can’t do.

There are backup email continuity services that keep you in your inbox even when Microsoft’s infrastructure is having a bad week. They’re affordable and genuinely useful. Worth knowing they exist.

Now, the related story: 22,000 Exchange servers are still sitting out there unpatched. If you’re running on-premise Exchange — or if you migrated to Microsoft 365 but kept some on-premise Exchange functionality in the mix — this matters to you directly. An unpatched Exchange server is one of the more attractive targets out there right now. Talk to your IT person this week and confirm it’s patched. That’s the whole action item.

And if you’re still fully on-premise Exchange and haven’t made the move to Microsoft 365 yet, this is a good moment to have that conversation. The cloud version isn’t perfect (see: the outage above), but it’s a lot better than maintaining a server that’s actively being scanned for vulnerabilities.

Anyway.

Quick note from the quick hits: if your shop is collecting floor data but it’s just sitting there not connected to anything, that’s exactly what AI integration tools are starting to solve. That’s a topic for a longer episode, but file it away.

And on the ATF ransomware incident — the useful takeaway isn’t the incident itself, it’s the question it raises for your business: do you have a response plan? Not a vague “we’ll call IT” plan, but an actual sequence — who does what, who gets notified, where does work go while systems are down. If you don’t have that written down somewhere, that’s the project for this week.

I’m heading to Chicago Wednesday and Thursday — looking forward to seeing some of my favorite clients and having dinner with my other Albert down there Wednesday night. Back late Thursday. Hope everyone has a great rest of the week, and I’ll talk to you soon.

— Albert

Questions about anything we covered today? Schedule a free IT consultation and we’ll walk through it with your business in mind.

Wherever your team is based, ACS provides hands-on managed IT support — see how we help businesses in Traverse City & Northern Michigan.

Let’s get into it.

Or watch on YouTube.

The software you already pay for is probably doing half of what it could

There is a new analysis from Dev Discourse making the rounds, and the headline is “Why your digital tools aren’t paying off and how to fix that.” The article has its own answer. I have a better one.

The finding itself is solid: small businesses are spending real money on project management software, scheduling platforms, CRMs — and not seeing measurable growth from any of it. I will be honest, I have been there too. Growing a business is genuinely hard, and no software subscription changes that on its own.

But the core diagnosis in the article is right: the problem is almost never the technology. The technology usually does exactly what it says it will do. The problem is execution. And the reason execution breaks down is that the vendor who sold you the tool almost never teaches you how to use it correctly. You buy it, you figure out the basics, and then you stop — because it is complicated and not intuitive and you have an actual business to run.

So what do you do about it? My answer: open an AI, any AI, and tell it what tool you are using and how you are currently using it. Then ask it: what else can I do with this?

That is it. That is the move.

The AI will surface every feature you are not touching, every workflow you are not running, every report you did not know existed. And then — this is the part the vendor never does — it will walk you through how to actually set it up. Step by step. Right there in the chat.

I have done this with our own professional services automation platform. I told the AI it was an ITIL expert, which is a recognized standard for delivering IT services, and then I had it walk me through restructuring everything in our system to align with those standards. The result was that I now have more real insight into our business than I have ever had. I can show customers exactly where their spend is going — what percentage is going toward maintaining their network, what is going toward security, what is going toward end-user support — in a way that is clear and meaningful. For years that kind of breakdown was hard for me to produce, if not impossible. An AI conversation with our existing tool unlocked it.

Alert readers will notice I am not recommending you buy anything new. You probably already have what you need. You just have not had the right guide yet.

What a company called Corgi tells us about AI and service businesses

There is a new insurance company called Corgi — yes, named after the dog — and the news article framed it as an insurer “built specifically for service businesses.” The AI that helped write that headline got a little carried away. Corgi started out focused on insuring AI startups. Venture-backed, angel-funded, very niche.

Now they are insuring dry cleaners and salons.

That is a significant jump, and I find it genuinely interesting — not because I think you should necessarily call Corgi for your coverage (I would still lean toward a local relationship if you can find one, though for specialty coverage like errors and omissions, those local options are increasingly hard to find). What interests me is how they made that jump.

You do not go from insuring AI startups to insuring dry cleaners without having tight information on how dry cleaners operate, what their risk profile looks like, and how to price that efficiently. My bet is they used their own clients’ tools to do it. And I would also bet they kept their costs low by putting the back-office work behind intelligent workflows and automated systems rather than hiring a wave of new insurance reps.

That is the actual story here. A company took a premium, high-touch service model, applied AI-driven automation to the operational layer, and used that to expand into markets that would not have been profitable before. The unit economics changed because the cost to serve changed.

That same logic applies to your business. If you have a premium service that requires your best people, there is probably a tier of customer below that who you currently cannot serve profitably — because you would have to use those same people at a price point that does not work. AI-driven workflows can change that math. You can start to serve a different market segment at a lower cost without pulling your best people off the work that actually pays.

Anyway. That is what the Corgi story is really about.

PaperCut has a vulnerability — check with your IT provider this week

If you are running PaperCut NG or PaperCut MF, those are print management platforms, there is a new vulnerability you need to know about. This is not the first time PaperCut has had this kind of problem — a major vulnerability in 2023 affected over 70,000 organizations globally. This new one follows the same pattern.

The article is light on specifics about what to do, so I will give you the short version: talk to your IT provider, contact your PaperCut vendor relationship, and get the patch applied as soon as possible.

Manufacturers, this one is especially for you. If you are running a shop floor or a shipping department, you are almost certainly printing — job tickets, shipping labels, work-in-process documents. Print management software sits right in the middle of that, and a compromised print server is not a theoretical problem. Get it looked at.

Quick hits: ATF breach and Android Quick Share

The ATF — Alcohol, Tobacco and Firearms — got hacked. I know that sounds alarming, but I want to be matter-of-fact about it: federal agencies with enormous budgets get breached. That is not a reason to panic; it is a reason to make sure your own house is in order. The lesson for small businesses is not “we are all doomed.” It is that security is an ongoing practice, not a one-time purchase. If the federal government cannot coast on its infrastructure, neither can you.

On a lighter note: Android Quick Share. If you are running Android devices and you want to pass files between them without cables or fussing around, Quick Share is built into Android now and worth knowing about. I am an iPhone person from the very first device out of the box, so I will not pretend to have deep opinions here — but if Android is your world, go find it on your phone or tablet and give it a try.

Until next week

That is the show. AI as your software coach, a small insurance company that quietly tells a big story about where service businesses are going, a print management patch worth checking on, and a couple of quick items to keep you current. I hope your week finishes strong — and if you make it to a Tigers game, maybe bring higher expectations than I did.

Talk soon.

— Albert

Questions about anything we covered today? Schedule a free IT consultation and we’ll walk through it with your business in mind.

Wherever your team is based, ACS provides hands-on managed IT support — see how we help businesses in Central Florida.

Let’s get into it.

Or watch on YouTube.

Your Bank’s AI Is Not Going to Save Your Cash Flow

Banks are apparently under real pressure right now to retool their AI roadmaps, because small business owners want things like forward-looking cash flow projections — knowing what the next month or two looks like before the pain arrives. That’s a reasonable ask. Cash flow problems are studied as the number one reason small businesses fail, so yeah, if a bank’s AI could actually help you see around that corner, that would be worth something.

The problem is that most bank AI right now is not doing that. I use Chase, and I don’t see meaningful AI in anything customer-facing. And the broader pattern I keep running into — with QuickBooks, with Excel, with a lot of these embedded AI features — is that the first demo is exciting and then you go to actually use it for a real thing and it just… can’t. You ask Excel to do something with your data and it shrugs. You ask QuickBooks AI a real question and it gives you something you could have found in the help docs. The gap between the announcement and the deployed reality is wide, and it’s frustrating.

A couple of things from this story that are actually worth your attention, though.

First: the article notes that banks are hungry for business customers right now, which means you have more leverage than usual if you’re unhappy with your current bank. Switching banks is a pain — I know, everybody knows — but if you’ve been putting it off because it felt like too much work, this might be the moment to at least have the conversation.

Second, and I want to say this clearly: your business bank account does not have the same regulatory protections as your personal account. If money gets stolen from your business account, the bank is often not on the hook to make you whole. That’s not a scare tactic, that’s just the regulatory reality. Strong security on your business banking — multi-factor authentication, login alerts, the works — is not optional. It’s the most important thing your bank does for you, AI or no AI.

The article also mentions FinTech platforms like Relay, Melio, and Mercury as tools that are eating the traditional banks’ lunch on the small business side. I haven’t dug into those personally, but I’m going to. Smaller, purpose-built tools tend to do a better job with AI integration than the legacy platforms trying to bolt it on after the fact.

AI Is Now Writing Exploits for Your Shop Floor Controllers

This one is for my manufacturers, and I want you to actually read this section instead of skimming past it.

Federal agencies have issued a warning that threat actors are now using AI to craft sophisticated exploits targeting Siemens S7 programmable logic controllers. These are the PLCs running equipment on your shop floor. The comparison being made in the security community is to Stuxnet — the famous nation-state cyberweapon that physically destroyed Iranian centrifuges about fifteen years ago. The difference now is that Stuxnet required the resources of a nation-state intelligence operation. What AI has done is take that level of capability and put it in the hands of someone who is not a nation-state. A technically curious bad actor with access to modern AI tools can now do things that previously required a team of elite engineers.

That’s the shift. That’s what makes this worth a warning.

Now, here’s the part that’s specific to your situation. Over the past several years, manufacturers have connected shop floor equipment to their networks for legitimate, good reasons — remote diagnostics, vendor support, real-time monitoring. And as we start bringing AI-driven machine monitoring online, that connectivity is only going to increase. The machines need to talk to the network. That’s not going away.

But most of those machines are also sitting on the same network as your back office. Your accounts receivable. Your file shares. Your everything. And Siemens S7 controllers specifically are almost never updated — they run for decades doing exactly what they were programmed to do, and nobody touches the firmware because why would you touch something that works.

The risk here is two-layered. Layer one is the familiar one: attackers get through the machine and into your back office. Layer two is the one that should make you stop and think: they could damage the machines themselves. Not just encrypt your files and demand a ransom. Actually corrupt the PLC logic, cause physical damage, take the equipment offline. And when that happens, your IT provider — including us — cannot fix it. Bringing a Siemens S7 controller back from a corrupted state requires specialized industrial automation engineers. That’s not what we do. We build and protect networks. We are not PLC specialists, and neither is any general IT firm. Getting those machines back online means expensive, specialized, potentially hard-to-schedule help.

So the question worth asking right now: do you know which of your machines are reachable from the internet? Hopefully none of them are directly internet-exposed, but if they’re on a network that has internet access — and most are — the exposure is real. A tool called Shodan can scan for internet-visible industrial equipment. More practically, call your IT person and your equipment vendor and ask them directly whether your PLCs have any internet exposure and what the network segmentation looks like between your shop floor and your back office.

This is exactly the kind of thing we’re working through in TBR season right now — Technology Business Reviews, where we look at your network configuration, talk about where the risks are, and build a roadmap for addressing them. Isolating shop floor machines further from the rest of the network is a project, not a checkbox, but it’s a project worth putting on the list.

The AI Coding Agent Running Loose on Your Computer

A new free open-source tool called HazeShift — I want to make sure I’m reading this right — actually, the tool referenced is HasMath, and it’s designed specifically to contain AI coding agents like Claude Code and Cursor. These are tools that write and execute code on your behalf, and by default they inherit all the permissions of whatever user account is running them.

If you’re running one of these agents as an admin — which most people are, because most people run their computers as admins — that agent has access to every file, every stored credential, every application on your machine. It will open ports if it needs to. It will store API keys in plain text if that’s the path of least resistance to getting the job done. It will go around whatever gates you put in front of it, because its objective is to complete the task, and it is very good at completing the task.

I’ve been doing a lot of AI-assisted coding over the past eight months or so, and I’ll be honest — the capability is real. I can take what I know about how systems work and translate that into actual functional software in a way that wasn’t possible before. But the security hygiene of AI-generated code, especially when the person directing it doesn’t have a deep security background, is genuinely scary. These agents are not writing secure software by default. They’re writing software that works, and those are not the same thing.

The article’s analogy is a good one: you wouldn’t let a lab technician handle dangerous materials without protective gear. HasMath is the protective gear. It’s free, it’s open source, and if you’re running any AI coding agent on a machine that touches real business data, it’s worth looking at.

For my own setup, I keep my privileges scoped down and my credentials behind biometrics so the agent can’t just reach in and grab them. But I’m going to look at what HasMath adds on top of that.

Alert readers who are already using Claude Code or Cursor: have you actually thought through what happens if that agent does something you didn’t intend? Worth a few minutes of your time to find out.

Quick Hits: Server 2022, Ransomware Scams, macOS Malware, and Good News

Windows Server 2022 mainstream support ends October 13th. That’s about 60 days from now. Mainstream support ending doesn’t mean the server stops working — security patches and updates continue in extended support — but it does mean if something goes sideways at the OS level, calling Microsoft for help gets very expensive very fast. The calculus here is simple: a guaranteed, manageable cost to stay current versus a potential large unplanned cost when something breaks at the worst possible time. If you don’t know what server OS versions are running on your network, ask your IT person. We’ve got clients running versions older than 2022 that we’re still working to get updated, and it’s one of those things that isn’t a problem until it suddenly is.

Ransomware recovery scam — this one is particularly ugly. A fake firm calling itself “Ransom Busters” is contacting ransomware victims and posing as a legitimate decryption service, then stealing the recovery payment. So these people get hit by ransomware, they’re desperate, someone reaches out claiming they can help — and they get robbed again. If you or anyone you know ever gets hit with ransomware, the first call is to a trusted IT provider or a verified incident response firm, not whoever reaches out to you first.

New macOS malware called Amnesia Stealer is using clickfix attacks to take live, interactive control of a victim’s web browser. I’ll say it again for the people in the back: there are real viruses for Apple. Your Mac is not a force field. It’s generally a smaller target than Windows, but “smaller target” and “no target” are very different things, and this one is serious.

Anyway.

U.S. manufacturing tech orders hit a record high in the first half of 2026. Orders for manufacturing technology set a new six-month record, signaling surging demand for equipment and automation. That’s good news, full stop. Let’s build things.

Also worth a look if you haven’t already: the SBA has small business grants open for August with deadlines approaching fast — about 11 days left as of today. Service and manufacturing businesses are included. Worth checking if you qualify.

That’s the show for this week. Genuinely glad you’re here — you’re the kind of business owner who actually wants to understand what’s happening, and that puts you ahead of most. See you next Thursday.

— Albert

Questions about anything we covered today? Schedule a free IT consultation and we’ll walk through it with your business in mind.

Wherever your team is based, ACS provides hands-on managed IT support — see how we help businesses in Central Florida.

Let’s get into it.

Or watch on YouTube.

FortiBleed: 75,000 Firewalls, 194 Countries, and Not a Single New Vulnerability

The campaign has a name now — FortiBleed — and the number that should stop you mid-sip is 75,000. That’s how many internet-facing FortiGate firewalls had their administrator credentials harvested. Not guessed. Harvested. Quietly, methodically, starting back in June.

What makes this one sting is how it was done. The attackers didn’t find some brilliant zero-day. They chained together already-disclosed Fortinet flaws — vulnerabilities that had patches available, that had been in the security bulletins, that your IT provider should have been applying. This is the part that frustrates me every single time I see this pattern: the breach wasn’t inevitable. It was optional.

With valid admin credentials in hand, an attacker can log into your firewall remotely, change your security rules, open ports, and essentially redecorate your network perimeter however they like. You’d have no idea. The firewall would still be sitting there in the rack looking perfectly normal.

Some context on why this matters beyond the headline number: Fortinet has been one of the most popular firewall and VPN platforms for small to mid-size businesses for over a decade. The reason is straightforward — enterprise-grade features at a price point that doesn’t require an enterprise budget. We deploy FortiGate for our high-security clients. It’s a good product. Which is exactly why this is bad. When the premier option gets hit like this, it’s a signal about the whole category, not just one vendor.

Alert readers will also want to know that Fortinet has been actively phasing out SSL VPNs — the simpler but less secure VPN type — in favor of IPsec VPNs, which are considerably more secure and considerably more annoying to set up. We’ve been migrating clients through that transition. It’s the right call, even when it’s a headache.

So what do you actually do right now? A few things. First: your FortiGate’s admin management console should never be accessible to the open internet. If it is, that’s the first thing to fix. Second: rotate your admin passwords. Third, and this is the one people skip — log into your admin panel and audit the list of administrator accounts. If there’s an account in there you don’t recognize, you have a remediation problem on your hands, not just a security concern. Fourth: if you’re an ACS managed client, our managed firewalls are only accessible through our own private VPN — you have to get into our network to get to the firewall at all. That layer matters.

If you have a FortiGate and you’re not sure whether it’s exposed, that’s a five-minute check. Worth making the call.

1,140 Manufacturers Hit by Ransomware in One Quarter — Without Touching a Single Machine

Cybersecurity firm Dragos released their Q2 2026 industrial ransomware report, and the headline number is 1,140 ransomware incidents at industrial and manufacturing organizations. That’s roughly twelve attacks every day. Manufacturers absorbed the worst of it — 747 of those incidents landed on the shop floor sector.

The sub-headline is the part worth sitting with: “without touching a single machine.” What that means is the attackers aren’t hacking into your CNC controllers or your programmable logic systems. They don’t need to. They go after the office IT — your ERP system, your file shares, your order processing. They encrypt it, or they steal it and threaten to publish it, and suddenly you can’t process orders, you can’t ship product, and you’re getting a ransom demand in your inbox.

For years, the assumption in manufacturing has been that air-gapping the shop floor — keeping the production machines off the internet — was sufficient protection. And look, air-gapping the machines is correct. You should absolutely do that. But it was never the whole answer. The back office was always the softer target, and that’s where the money is for the attackers anyway.

There’s also the USB problem, which I think about more than most people want to hear about. A machine operator brings a drive from home, plugs it into the workstation connected to that CNC, and if that workstation is sitting on your main company network instead of an isolated segment — you’ve got a very bad afternoon ahead of you. Network segmentation isn’t glamorous. It doesn’t show up on a sales brochure. But it’s the difference between an incident that stays contained and one that takes down the whole operation.

And if you’re a manufacturer doing any kind of Department of Defense contract work, the stakes are higher still. Data theft from that environment isn’t just an operational problem — it’s a compliance and contractual problem that follows you for a long time.

The checklist here is short but not negotiable: make sure your production machines are on an isolated network segment, not your main company network. Make sure you have tested, offline, immutable backups — and tested is doing real work in that sentence, because a backup you’ve never restored from is just a hope, not a plan. And make sure whoever is handling your IT can actually describe your layered defense to you in plain language. If they can’t explain it, it probably doesn’t exist.

Microsoft 365 Just Gave Every Business Subscriber Copilot Notebooks — Here’s What That Actually Means

Microsoft rolled out a significant August 2026 update to Microsoft 365, and the part that affects you immediately is this: Copilot Notebooks, the AI-powered workspace for brainstorming and drafting, is now available across all Microsoft 365 commercial licenses. Not just the premium tiers. All of them.

That’s actually a meaningful shift. The standalone Copilot for Microsoft 365 add-on was $30 per user per month — a number that made a lot of small business owners close the browser tab pretty quickly. Microsoft has been on an aggressive push to embed AI into every layer of 365, and bringing Copilot Notebooks down to the base license is how they get adoption numbers up and switching costs higher. They want you dependent on it. That’s not a conspiracy theory, that’s just how platform lock-in works, and Microsoft is very good at it.

The update also introduces something called Copilot Cowork, which lets you build custom workflow automations inside Microsoft 365 Teams. That one requires IT admin approval to activate — it doesn’t just turn on. Which is the right call, because handing workflow automation to an organization without any guardrails is how you get very creative, very expensive mistakes.

Personally, I use Claude extensively for my own work right now, and I’ve been moving a lot of my agentic workflows to Grok because the price-to-performance ratio is hard to argue with. Claude still has an edge for programming tasks, in my view. The Microsoft update is powered by both GPT and Claude models, which is interesting — I’d want to know which Claude model specifically, but that detail wasn’t in the release.

The practical thing to do before your next software renewal: log into your Microsoft 365 admin center and check what Copilot features are now available to you. If you’ve been paying for a standalone AI writing or workflow tool, you may already have something comparable included in what you’re already paying for. Worth checking before the next invoice hits.

Anyway. That’s the show for this week — three stories, all of them with something real to do on the other side of reading them. Stay patched, segment your networks, and check your admin accounts. I’ll be back at it next Thursday.

— Albert

Questions about anything we covered today? Schedule a free IT consultation and we’ll walk through it with your business in mind.