The same week, two people who don’t know each other and operate in different worlds said the same thing.

Jason Shuman, seed investor at Primary Venture Partners in New York, put it this way:

“Silicon Valley thinks AI agents are a $20/mo self-serve subscription. Main Street is paying local agencies $10,000 just to turn them on.”

Aaron Levie, CEO of Box (a company worth $5B+), said this:

“We dramatically underestimate how much change management it is going to take to automate most knowledge worker tasks.”

An investor who spends his days with SMBs and startups. A CEO who spends his with enterprise and Fortune 500. Neither cited the other. And both landed in the same place: the real money in AI isn’t in selling the software. It’s in installing it.

Two signals from opposite sides of the market

Shuman shared data that distills the problem into three numbers. 54% of SMBs don’t have internal AI expertise. 41% have data too poor for AI to work. And 41% already prefer buying AI through a local technology provider.

Three data points that tell a story Silicon Valley doesn’t want to hear. You can’t “1-click install” a genius AI agent into a messy CRM, a 15-year-old server, or a database where half the fields are free-form text with no structure. It doesn’t work. And most real-world businesses are in exactly that situation.

Levie sees it from the other side — large enterprises. And the diagnosis is identical: data trapped in legacy systems without decent APIs, missing context for AI to do its job, teams that aren’t technical and won’t become technical tomorrow. The change management required to automate knowledge work is bigger than anyone is admitting.

But Levie adds something that matters: “This is actually good news if you’re building right now because the opportunity is to build the software bridges to make this easier, or to build new services firms to help with this change management.”

The opportunity isn’t selling more subscriptions. It’s being the bridge between what AI promises and what a company actually needs to do to make it work.

Every technology revolution creates an installer class

This isn’t new. I’ve seen it three times in my career, and history records at least one more.

Electricity, 1880-1920. While Edison and Westinghouse sold the core technology — generators, cables, light bulbs — the real work was done by thousands of local electricians who wired buildings one by one. In 1901, 49 contractors from 18 cities founded NECA (National Electrical Contractors Association) at the same event where Thomas Edison was inaugurating the electrical illumination of the Pan-American Exposition in Buffalo. By 1910, more money was being spent on wiring buildings than on building power plants.

PCs, 1980-2000. IBM and Microsoft provided the boxes and the software, but the real money flowed through the channel. VARs (Value-Added Resellers) and MSPs (Managed Service Providers) who configured networks and managed servers built a massive industry. Today, in the US alone, that channel generates over $500 billion annually according to GTIA (formerly CompTIA).

Cloud, 2010-2020. The rise of AWS, Azure, and Google Cloud created a services market even larger than the infrastructure itself. Accenture, Deloitte, and thousands of boutique consultancies captured more in migration fees than AWS earned in hosting. The work wasn’t spinning up a virtual server — it was migrating data from 20-year-old systems, retraining teams, and redesigning processes that hadn’t been touched in decades.

In every case, the pattern is the same. The technology is generic. The last mile is local, messy, and requires trust. The manufacturer wins on scale. The installer wins on relationship.

The Gartner data from February 2026 confirms this at the macro level: global IT services spending reaches $1.87 trillion in 2026, while software spending sits at $1.4 trillion. The services layer exceeds the product layer by $470 billion. That’s not an anomaly — it’s the permanent structure of how technology gets adopted.

Why self-serve doesn’t work with AI

There’s a distinction that matters and that Shuman identifies well. When your company adopted Slack, Zoom, or Google Workspace, you added a new tool. You didn’t touch what you already had. The database stayed the same. The processes stayed the same. The roles stayed the same. You opened a new app and started using it.

AI doesn’t work that way.

Implementing AI in a real operation means touching existing data — cleaning it, connecting it, structuring it. It means redesigning decisions that until now a person was making. It means the sales team, or operations, or finance, changes how they do things they’ve been doing for years.

Collaboration tools are additive. AI is transformative. Additive can be self-serve. Transformative can’t.

And Shuman’s data confirms it from the demand side: if 41% of SMBs have data too poor for AI to work, no $20/month subscription solves that. Someone has to go in, see what’s there, clean it up, and connect it. Not a chatbot. A person with context about your business.

BCG found that 70% of success in AI projects depends on people and processes, not technology. That’s not an accident. It’s the natural consequence of AI touching everything that already exists — and what already exists is messy.

Three installer models (and which one wins)

If the AI installation market is real — and the data says it is — then the question is who captures it. I see three models competing.

Big consulting. Accenture, Deloitte, McKinsey. Projects lasting 6 to 12 months, teams of 15 people, budgets starting at six figures. Works for Fortune 500. Irrelevant for the 54% of SMBs Shuman describes — they can’t pay $500K for an AI pilot, and they shouldn’t.

The local agency or MSP. Projects from $5K to $50K, direct relationship, proximity. This is Shuman’s “$10,000 just to turn them on.” It’s the company that already manages your email and firewall, now adding AI agent deployment. The problem: many MSPs are still learning AI themselves. Supply hasn’t caught up with demand.

The vertical specialist. Not geographic, but industry-specific. The company that knows dental practices, or trucking logistics, or restaurant chains — and builds AI implementations specific to that vertical. It needs people who combine business knowledge with technical capability — what we at IQ Source call agent maestros.

My bet is that models two and three win for SMBs and mid-market companies. Big consulting stays with enterprise. And the biggest opportunity is at the intersection: the vertical specialist who operates with local agency costs.

Sequoia Capital quantified this with a 6:1 ratio — for every dollar in software, six in services. What Shuman and Levie add is the demand perspective: SMBs are already looking for these installers. The 41% who prefer buying through a local provider aren’t waiting for Silicon Valley to sell them a subscription.

The installer isn’t a technician — it’s a translator

When I read Shuman’s and Levie’s tweets, I didn’t learn anything new. I recognized something I’ve been seeing for over 25 years through every technology cycle: the tool is generic, the integration is specific, and the winner is the one who knows the client’s terrain — not the one who knows the product.

But this time there’s a difference. With electricity, PCs, and cloud, the installer was essentially a technician. They knew how to wire buildings, configure servers, or migrate databases. The work was mechanical and predictable.

With AI, the installer’s job is different. It’s not enough to know how to connect an API. The AI installer has to understand how the business operates — the real workflows, not the documented ones — and design where an agent adds value and where it creates risk. They need to know that a purchase approval flow that has 4 steps on the org chart actually has 11 in reality — including informal approvals over WhatsApp and a manual check against a spreadsheet someone updates on Friday afternoons.

That profile isn’t an engineer. It isn’t an operations manager either. It’s something new: someone who translates between the business and AI agents. Jason Calacanis called it “the maestro of AI agents” — the person who understands a business deeply enough to deploy and manage agents without writing a line of code.

At IQ Source, that’s our job. And we’ve turned it into a service.

What we do: AI Maestro

Our AI Maestro service is exactly what Shuman and Levie are describing — the installer that connects the promise of AI with real-world operations.

When a company reaches out, the first meeting is never about which model to use. It’s about how the business operates today. Who makes which decisions, with what data, at what moment. And where the waste is that an agent could eliminate.

Roughly 3 out of 10 times, the answer is that AI isn’t what they need. Sometimes the problem is data — it’s dirty, scattered, or there simply isn’t enough of it. Sometimes it’s process — the manual operation doesn’t work well, and automating chaos just gives you faster chaos. In those cases, we say no. And that builds more trust than any demo.

When we do move forward, the work looks like this:

Process archaeology. Not the polished flowchart someone made for a presentation. We sit with the people who do the work, follow documents through the system, and map the real process — with its shortcuts, its exceptions, and its tribal knowledge. You can’t automate what you don’t understand. And you can’t understand it from a process document.

Cleaning and connecting data. The 41% with bad data that Shuman mentions is what we see every week. Before any agent, you have to clean, structure, and connect. This step isn’t glamorous. It’s the one that determines whether the AI works or not.

Deploying agents inside the workflow. Agents can’t be another tab that people open when they remember. They have to live inside the tools the team already uses — their CRM, their ERP, their ticketing system. We design that integration so AI becomes part of the work, not an extra step.

Walking alongside the change. Levie is right: change management is what everyone underestimates. It’s not enough for the agent to work — the team has to change how they work. That takes time, patience, and someone who’s there when things don’t go like the demo. The agent maestro is that person.

The 41% who already prefer local providers understand something the rest haven’t yet: that trust, context, and knowledge of your operation matter more than which model you picked.

If you already have AI tools you bought but nobody’s fully using — or if you’re figuring out where to start — send us the list. We’ll tell you which ones make sense for your operation, which need an agent maestro, and which you can cut.

Learn about our AI Maestro service