Director or Doer? The AI Question Nobody’s Asking

Most of the AI conversations I have these days start the same way. Someone leans in and quietly asks, “Do you think AI is going to take my job?” I understand the worry — it’s everywhere, and it’s loud. But I think it’s the wrong question. The one worth asking is sharper and far more uncomfortable. Are you using AI, or is AI using you? That single reframing changes the whole game. And the window to land on the right side of it is narrowing faster than most people realise.

The Doer Trap

I see the Doer pattern everywhere. Someone types a rushed prompt, reads whatever comes back, tidies up a comma or two, and ships it. The email goes out. The deck gets shared. The summary lands in a meeting. The person feels productive because something got done — but they didn’t really direct any of it. The tool picked the angle, the structure, the tone, even the conclusion. They just drove the delivery truck.

The thing that makes this dangerous is that it feels like progress. Output is going up. Calendars are clearing. But the thinking is going down. The muscles that matter — judgement, taste, point of view — quietly shrink while everyone is busy celebrating how much faster the work moves. If AI is setting the pace, choosing the framing, and deciding what “good” looks like, you are no longer in charge of your own work. You are assisting it.

The Director Shift

The people I watch pulling away from the pack work very differently. They treat AI the way a good manager treats a capable team. They brief it properly. They tell it the audience, the constraint, the outcome they want, and what to leave out. They read the output the way an editor reads a draft — with scepticism, not relief. They push back. They ask it to try a sharper angle, to argue the opposite, to shorten by half. They know what great looks like before they ask for it, and they recognise when the answer is merely adequate.

Being the Director is harder. It takes domain knowledge, taste, and the patience to iterate. But the work that comes out the other side is genuinely yours. The ideas are yours, the standards are yours, the reasoning is yours. AI is doing the heavy lifting on the mechanics while you do the heavy lifting on the thinking. That’s the right shape of the partnership.

The Window Is Closing

Here’s what I think people underestimate. The gap between Directors and Doers is compounding. Every week spent actively learning how to brief, evaluate, and steer these tools is a week of skill you’re banking. Every week spent passively accepting output is a week of skill you’re quietly losing. Six months from now, a year from now, that gap will be visible from across the room — in the quality of decisions, the confidence of arguments, the crispness of output.

The people who dig in now, who actually invest the hours to learn this properly, aren’t just getting better at AI. They’re becoming more valuable than they were before AI existed. Their judgement is sharper. Their output is broader. Their leverage is higher. The people waiting for it to settle down are going to wake up behind, and it will take a lot more than a weekend of prompting tutorials to catch up.

So I’d stop asking whether AI is coming for your job. Ask instead who’s running whose day. Because that answer — today, this week, this month — is the one that decides where you end up.

When Your LLM Goes Down: Are MSPs Designing a New Single Point of Failure?

Over the past year, I’ve watched something fascinating—and slightly uncomfortable—happen inside MSPs and their clients’ businesses. AI tools, particularly Microsoft 365 Copilot, have gone from “interesting experiment” to “critical part of how work gets done” at a pace I don’t think many people fully appreciate yet.

And that raises an uncomfortable question we haven’t really answered:

What happens when the LLM isn’t there?

Not slow. Not “a bit less helpful.”
Actually unavailable.

AI Has Quietly Moved Into the Critical Path

In some of the environments I’m seeing, Copilot isn’t just helping draft emails or summarise meetings. It’s shaping decisions.

Staff are using it to draft client responses, interpret data, build proposals, prepare board slides, and make sense of complex information faster than they ever did before. Managers are using it to think through options, not just document outcomes.

That’s important, because it means AI has crossed a line. It’s no longer a convenience layer. It’s becoming part of the business process itself.

From an MSP perspective, that should set off the same internal alarm bells as any other critical dependency. Because if your client’s process assumes Copilot is available, then Copilot downtime is no longer “an inconvenience”. It’s downtime.

The New Form of Business Continuity Risk

We’re very good, as an industry, at talking about disaster recovery in traditional terms. Backups. Redundancy. Failover. RPOs and RTOs.

But AI introduces a different kind of risk—cognitive dependency.

Here’s a simple scenario I’ve already seen play out in smaller ways:

A staff member is used to Copilot summarising long email threads before client calls. One day it’s unavailable. They’re still expected to run the meeting, but they haven’t read the full thread because the process evolved around “the AI will summarise it”.

No data was lost. No system was breached. But productivity drops, confidence drops, and errors creep in.

Now scale that to proposal preparation, reporting, or internal decision-making processes that assume AI assistance.

We haven’t lost data—but we’ve lost thinking capacity under time pressure.

“The AI Will Be Back Soon” Is Not a Strategy

One of the more dangerous assumptions I hear is:
“Microsoft will fix it quickly.”

Maybe. Probably. But that’s not business continuity planning. That’s hope.

As MSPs, we need to start asking different questions during AI discussions:

What manual process exists if AI is unavailable for a day?
Do staff know how to complete the task without AI, or have we trained that muscle out of them?
Which workflows are AI‑assisted—and which are AI‑dependent?

This isn’t about rejecting AI. I’m fully in favour of using Copilot when it genuinely improves outcomes. But professional-grade technology adoption has always meant understanding failure modes, not just success stories.

Designing AI‑Resilient Workflows

The smarter MSPs I’m working with are starting to treat AI like any other tier‑one system:

Document the “AI unavailable” version of key workflows
Set expectations with clients that AI enhances productivity but is not guaranteed
Train staff to validate, understand, and reconstruct work without AI assistance
Decide consciously where AI is optional versus where it must never be the only path

Ironically, the organisations doing this best often get more value from Copilot, not less. Why? Because they understand it as an accelerator—not a replacement for thinking.

The Question MSPs Should Be Asking Right Now

AI isn’t going away. Dependency will increase, not decrease. That makes this a leadership issue, not a technical one.

So here’s the question I think every MSP owner should be asking themselves:

If Copilot vanished tomorrow, which of my clients’ processes would break—and would they even realise why?

If the answer makes you uncomfortable, that’s a good thing.

That discomfort is the early warning system telling you it’s time to evolve disaster recovery thinking for the age of AI.

AI Didn’t Remove Programming – It Lowered the Bar

One of the most dangerous misunderstandings I hear is:
“AI means we don’t need programming anymore.”

The opposite is true.

We need more programming literacy—just a different kind.

AI doesn’t replace logic, structure, or clarity. It amplifies them. When an AI tool “writes code” for you, what it’s really doing is translating your intent into something executable. If your intent is vague, messy, or logically broken, the output will be too.

MSPs already see this in practice:

A poorly described Power Automate flow that works once and then quietly breaks.
An AI-generated script that technically runs but makes unsafe assumptions.
A Copilot prompt that looks clever but produces useless business output.

The common issue isn’t the tool. It’s the thinking behind the instructions.

Understanding basic concepts—inputs, outputs, conditions, loops, exceptions—has never been more important. The difference now is you don’t need to memorise syntax. You need to think clearly and explain cleanly.

This Is a Business Advantage, Not a Technical Party Trick

Here’s where many MSPs miss the opportunity.

They see AI-assisted “programming” as something clever techs play with internally. In reality, it’s fast becoming a deliverable business capability.

Think about your SMB clients:

They know their processes are inefficient.
They can explain what they want, but not how to build it.
They don’t want a six‑month dev project for a simple workflow problem.

An MSP that can sit with a client, map a process in plain English, and turn it into an automated solution is no longer just “support”. You’re helping redesign how the business operates.

And the simplicity is the point.

A one‑page English description that becomes:

A ticket triage workflow
An onboarding checklist generator
A management report assembler
A light internal chatbot using their own documents

None of that needs hardcore development skills anymore—but all of it still needs structured thinking.

Your Team Doesn’t Need Coding Skills – They Need Programming Awareness

This is where MSP leaders need to be deliberate.

You don’t suddenly need Python experts across your service desk. What you do need is:

Staff who can break problems into steps
People who can explain outcomes unambiguously
A shared understanding of how logic flows

If your team can already document SOPs well, they are halfway there.

I’ve seen MSPs get real value by:

Treating AI prompts like mini specifications, not chat questions
Reviewing AI-generated automations as a team, not blindly deploying them
Teaching junior staff how to describe a problem, not just which tool to click

Those are capability investments, not tool training.

The MSPs Who Win Will Treat This as a Core Skill

We’ve crossed a line. Programming is no longer gated by language barriers—it’s gated by thinking quality.

That changes what “technical literacy” means for MSPs.

The firms that thrive over the next few years won’t be the ones chasing every new AI tool. They’ll be the ones that:

Build strong internal habits around logical thinking
Help clients translate business problems into clear instructions
Package simple automation as repeatable, billable outcomes

If English is now the language of code, the question is simple:

Are you teaching your people how to speak it clearly—or assuming the tools will do that for them?

That’s a strategic choice every MSP leader needs to make, sooner rather than later.

Most MSPs Don’t Need Reinvention — They Need Better Systems

I spend a lot of time talking to MSP owners who feel stuck. Revenue is steady but flat. The team is busy, but not always effective. Sales relies on the same few people. Marketing is inconsistent. Documentation lives in too many places and still gets ignored.

What strikes me is this: most of these businesses are already doing a lot of things right. They’re just doing them manually, repeatedly, and inconsistently.

The uncomfortable truth is that most MSPs aren’t a full transformation away from growth. They’re only a few small system changes away from it. And right now, Microsoft 365 Copilot is exposing exactly where those gaps are.

Not because it’s “AI”.
But because it forces you to confront how broken your internal systems actually are.

The Real Constraint Isn’t Effort — It’s Repeatability

Here’s the pattern I keep seeing. An MSP wins business because one or two senior people are excellent. They know how to sell. They know how to explain value. They know how to scope work. They know how to write proposals that close.

The problem? None of that is systemised.

So every deal depends on hero effort. Every proposal is written from scratch. Every quarterly review relies on memory. Every new hire takes months to onboard because the knowledge lives in someone’s head.

This is where Copilot becomes uncomfortable — in a good way.

When Copilot is dropped into a messy tenant, it doesn’t magically fix anything. It simply reflects reality back at you. Disorganised docs stay disorganised. Inconsistent processes stay inconsistent. Scattered information stays scattered.

But when the systems are right? That’s when the leverage appears.

When Systems Start Doing the Heavy Lifting

I’ve seen MSPs get real traction when they stop thinking of Copilot as a feature and start using it as a force multiplier for their existing wins.

One example: sales proposals.
If your proposals are consistent, well-written, and stored properly, Copilot can help staff generate first drafts that sound like the business — not like a junior guessing. That doesn’t replace sales skills. It removes friction.

Another example: customer communication.
When meeting notes, action items, and follow-ups live in the right place, Copilot turns conversations into continuity. Clients feel like the business is organised, responsive, and on top of things — even when the team is stretched.

The biggest shift, though, is internal.
When onboarding guides, security standards, SOPs, and client histories are actually usable, Copilot acts like institutional memory. New staff ramp faster. Senior staff stop being bottlenecks. Decisions get made with context, not gut feel.

That’s not automation for the sake of it.
That’s systems enabling growth without burnout.

Autopilot Isn’t About Less Work — It’s About Better Work

Let’s be clear: autopilot doesn’t mean switching off. It means the business stops relying on constant pushing just to maintain altitude.

When systems handle the routine thinking — drafting, summarising, correlating information — people get to focus on judgement, relationships, and strategy. The things MSPs say they value, but rarely have time for.

Copilot doesn’t close deals by itself.
But it supports a system that does.

It doesn’t magically scale marketing.
But it makes consistency achievable.

It doesn’t hire great staff.
But it makes working in your business less chaotic — which is how you keep them.

The Question Every MSP Should Be Asking

Instead of asking, “What can Copilot do?”, the better question is:
What would break if our best people were unavailable next week?

Where the answer is “everything”, that’s where the system is missing.

Copilot doesn’t create discipline.
It rewards it.

For MSPs willing to invest in structure — not just tools — this is the closest thing I’ve seen to pushing an autopilot button. Not because it removes effort, but because it finally turns what you’re already good at into something that scales.

And that’s where real growth starts.

Copilot image generation in PowerPoint

One of the ways I like to test new image models is to give them the same prompt to create an image of an expresso machine. You can see the last iteration here:

https://blog.ciaops.com/2026/05/05/revisiting-copilot-image-generation-analysis/

More advanced image options have made their way into PowerPoint and it allows me to use the image model of my choice per:

So to test, I took exactly the same prompt I had used before to create images and used it inside PowerPoint with each model. Here’s the result from using GPT-Image-1.5

and Flux.2 Flex

These output are now beginning to revival dedicated image creation tools and they are backed right into PowerPoint and probably soon, all other desktop apps thanks to Copilot!

As a slight variation, I asked Copilot in PowerPoint to create a complete slide rather than just a single image and here’s what I got:

GPT-Image-1.5

and Flux.2 Flex

Inspect What You Expect: Why MSPs Can’t “Set and Forget” Copilot (or Anything Else)

One pattern I see repeatedly in MSP businesses—especially as they start adopting Microsoft 365 Copilot—is the quiet belief that once something is delegated, the job is done.

Hand it to a technician.
Hand it to an admin.
Hand it to an AI tool.

Then move on.

That approach has always been risky. With Copilot in the mix, it becomes outright dangerous.

Not because Copilot is untrustworthy—but because systems don’t improve unless you observe them. And as an MSP, improving systems is literally your job.

Delegation Without Inspection Is How Problems Hide

Most MSPs already understand this with infrastructure. You don’t deploy backups and just hope they work. You test. You get alerts. You look for drift.

But when it comes to productivity work—emails, reporting, meetings, content creation—we suddenly relax.

I’m seeing MSPs roll out Copilot, show users a few prompts, and then disappear. No feedback loop. No measurement. No review of outputs or behaviours.

Weeks later, the questions start:

“Why are users still asking basic questions?”
“Why hasn’t productivity improved?”
“Why does Copilot feel underwhelming?”

The issue isn’t the tool. It’s the lack of inspection.

Copilot Changes the Work—So You Need New Sensors

Copilot doesn’t just speed things up. It changes how work happens.

People delegate thinking earlier.
Drafts appear faster.
Decisions are made with less friction—and sometimes less reflection.

That’s powerful, but only if you can see what’s going on.

This is where I introduce the idea of sensors.

Sensors are simple mechanisms that tell you when reality drifts from expectation. They’re not about distrust—they’re about visibility.

In Copilot terms, that might look like:

A short weekly check‑in where users paste an example output that helped (or failed).
A dashboard showing adoption signals across apps, not just license counts.
A Teams message when usage patterns drop after the initial rollout.
A review cadence where managers validate whether Copilot‑created artefacts are actually being used.

None of this is complex. Almost no one does it.

AI Amplifies Weak Processes First

Here’s the uncomfortable truth: Copilot makes good systems better and bad systems louder.

If documentation is outdated, Copilot spreads outdated thinking faster.
If decision rights are unclear, Copilot accelerates confusion.
If users don’t know what “good” looks like, Copilot produces more confident mediocrity.

Inspecting outcomes—not effort—is how you catch this early.

I’ve worked with MSPs who expected Copilot to “lift capability” across the board. What actually happened was more revealing: high performers got better, while poor habits became more visible.

That visibility is a gift—if you’re looking for it.

Growth Comes From Feedback Loops, Not Trust Falls

Whether you’re an MSP of five people or fifty, growth doesn’t come from hiring smarter people or deploying smarter tools. It comes from tightening feedback loops.

That’s why mature MSPs obsess over:

Red/green indicators
Exception reporting
Notifications when something deviates from normal

The same thinking now applies to knowledge work.

Copilot isn’t a project you “finish”. It’s a system you tune. And tuning only works when you inspect what you expect.

The Takeaway for MSP Leaders

If you’re advising clients—or running your own MSP—don’t treat Copilot like a magic upgrade.

Treat it like any other core system:

Define what “good” looks like
Build simple sensors
Review outputs, not intentions
Adjust the environment, not just the prompts

Trust is fine. Visibility is better.

If you’re not inspecting, you’re guessing. And guessing doesn’t scale—especially in an AI‑assisted world.

The Real Reason Copilot “Didn’t Work”? No One Defined What Success Looked Like

I keep hearing the same complaint from MSPs experimenting with Microsoft 365 Copilot.

“It didn’t really land.”
“The team didn’t get much value.”
“We turned it on, but outcomes were mixed.”

When I dig into those conversations, the issue is almost never licensing, configuration, or even training.

It’s much simpler—and more uncomfortable.

No one explained the criteria for success.

A Team Can’t Execute a Standard They’ve Never Seen

I’ve watched this play out inside MSPs and their clients more times than I can count.

Copilot gets enabled. People are encouraged to “use AI.” Expectations are implied, not stated. Then weeks later, leadership wonders why email quality is inconsistent, reports still take too long, or meetings haven’t magically improved.

Copilot didn’t fail. The organisation did.

Humans—and AI—perform best when “good” is clearly defined. If you don’t articulate what a successful outcome looks like, Copilot will happily produce something, but it won’t necessarily produce the right thing.

This is where Copilot quietly exposes a weakness many MSPs already have: undocumented standards.

Copilot Forces the “Definition of Done” Conversation

One of the most valuable things Copilot does isn’t writing content or summarising meetings. It forces people to think clearly before they ask.

When someone prompts Copilot effectively, they’re doing implicit work:

What is the purpose of this output?
Who is it for?
What would “finished and acceptable” actually look like?

Without that clarity, prompts drift, outputs vary, and frustration sets in.

I now encourage MSPs to write down three to five criteria that define “done” for common tasks before encouraging Copilot use.

Not documentation theatre. Just enough clarity to guide behaviour.

A Practical MSP Scenario You’ll Recognise

Take a simple task: internal client update emails.

Without a definition of done, Copilot outputs range from overly wordy to dangerously vague. The problem isn’t AI—it’s ambiguity.

Now imagine the standard is written down:

Clear summary of what was done (in plain language)
Any risks or follow‑ups explicitly called out
No technical jargon unless requested
Suitable to forward directly to a non‑technical client
Under 200 words

Suddenly, Copilot becomes consistent, fast, and useful. Junior staff improve overnight. Senior staff stop rewriting everything. The standard becomes repeatable.

Copilot didn’t create the quality. The criteria did.

Why This Matters More Than the Tech

MSPs love tools, but tools don’t fix thinking problems.

Copilot changes the way people work by making fuzzy expectations painfully visible. If staff don’t know what a “good” report, ticket update, or proposal looks like, Copilot will simply amplify that uncertainty at scale.

The MSPs seeing real productivity gains are doing something different. They’re treating Copilot as a thinking partner, not an output machine.

They define success first, then let Copilot help execute it faster and more consistently.

That shift—from “do the task” to “meet the standard”—is where the real business impact sits.

What I’m Advising MSP Leaders to Do Now

Before your next Copilot rollout, pause.

Pick three high‑value tasks your team does daily. For each one, write down three to five simple success criteria. That’s it.

Not policies. Not 12‑page SOPs. Just clarity.

Then show the team how Copilot supports that standard.

The Takeaway

If Copilot “isn’t delivering value,” don’t start by blaming the tool.

Ask a harder question instead:

Did we ever explain what success actually looks like?

Because a team can’t execute a standard they’ve never been shown—and Copilot will expose that gap faster than any consultant ever could.

If you get the definition of done right, Copilot becomes a force multiplier. If you don’t, it just makes the mess more obvious.

And honestly? That might be exactly the wake‑up call MSPs need.

Claude Cowork vs Copilot Cowork: why the Microsoft answer wins for SMB

I’ve watched a lot of clients spend the last twelve months stitching together AI tools that don’t talk to each other. A Claude tab here. A ChatGPT tab there. A Copilot tab somewhere in the middle. Then a folder of CSVs they keep dragging in and out of each one.

That’s not a workflow. That’s a tax.

So when Anthropic shipped Claude Cowork and Microsoft shipped Copilot Cowork in roughly the same window, the question landed in my inbox: which one do we tell our clients to use?

I’ll save you the suspense. For an SMB already paying for Microsoft 365, it’s not close.

What is Cowork, really?

Cowork is the bit that does the work, not the bit that talks about it. You give it an outcome — “draft the quarterly update from these meeting notes and send it to the leadership team” — and it goes off and does the thing.

That’s the shared idea. Both products own it. The split is in where the work happens.

Claude Cowork lives on your desktop. You mount a folder, drop your files in, and Claude runs in a sandbox on your machine. It doesn’t see your inbox. It doesn’t see your calendar. It doesn’t see your Teams chats unless you’ve copy-pasted them in. You bring the data to the model.

Copilot Cowork is the inverse. It already lives inside Microsoft 365, grounded in your Outlook, Teams, SharePoint, OneDrive, and calendar through Work IQ. You don’t mount anything. The model is already where your data lives.

Notice what’s missing? The mounting step. The “let me copy this folder over” step. The “hang on, I need to paste in the email thread” step.

For SMBs, that’s the whole game.

Step-by-Step: getting Copilot Cowork going

If you’re licensed for Microsoft 365 Copilot and enrolled in the Frontier preview, the start is short.

Open Cowork in Microsoft 365

Browse to m365.cloud.microsoft, sign in, and pick Cowork from the agent list. If you don’t see it, check Frontier enrolment under Copilot settings.

Describe the outcome

Skip the prompt-engineering nonsense. Talk like a person.

Read my inbox from this week, find anything tagged from a client,
draft a Friday wrap-up email summarising open items, and post a
short version into the Operations channel in Teams.

Notice what’s missing? Any reference to a file path. Any “first export your inbox to CSV” step. Cowork already has the inbox, the calendar, the Teams channels, and the SharePoint files. It just needs the instruction.

Approve the action

Cowork shows you exactly what it’s about to send, post, or schedule before it does it. You hit Send, Post, or Cancel. The full flow is in the getting started doc if you want to walk a client through it.

Set the schedule

Want it to do this every Friday at 4pm? Schedule the prompt and walk away. Copilot doesn’t get tired. Use that.

Why this actually changes behaviour

Claude Cowork is a beautifully built tool. For a developer or a data analyst on a Mac with a folder full of CSVs, it sings. I’m not knocking it.

But that’s not the SMB picture. The SMB picture is a bookkeeper, a sales lead and a director who all live inside Outlook and Teams from 8am to 6pm. Their data isn’t in a folder on their desktop. It’s in their mailbox, their channel chats, their SharePoint sites and their meeting transcripts.

“But couldn’t we just pipe our M365 data into Claude?”

You could. You’d be paying twice — once for M365, once for Claude — and you’d be exporting business data into a different vendor’s environment to do work the platform you already own can do natively.

That’s not a productivity gain. That’s a procurement problem.

Here’s the real win. Copilot Cowork sits behind the same Entra identity, the same conditional access, the same Purview labels and the same retention policies your tenant already runs. The governance story is already built. There’s no second tool to license, secure, train, or audit.

My recommendation? If you’re an MSP and you’re not walking your SMB clients through Copilot Cowork, you’re leaving value on the table — theirs and yours.

Meet people where they already are.

Cowork isn’t a second AI app for your clients to learn. It’s the work, finally getting done in the place it was always supposed to happen.