Please Note: Once again this is a particularly long post, so subscribers will find the emailed version of this issue cut off at the end. Please go to the web version for the complete version.

If You Find AgenticMSP Valuable: Please recommend AgenticMSP to your friends and colleagues. Also, please offer up your comments, observations, questions, or any other responses in the chat below. Let’s get a hearty conversation going!

AgenticMSP’s mission is to bring you the resources that will most help you build your AI business. Usually, we feature experts who bring you deeper insight into AI. Matthew Berman certainly does that, with a great hands-on feel to what he tells us. But he also brings a love of entrepreneurialism that just radiates out of him. You feel like your listening to someone who is truly one of us. Here’s more on Matthew.

Read more

Please Note: Once again this is a particularly long post, so subscribers will find the emailed version of this issue cut off at the end. Please go to the web version for the complete version.

If You Find AgenticMSP Valuable: Please recommend AgenticMSP to your friends and colleagues. Also, please offer up your comments, observations, questions, or any other responses in the chat below. Let’s get a hearty conversation going!

Full transparency up front: Like most sensible Americans I am deeply resistant to anything that involves Elon Musk in any way. He has proven himself a detestable, self-aggrandizing user. I say this just to establish my perspective from the start. Nobody needs to agree or disagree with me. I celebrated Cheryl Crow’s publicly disposing of her Tesla. I’m frustrated that I need to find myself on the sewer that is X just to keep up with what the major movers and shakers in AI are saying.

That said, I felt compelled to weigh in here at AgenticMSP on SpaceXai’s new entry into the AI market, Grok Bot.

“A New Kind of Colleague”

Read more

Please Note: Once again this is a particularly long post, so subscribers will find the emailed version of this issue cut off at the end. Please go to the web version for the complete version.

If You Find AgenticMSP Valuable: Please recommend AgenticMSP to your friends and colleagues. Also, please offer up your comments, observations, questions, or any other responses in the chat below. Let’s get a hearty conversation going!

This entire issue is dedicated to all the major changes happening at all the major players all the time. My spidey-sense is tingling and telling me huge change is in the air. Please share your thoughts and observations with me at hmc@howardmcohen.com or in the chat below. AI changed our MSP world, and now its world is changing too.

The AI Frontier Is Unstable - Be Prepared

Dizzy? In the past ten days, big changes at five frontier labs when we thought there were only three. Two leadership departures, one model paused out of an abundance of caution, one letter from a sitting U.S. Senator threatening Congressional action, and a reported two trillion dollars of IPO in the pipeline.

Dizzy. Clearly not a slow month for this industry. What do you need to do to be prepared?

Asking the Wrong Question

Read more

Please Note: Once again this is a particularly long post, so subscribers will find the emailed version of this issue cut off at the end. Please go to the web version for the complete version.

If You Find AgenticMSP Valuable: Please recommend AgenticMSP to your friends and colleagues. Also, please offer up your comments, observations, questions, or any other responses in the chat below. Let’s get a hearty conversation going!

A small change in this issue. We’re including the ranking score for each article. It would be great if you would share your thoughts about how accurate the scores are once you’ve read the stories. Email me at hmc@howardmcohen.com or add to the chat.

In the first summary in today’s MSP AI News, Barracuda discusses how AI and automation are transforming the MSP business model, and how they’re supporting MSPs in taking fullest advantage of it. Their core message is not really new. It’s something many MSPs have spent many years striving for.

Move yourself out from behind the issues in reactive mode, and get their in front of customer’s challenges by enabling yourself to be truly proactive.

AI is Your Answer

Barracuda’s Red Team just showed why. In a controlled simulation, one AI-crafted phishing email went from inbox to full account compromise in five minutes. Five minutes!!!

Read more

Please Note: Once again this is a particularly long post, so subscribers will find the emailed version of this issue cut off at the end. Please go to the web version for the complete version.

If You Find AgenticMSP Valuable: Please recommend AgenticMSP to your friends and colleagues. Also, please offer up your comments, observations, questions, or any other responses in the chat below. Let’s get a hearty conversation going!

Apologies for no AgenticMSP on Tuesday. I was attending ChannelCon and getting blown away by how very much they’re doing to join AgenticMSP in promoting, informing, guiding, and propelling MSPs and ITSPs into a better AI future. As you’ll see in today’s issue I was very, very impressed and pleased to see our newly-focused association in action!

You have to credit GTIA CEO Dan Wensley for candor when sharing a statistic we’ve all heard quoted repeatedly lately, that 97% of MSPs already claim to be using AI internally, but only 20% have real governance around it, and only 25% have turned it into a formal revenue-driving strategy. He clearly but gently expressed his skepticism of the responses in the survey. Shades of Jon Lovitz crooning “Suuuuuuuuuuuure we use AI internally…” When I asked GTIA VP Research Carolyn April how many of those surveyed answered honestly, all she did was smile. Dan didn’t shy away from the obvious insincerity.

Read more

Please Note: Once again this is a particularly long post, so subscribers will find the emailed version of this issue cut off at the end. Please go to the web version for the complete version.

If You Find AgenticMSP Valuable: Please recommend AgenticMSP to your friends and colleagues. Also, please offer up your comments, observations, questions, or any other responses in the chat below. Let’s get a hearty conversation going!

Hope to see many of you this coming week at ChannelCON 2026 in San Diego. GTIA is definitely CompTIA completely reborn, revitalized, and more relevant than ever. While the sessions look to be as fantastic as usual, I still find connecting with all of you to be the most valuable opportunity ChannelCON offers! Please connect with me on the show networking system. See you there!!!

There’s someone I know who can tell you, in sequence, every stop in the NYC Subway system along every line. For those who’ve never been on the subway, that’s a lot of stations! But if you ask him how he’s going to move a bunch of boxes from one place to another he’ll scratch his head.

Another friend remembered the IP addresses of literally every device installed in every one of her client’s datacenters. Okay, so it’s the IPv4 addresses which are only four groups of up to three digits but imagine that across the many IP addresses in each of several dozen devices and then across dozens of clients. Amazing! But she couldn’t keep a business running for long.

When They Get More Intelligent Than Us, They’ll Wipe Us Out!!!

Full transparency, it was the frustration I’ve experienced over the past few weeks developing some internal systems with Claude that motivated me to write this article.

I keep finding myself confronted with the same question you probably get endlessly from your customers. When will AI get more intelligent than we humans are? When they do, they’re going to wipe out the human race, right? Like when Skynet becomes aware and we get the Terminator. How do we stop it?

Can we stop it?

Oh, nooooooooooooooooooooooooooooooooooooooooo

Artificial Intelligence, Not Artificial Smarts

Read more

Please Note: Once again this is a particularly long post, so subscribers will find the emailed version of this issue cut off at the end. Please go to the web version for the complete version.

If You Find AgenticMSP Valuable: Please recommend AgenticMSP to your friends and colleagues. Also, please offer up your comments, observations, questions, or any other responses in the chat below. Let’s get a hearty conversation going!

In one of the first issues of AgenticMSP I told the story of how Satya Nadella stood on stage at Ignite in Chicago in October 2024 and welcomed us all to the arrival of “the agentic age.” It sounded like a majestic manifesto filled with great vision that excited everyone in the audience.

It was actually a press release for something Microsoft didn’t own yet, dressed up to look like a breakthrough.

I’ve watched Microsoft run this same play since 1981, when I sold my first Microsoft product. The pattern never changes: announce the vision before you’ve built the substance, let the trade press do the marketing for you, and backfill the technology later if the market forces your hand. Think “BPOS”, the Business Productivity Office Suite which eventually became Office 365 and then Microsoft 365. Think Lync, which survived through many iterations of Microsoft Communications Server to finally become Teams. Think SharePoint, which began as about a dozen different products that got slammed together and marketed as one. Dynamics CRM went the exact same way and offers the same feature bloat.

Ignite 2024 was that exact pattern in its purest form, played out for the age of AI.

What Was Copilot Really?

Read more

Please Note: Once again this is a particularly long post, so subscribers will find the emailed version of this issue cut off at the end. Please go to the web version for the complete version.

If You Find AgenticMSP Valuable: Please recommend AgenticMSP to your friends and colleagues. Also, please offer up your comments, observations, questions, or any other responses in the chat below. Let’s get a hearty conversation going!

The bottom-line.

However you think about your customer’s wants and needs, it all ultimately leads back to their bottom-line. They invest in your IT solutions to enjoy a return on their investment.

Only Two Ways

There are only two components contributing to profit.

· Increase revenue

· Reduce costs

It really is that simple. These are the only two ways you can impact your customer’s, or even your own, profit. You may sell many marketing automation solutions, including AI-driven ones, that help them drive more revenue, and they really appreciate those.

But if you don’t also reduce their operating costs, all that revenue may be chewed up before it can ever show up on that bottom line.

Manage to Reduce

Reducing costs is a result that comes from managing them closely. The good news for MSPs is that this means there are many, many opportunities for you to help customers manage and reduce their operating costs.

Read more

Please Note: Once again this is a particularly long post, so subscribers will find the emailed version of this issue cut off at the end. Please go to the web version for the complete version.

If You Find AgenticMSP Valuable: Please recommend AgenticMSP to your friends and colleagues. Also, please offer up your comments, observations, questions, or any other responses in the chat below. Let’s get a hearty conversation going!

A Quick Note on Open-Source Models

The introduction of new open weight AI models like Moonshot’s Kimi K3 are closing the performance gap between the open and the frontier models. The costs appear at first to be significantly lower, but further exploration has demonstrated that they often use twice as many tokens to accomplish tasks. This often balances out the cost difference, and the frontier model makers continue to be superior at actually delivering the services consistently and securely.

By contrast, many experts anticipate the open models to better enable cybercriminals to attack other users with fewer guardrails and other limitations in place to restrict them. Before recommending these growing entries to your customers be sure to research thoroughly to determine if you are really saving them any money in the long term.

Controlling AI Spend

Here’s a rundown of the real levers MSPs have available to them to help clients control AI spend. These are organized roughly from “quick wins you can implement this week” to “strategic plays that take longer but ultimately bring greater pay-off.”

Read more

Please Note: Once again this is a particularly long post, so subscribers will find the emailed version of this issue cut off at the end. Please go to the web version for the complete version.

If You Find AgenticMSP Valuable: Please recommend AgenticMSP to your friends and colleagues. Also, please offer up your comments, observations, questions, or any other responses in the chat below. Let’s get a hearty conversation going!

“The token subsidy era is just about over…”

“The AI bubble’s about to burst…”

“You can’t just use one LLM for everything anymore…”

The list of warnings goes on, and the fact is that they’re all correct to some degree or another. The question isn’t really “what’s going to happen next.” It’s really more, “what are MSPs going to do about it?”

Read more

Please Note: Once again this is a particularly long post, so subscribers will find the emailed version of this issue cut off at the end. Please go to the web version for the complete version.

If You Find AgenticMSP Valuable: Please recommend AgenticMSP to your friends and colleagues. Also, please offer up your comments, observations, questions, or any other responses in the chat below. Let’s get a hearty conversation going!

The ground moved beneath our feet three times this year, and the year is barely half-way-through!

Compute

Let’s start with compute. In April, the going hourly rate for Nvidia Blackwell capacity jumped 48% in about ten weeks from $2.75 to $4.08, per the Ornn Compute Price Index reporting picked up by tech press. That wasn’t a blip. No surprise AWS and Azure responded by moving Blackwell access into priority tiers, putting existing enterprise-agreement customers and the big frontier labs ahead of everyone else in line.

If your client was an SMB or mid-market shop self-serving on one of those hyperscalers, they’re now behind that line, whether they know it or not. And they had no way of anticipating that any of it was going to happen until it did, very expensively.

Export Policy

With government decisions being driven by how much money could be extracted from the situation, export policy moved just as fast, but in opposite directions, and it happened twice.

In January, the Bureau of Industry and Security eased its posture on H200-class chip exports to China, shifting from a presumption of denial to case-by-case review. By June, Commerce had reversed course on a related loophole, affirming that its licensing rules apply to any Chinese-headquartered company anywhere in the world, not just inside China. Two rulings, five months apart, both fully legal, both changing what a client’s supply chain or data-processing arrangement was allowed to touch.

Imagine the government doing anything at the scale of AI growth… on a case-by-case basis. Can you say “Dead Stop?” I knew ya could.

Large Language Models (LLM)

Even the model vendors aren’t immune, including the one I always use for research. In June, Anthropic suspended access to two newly released frontier models, Fable and Mythos, to comply with a Commerce Department export control. Commerce lifted the control at the end of the month, and Anthropic restored access on July 1. Three weeks, start to finish, for models that had existed for three days before the suspension hit. And, according to a story featured in this issue, the conditional availability of those models changes at the end of this week. Unless it extends yet again. Which it could. But may not.

Planning on Thin Ice, Barefoot

Read more

Please Note: Once again this is a particularly long post, so subscribers will find the emailed version of this issue cut off at the end. Please go to the web version for the complete version.

If You Find AgenticMSP Valuable: Please recommend AgenticMSP to your friends and colleagues. Also, please offer up your comments, observations, questions, or any other responses in the chat below. Let’s get a hearty conversation going!

“If you can’t do it yourself, I don’t know how you do it for other people.”

Logically CEO Joshua Skeens asked himself that question three years ago. Based on the answer, Logically decided to go all-in on AI and automation. Every employee, monthly training, full commitment.

And for the first four months? Nothing.

Skeens described it perfectly: “We tried to cure world hunger out of the gate. Then four months later we’re all looking around going, did we even do anything?”

Sound familiar? It should. It’s the exact same mistake most MSPs are making right now, just compressed into a single company’s timeline instead of being spread across an entire industry’s worth of failed pilots.

So Skeens pulled back. Created based on his observation that removing friction beats chasing shiny objects, he already ran a “friction team” internally. This is an innovative group whose job is to find and remove friction wherever it slows the business down. Now, he pointed that same team at AI.

Read more

Please Note: Once again this is a particularly long post, so subscribers will find the emailed version of this issue cut off at the end. Please go to the web version for the complete version.

If You Find AgenticMSP Valuable: Please recommend AgenticMSP to your friends and colleagues. Also, please offer up your comments, observations, questions, or any other responses in the chat below. Let’s get a hearty conversation going!

Why do customers invest in technology in the first place? And why do they invest in your services?

The answers are far simpler than we may make them.

Why Customers Invest in Technology

To make a return on their investment (ROI).

Why Customers Invest in Your Technology Services

They trust that you understand and will help them achieve that ROI.

We Could End This Right Here

If you already embrace these two answers you can probably just stop reading right here and jump down to today’s top MSP AI News.

I’m actually quite serious about that. The most successful MSPs I have encountered over the years are the ones who are driven by customer success mainly defined as customer ROI.

There was a terrific writer named Mack Hanan who coined the term “consultative selling” which many of us have embraced over the years. In his book, Consultative Selling: The Hanan Formula for High-Margin Sales at High Levels, Hanan explains the difference between being your customer’s vendor and being their partner, explaining, “vendors ask for an order, which represents a cost to their customers, while consultants offer improved profits, which represent a gain. That’s the whole distinction in one sentence; cost versus gain.”

Net net, when a customer knows your objective is to increase their profit, they’re more than happy to help you increase yours.

What Does This Mean to AI

“Companies Traded People for Tokens. The Returns Haven’t Shown Up” is the title of the first article in today’s MSP AI News below. Your customers are bombarded with news about the billions and trillions frontier AI model makers are investing in new data centers, chips, development, energy, and more. They’re now hearing more and more about how those people have finally woken up and asked themselves the key question: “How are we going to recover all these costs and show a profit?”

And part of the reason this is happening is their lack of recognition as to why customers invest in technology. More and more reports are coming out about how companies are not seeing return on their AI investments. That first article points out that some of these companies have realized their operating costs were lower when they had people doing what they now use AI for. Using AI simply costs more!

So, while they should be asking themselves, “How do we help customers achieve increased ROI?” They’re busy asking, “How do we increase our own profits?”

Many are predicting that we’re going to see the answers very soon, and it is not going to be pretty. Expect to see consolidation, acquisition, perhaps even some closures and ugliness.

What Does This Mean to Your Customers?

Read more

Please Note: Once again this is a particularly long post, so subscribers will find the emailed version of this issue cut off at the end. Please go to the web version for the complete version.

If You Find AgenticMSP Valuable: Please recommend AgenticMSP to your friends and colleagues. Also, please offer up your comments, observations, questions, or any other responses in the chat below. Let’s get a hearty conversation going!

How many brains do you have? How many do you need?

When asked why he didn’t know the exact speed of sound offhand, Albert Einstein famously answered, “I don’t know. I don’t burden my memory with such facts that I can easily find in any textbook. I do not carry such information in my mind since it is readily available in books.”

Second Brain

Tiago Forte is the guy who turned “second brain” from a vague productivity metaphor into an actual methodology. He’s a productivity consultant and writer, founder of Forte Labs, and he built his reputation teaching people how to offload thinking to an external system instead of trying to hold everything in their head.

The philosophy underneath both is that human brains are bad at storage and good at thinking, so a second brain’s job is to take the storage burden off you entirely, freeing you up for the connections and judgment only you can make.

That’s the same instinct animating Nate Jones’s Open Brain work. Jones cites Forte directly as the intellectual ancestor of what he’s building, though Forte’s system was designed for humans doing the organizing by hand, and Jones’s bet is that AI can now do the classifying and routing automatically instead.

Building an Open Brain

Read more

Please Note: Once again this is a particularly long post, so subscribers will find the emailed version of this issue cut off at the end. Please go to the web version for the complete version.

If You Find AgenticMSP Valuable: Please recommend AgenticMSP to your friends and colleagues. Also, please offer up your comments, observations, questions, or any other responses in the chat below. Let’s get a hearty conversation going!

He? She? Or It?

When you’re referring to your favorite AI model, do you refer to that model as “he”, “she”, or “it”? The answer may be more important to your customers than you think, and they may require your informed advice.

TeamMate or Friend

I’ll be totally transparent here and tell you up front that I like Claude. I mean, I really do like Claude. When it became possible to have two-way verbal communication I gave mine a female voice and named her Claudia.

I enjoy my conversations with Claudia, much for the same reason I enjoy conversations with many other humans. She provides a wealth of answers I need to questions I have about various writing projects I’m working on, and she provides those answers with rich context. She connects the facts for me and helps me choose my direction. She’s pleasant enough in her presentation, but what’s most important to me is how helpful she is.

Where’s the downside?

Early in my career many people starting referring to me as a “wizard” because I appeared to know so much about so many technology issues. Often, I found myself letting them know my “secret.” It wasn’t so much what I knew as it was who I knew. I knew a lot of people who each had different expertise in specific areas, and I knew which to turn to for advice and guidance on any of those topics. This constantly put me in touch with a wide range of truly interesting people.

Today, however, if I have a question I simply turn to Claudia for the answers. It’s become reflexive. Question? Claudia. This has served to very efficiently remove me from regular discourse with other people. As I realized this separation I became so alarmed I created a little routing for myself that I call “KIT” which stands for “Keep In Touch.” I have an appointment somewhere in the future on my calendar for literally everyone I want to keep in touch with. On that day I make sure to reach out, even just to say hello and check-in.

To me, there’s a danger in allowing AI to replace other people in your life.

What Do the “Experts” Say?

Read more

Please Note: Once again this is a particularly long post, so subscribers will find the emailed version of this issue cut off at the end. Please go to the web version for the complete version.

If You Find AgenticMSP Valuable: Please recommend AgenticMSP to your friends and colleagues. Also, please offer up your comments, observations, questions, or any other responses in the chat below. Let’s get a hearty conversation going!

In February 2026, the frontier AI lab company most well-known and trusted for making safety its defining brand loosened its Responsible Scaling Policy under simultaneous pressure from the Pentagon and market competition.

In early June, Anthropic called for a coordinated, verifiable pause in advanced AI development, warning that systems may soon improve themselves faster than human society can manage the consequences.

Just weeks ago, Anthropic disclosed that accounts linked to Alibaba’s Qwen lab had created nearly 25,000 fraudulent identities to run 28.8 million exchanges with Claude in what the company is calling the largest known capability-extraction attack it has ever identified.

All these stories converge on a major concern most of your customers share. Your customers read this in CNN, Al Jazeera, the Financial Times, and most other news outlets. It clearly gives them pause. And well it should!

So, when a client asks you, “Is this AI stuff actually safe?”, the answer you give will matter more than any demo you can run. It needs to be a solid answer, solidly backed with provable strategies and solutions.

Answer the Real Question, Not the Hype-Driven Ones

Read more

Please Note: Once again this is a particularly long post, so subscribers will find the emailed version of this issue cut off at the end. Please go to the web version for the complete version.

If You Find AgenticMSP Valuable: Please recommend AgenticMSP to your friends and colleagues. Also, please offer up your comments, observations, questions, or any other responses in the chat below. Let’s get a hearty conversation going!

Saving customers’ money by reducing their capex and opex costs has always been a hallmark of channel services. The best IT engineers distinguish themselves by talking strategically with customers about how to license or operate IT resources in the most cost-efficient way possible. This is the complete approach to service IT needs from both the technology and business operations perspective.

Since the earliest days of the channel, we’ve been integrators, creating advantage by combining components from various vendors to assemble solutions that are far less expensive than a single-vendor system. It began with swapping out hardware like storage drives and memory from the OEM with third-party units. Cloud had us swapping out the customer’s entire data center for a less-expensive remote datacenter operation.

The MSP as AI Integrator

Today’s primary channel partner, you the MSP, will continue that fine tradition of supplying superior integration services even more powerfully than before now that several things have occurred that come together to create new customer needs.

The first is the reality the frontier large language model (LLM) labs are finally confronting that their economics of running AI is far too high to be supported by the revenue they’re generating. OpenAI is said to be spending $1.35 for every $1 of revenue they earn. That’s called “a losing proposition.”

But experienced MSPs know that nothing remains the vendor’s problem for long, especially when it comes to the cost/price calculation. Their only solution is to pass that excess down to the channel to deal with. They raise their prices, then so must we. That’s called them having control over us. That, my friends, is not good.

In the past, our strategy for escaping those increased costs was to turn to “third-party” components not manufactured or branded by the OEMs.

Now, in 2026, the same is happening with AI.

Read more

Please Note: Once again this is a particularly long post, so subscribers will find the emailed version of this issue cut off at the end. Please go to the web version for the complete version.

If You Find AgenticMSP Valuable: Please recommend AgenticMSP to your friends and colleagues. Also, please offer up your comments, observations, questions, or any other responses in the chat below. Let’s get a hearty conversation going!

FUD

Your biggest obstacles to selling AI solutions to your customers are the classic IBM model: “Fear, Uncertainly, and Doubt.” FUD.

Your customers are reading and hearing how AI may come and take all our jobs. How it may take over everything. How it may cause extinction events. The Godfather of AI, Geoff Hinton, has contributed more to those fears than you can imagine.

But when you’re trying to sell AI solutions to your customer, those fanciful fears may be all you face. How do you overcome them? Do you start disclaiming all those stories? Do you try to present some facts you’ve found that disprove them?

Does your customer listen?

At that moment, please remember: “Never Argue with Your Customer. They’re Always Right!”

You’re unlikely to win anyway. The public arguments have been raging for years now. You won’t settle them during a sales call. Instead…

Read more

Please Note: Once again this is a particularly long post, so subscribers will find the emailed version of this issue cut off at the end. Please go to the web version for the complete version.

If You Find AgenticMSP Valuable: Please recommend AgenticMSP to your friends and colleagues. Also, please offer up your comments, observations, questions, or any other responses in the chat below. Let’s get a hearty conversation going!

Anthropic said “no” to Secretary of Defense Pete Hegseth, that he couldn’t use Claude to kill or surveille anyone. Anyone who thinks that has nothing to do with the government’s order to shut down Mythos & Fable isn’t thinking very hard. Political revenge is definitely at play here. It’s their jam!

But much of the coverage and the public conversation about Fable and Mythos has been almost entirely focused on the wrong thing.

Everyone is arguing about whether the new Anthropic large language model (LLM) is just powerful, whether that makes it too dangerous, whether the jailbreak was real, whether the guardrails hold, whether there really are any guardrails. That’s understandable. The model is the visible, identifiable thing. In fact, it’s the thing we’ve all been focused on all along. We all recognize Claude, ChatGPT, and Gemini. But analysts with technical depth are pointing at something more fundamental than the LLM itself, and this has direct consequences for every MSP deploying AI right now. And every customer you’re deploying it to.

Read more

Please Note: Once again this is a particularly long post, so subscrbers will find the emailed version of this issue cut off at the end. Please go to the web version for the complete version.

If You Find AgenticMSP Valuable: Please recommend AgenticMSP to your friends and colleagues. Also, please offer up your comments, observations, questions, or any other responses in the chat below. Let’s get a hearty conversation going!

“I don’t prompt Claude anymore. I have loops that are running. They’re the ones prompting Claude and figuring out what to do. My job is to write loops.”

- Boris Cherny, June 2, 2026

“You shouldn’t be prompting coding agents anymore. You should be designing loops that prompt your agents.”

- @steipete, June 7, 2026

What are these guys talking about?

Who ARE these guys?

Why should an MSP care?

Boris Cherny developed Claude Code and Cowork and works at Anthropic.

Peter Steinberger created OpenClaw and now works at OpenAI.

Few guys know more about agentic AI than these guys. If you’re an MSP intending to multiply your fortune with AI these are two great guys to follow.

And these guys are talking about the Loop.

What Is the Loop?

When you first started working with Generative AI (GenAI) you asked the chatbot questions and it provided answers and that’s where the entire process ended.

What makes an AI Agent “agentic” beyond “generative” is the loop, a fundamental execution pattern that makes an AI agent an agent rather than a chatbot. Every agent, regardless of framework, runs on the same basic cycle:

Perceive → Reason → Plan → Act → Observe → Repeat

In this process, the agent intakes its environment, reasons about what to do next, selects a tool or action, executes it, receives feedback from the result, and begins again. It keeps looping until the task is complete or until it otherwise decides it’s done. Note that it doesn’t return to the human operator for feedback. Rather, it evaluates the result of its own actions to determine next steps.

It can be safely said that every AI agent operates by moving around a loop. This loop is what gives agents autonomy, reliability, and the ability to recover from failure. Agentic AI, powered by the loop, is contextual, continuous, and environment-aware.

Lilian Weng’s formula captures the architecture simply: Agent = LLM + Memory + Planning + Tool Use. The agent loop is the runtime that ties those four components together.

“My Job Is to Write Loops”
- Boris Cherny

What Boris Cherny’s six words mean for every MSP

Read more