The Service-as-Software Model Has a People Problem

If you follow the business side of AI even loosely, you have probably heard some version of the phrase "services-as-software" this year. It’s become one of the most talked-about ideas in venture capital, and since it touches almost every business that pays people to get work done, I think it deserves a plain-English explanation and an honest critique.

Here’s the idea in simple terms: for the last twenty years, software companies have mostly sold tools. You pay a monthly fee per user for something like accounting software or a scheduling platform, and your own people use that tool to do the work. The software makes them faster, but they’re still the ones doing the job. Services-as-software flips that arrangement around, and instead of selling you a tool for your bookkeeper, an AI company sells you the bookkeeping itself, delivered mostly by AI agents, and charges you for the finished result.

Sequoia Capital, one of the most influential investment firms in Silicon Valley, made the case for this in an essay published in March. Partner Julien Bek wrote, "the next $1T company will be a software company masquerading as a services firm," and he summed up the shift in a single line: "A copilot sells the tool. An autopilot sells the work." A copilot, in his framing, is AI that helps a professional do their job. An autopilot is AI that does the job and hands you the outcome.

I’ve read that essay several times, and I agree with more of it than you might expect from someone who runs an AI transformation firm. AI really can handle a remarkable share of the routine, information-heavy work that businesses have always paid people to do, and some of that work is undoubtedly going to move to software. Where I part ways with the thesis is in what it leaves out. The services-as-software pitch carefully counts the labor AI can replace, but gives very little weight to the labor AI depends on. What I mean by this is the people inside a business who have to trust what the AI produces, change the way they’ve worked for years, catch its mistakes, and keep using it long after the vendor has moved on to the next customer. In my experience across dozens of successful AI transformation engagements, those people decide whether an AI investment pays off, but the business models getting funded right now treat those people almost as an afterthought.

Why investors are so excited

To understand the enthusiasm, it helps to look at the numbers investors are working with. Sequoia's essay points out that "for every dollar spent on software, six are spent on services." In other words, businesses spend far more paying people (employees, contractors, agencies, outside firms) than they do on software licenses. Foundation Capital, another prominent venture firm, estimates that the pool of spending AI agents could eventually take over is about $4.6 trillion a year, made up of salaries and outsourced services. For comparison, it puts the traditional software market at roughly $200 billion. If you’re an investor, the appeal is obvious. Software was a big market, and this looks like a market more than twenty times its size.

What I want business owners to notice is what that $4.6 trillion is actually made of. It’s a headcount number. Primarily, it’s payroll and service contracts, which means it’s actual people. The pitch treats those people as the cost that AI will capture.

The trouble is that many of those same people are the ones who have to say yes before any of it works. Think about the dispatcher who has to trust the AI's schedule enough to stop double-checking it by hand, or the controller who has to put their name on a reconciliation the AI prepared, or the project manager who has quietly kept the "real" schedule in a spreadsheet on their desktop for six years and now has to let it go. On an investor's slide, those people are the cost being removed. Inside an actual company, they’re the ones who make or break the rollout. A business model that overlooks that is, in my view, built on the wrong assumptions.

We have already seen how this plays out

We don’t have to guess what happens when a company focuses on the cost of labor and underestimates the people around it, because Klarna, the buy-now-pay-later company, ran the experiment in public. In February 2024, Klarna announced that its new AI assistant was doing the work of 700 customer service agents and handling the majority of its customer chats. It quickly became the most cited example in the entire services-as-software conversation, proof that AI could take over a whole service function.

About a year later, the story changed. In May 2025, CEO Sebastian Siemiatkowski told reporters that the AI-first approach had led to "lower quality" service and that cost had carried too much weight in the decision. He said, "investing in the quality of human support is the way of the future for us," and that it’s, "so critical that you are clear to your customer that there will always be a human if you want." Klarna started hiring people again.

I give Klarna a lot of credit for saying that out loud, since most companies quietly bury a reversal like that. I think the lesson is a simple one: a service is a relationship, and every relationship has people on both sides of it. When you automate one side without thinking carefully about the other, quality slips in ways that rarely show up on a dashboard until customers start leaving.

A lesson from a hundred years ago

There’s a piece of economic history that I think every business owner should know right now, because it explains the current moment better than anything written this year. In 1990, a Stanford economist named Paul David published a paper called "The Dynamo and the Computer," which looked at how long it took electricity to actually make factories more productive.

Electric power was commercially available in the early 1880s. You would expect factories to have become dramatically more efficient soon after, but for roughly two decades, electricity did almost nothing for their productivity. The reason is that most factory owners simply took out their steam engine and put a big electric motor in its place. Everything else stayed the same. The same belts and shafts ran through the same multistory buildings, the machines were arranged the same way, and people did the same jobs. They had a new power source running the old way of working.

The real payoff didn’t arrive until the 1920s, when American manufacturing productivity finally surged. What changed was the factory itself. Owners began building single-story plants laid out around the flow of materials. They put small motors on individual machines so each worker could start and stop their own equipment. That new layout required workers who were trained differently and trusted with more control over their own work. Electricity made the new factory possible, but it was the reorganization of people and work that produced the gains, roughly forty years after the technology first showed up.

I believe most companies are in the steam-engine-swap phase with AI today. We’ve bolted a powerful new motor onto the old way of working and we are waiting for productivity to appear. The redesign is the real transformation, and redesign is people work: new roles, new habits, new levels of trust, and new training. Software can help that redesign happen faster, but someone still has to lead the people through it.

The research keeps pointing in the same direction

This is more than a hunch or a history lesson. Several of the most respected research firms in the world keep arriving at the same conclusion.

Boston Consulting Group has a long-standing rule of thumb for AI projects, often called the 10-20-70 rule. It holds that roughly 30 percent of the effort in a successful AI transformation goes to the algorithms, technology, and data, and about 70 percent goes to people, processes, and organizational change. If you only remember one number from this post, I would make it that one. Most companies budget as if those percentages were reversed.

BCG's newest study, published in July, adds some sharp detail. Only about 6 percent of companies qualify as what BCG calls "AI leaders," meaning companies that are getting real, measurable value from AI. At those leaders, 13 percent of employees have AI-related skills, compared with just 1 percent at the companies lagging behind. BCG put it this way: "What cannot be bought is the organizational capability to deploy them into the specific economics of a business." And they added, "This talent transformation does not happen organically. It requires deliberate organizational change."

Researchers at MIT reached a similar conclusion from a different angle. Their massively popular viral  2025 study found that about 95 percent of organizations were seeing no measurable impact on their profit and loss from generative AI, despite tens of billions of dollars in spending. The same research found something that surprised a lot of people: over 90 percent of workers were already using personal AI tools like ChatGPT for their jobs, while only about 40 percent of companies had bought an official subscription. When you put those two findings side by side, the picture is pretty clear: employees are already curious and already experimenting on their own. What most companies are missing is the structure, training, and support to turn those scattered experiments into shared, dependable ways of working.

Gartner, the technology research firm, has also weighed in. It predicts that more than 40 percent of projects built on AI agents (software that can take actions on its own, like sending an email or updating a record) will be canceled by the end of 2027, because of rising costs, unclear business value, or weak risk controls. Gartner also estimates that only about 130 of the thousands of companies selling "agentic AI" are offering the real thing.

These are different firms using different methods, and they all describe the same pattern. The technology tends to be ready long before the organization is.

Even the investors see it

What I find most telling is that the investors backing services-as-software acknowledge this in their own writing. Foundation Capital's review of the category's first year says that success depends on how deeply a product is implemented inside a customer's business, and it describes engineers working on-site with customers to document the "unspoken tribal knowledge" that lives in people's heads. Sequoia's essay draws a line between intelligence, which AI increasingly handles well, and judgment, which today still sits with people. And Fortune's coverage of the thesis in April flagged organizational inertia, the tendency of companies to keep doing things the way they always have, as one reason established service providers may hold their ground.

So the people funding this wave already recognize the human side of the equation. They tend to list it as a risk to manage. At Avolis, we see it as the core of the work, and that difference in perspective leads to a very different kind of company.

What "putting people first" looks like in practice

"Put people first" is easy to say and even easier to cut when budgets get tight, so I want to be specific about what it means, especially for the operations-heavy businesses we work with every day.

It starts with training that actually fits the job. A field technician and an accounts receivable clerk need completely different AI skills, and a generic webinar rarely changes how either of them works. Training built around each person's real, day-to-day workflow is what changes Monday morning.

It continues with support after launch. The first few weeks after a new system goes live are when adoption usually succeeds or fails. Someone needs to be reachable when the tool does something unexpected, and that someone needs to understand the business as well as the software, so they can tell the difference between a technical glitch and a process that needs rethinking.

It includes encouragement and visible wins. People tend to adopt what their respected coworkers adopt. Finding early champions on a team, celebrating what they figure out, and helping those wins spread is one of the most effective things any AI rollout can do, and it is almost entirely human work.

It requires clear ownership. When the AI gets something wrong on an important account, a person has to answer for it and fix it. That kind of accountability is exactly what makes a team willing to trust automation in the first place.

And it means capturing what people know. Every time an employee corrects the AI, they are contributing expertise. When that knowledge is written down as shared instructions, standard procedures, and clear rules about what AI is allowed to do, what needs a human's approval, and what information should never leave the building, the company starts to own its AI capability. That ownership is also what lets a business switch to better tools down the road without starting over from scratch.

None of this comes in a software download, and all of it determines whether the download was worth buying.

Where Avolis stands

I want to be direct about why we built Avolis the way we did.

We are an AI transformation firm, and we use every capable tool on the market. We have a team dedicated exclusively to finding new technology, learning how to use it, and discovering how we can deliver that technology to our customers. We also believe the technology is the smallest part of the equation. Our approach has three phases: Diagnose, Build, and Embed. The Embed phase is the part most of this industry is now trying to automate away. It is the weeks and months of training, support, follow-up, and encouragement that turn a working system into the way a company actually operates. We do that work as forward-deployed partners, which simply means we show up in person, inside the business, alongside the people who have to live with the change. To put it simply: people first, always. 

We measure ourselves by what your team is still using six months after we arrive. That’s a harder standard to meet than a launch date, and it’s the one that actually shows up in your bottom line.

I expect services-as-software to win plenty of truly repetitive, easily outsourced work, and I think that is a good thing. When the goal is transforming how your own company operates, though, what you are really buying is a change in how people work. That kind of change takes people to lead it.

If you want an honest read on where your team stands today, and what it would take to bring your people along, a good place to start is our AI Readiness Evaluation.

Sources

• Sequoia Capital, Julien Bek, "Services: The New Software" (March 5, 2026): https://sequoiacap.com/article/services-the-new-software

• Fortune, "This Sequoia partner thinks AI-enabled services are the new software" (April 21, 2026): https://fortune.com/2026/04/21/services-are-the-new-software-sequoia-venture-capital-julien-bek-ai-native-eye-on-ai/

• Foundation Capital, "The $4.6T Services-as-Software opportunity: Lessons from the first year" (July 3, 2025): https://foundationcapital.com/ideas/the-4-6t-services-as-software-opportunity-lessons-from-the-first-year

• Entrepreneur, "Klarna CEO reverses course by hiring more humans" (May 2025): https://www.entrepreneur.com/business-news/klarna-ceo-reverses-course-by-hiring-more-humans-not-ai/491396

• Paul A. David, "The Dynamo and the Computer" (1990): https://gwern.net/doc/economics/automation/1990-david.pdf

• Tim Harford on David's research, via History News Network: https://www.hnn.us/article/paul-david-it-took-decades-for-the-economic-impact

• BCG, "AI Talk Is Cheap. Value Creation Is Rare." (July 9, 2026): https://www.bcg.com/publications/2026/how-ai-leaders-create-competitive-advantage

• MIT NANDA, "The GenAI Divide" coverage, Virtualization Review (August 19, 2025): https://virtualizationreview.com/articles/2025/08/19/mit-report-finds-most-ai-business-investments-fail-reveals-genai-divide.aspx

• Gartner, "Over 40% of Agentic AI Projects Will Be Canceled by End of 2027" (June 25, 2025): https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027