Agentic AI can’t do it alone: the power of people-in-the-loop service models

The buzz around agentic AI — a type of artificial intelligence that includes sophisticated reasoning to autonomously solve complex, multi-step problems — has sparked some bold claims about how it’s going to transform the $4.6 trillion professional services market. 

Thought-provoking perspectives emerging from venture firms like Foundation Capital and Decibel paint a future where agentic AI transforms large swaths of services into a “services-as-software” (yes! SaS with one “a”) market with autonomous AI agents running the show. But is that vision realistic?

The tech industry loves a good revolution, and the narrative around agentic AI often frames the future as a binary choice: humans or technology. The truth is, enterprise services are messy, complex, and full of edge cases that AI alone can’t handle. 

What's often overlooked is that customers aren't buying agents or AI — they're buying outcomes. When you get your car repaired do you care how many mechanics worked on it? Or the tools they used? The measures of success are the results, the time it takes to get results and the cost. The number of agents or systems involved along the way? That's just implementation detail.

In my view, despite its promise of autonomy, agentic AI when applied to industry-specific challenges in enterprise environments, will continue to fall short due to the real-world differences  of enterprises. This includes the variability in their processes and the old-fashioned need for human trust. I believe the real opportunity lies in combining the best of both. Humans bring empathy, judgment, and adaptability, while technology offers speed, the opportunity to automate and enforce operational processes (precision), and scalability. Together, they create something far greater than the sum of their parts — a "people-in-the-loop" model, where humans and AI work together to deliver high-quality outcomes.

What is “services as software”?

Services as software is often described as the next evolution of SaaS, where AI-driven automation handles the tasks traditionally performed by users. The idea is to transform services into software products by eliminating human involvement, allowing companies to scale faster, cut costs, and improve efficiency. Sounds like magic, doesn't it?

While this vision is bold, it oversimplifies the problem.

Imagine, for example, an AI agent that handles the entire mortgage underwriting process — pulling credit reports, verifying employment, analyzing bank statements, assessing property values, and generating loan decisions without any human review. The appeal? That $4.6 trillion that’s currently spent on lower-margin people-intensive services turns into high-margin software products. Meanwhile the people can go on to do higher-value things.

The problem, though, is that services like these aren’t static — they're dynamic — and they require human adaptability to reliably get to quality outcomes customers are expecting.

Fixify’s perspective on “services as software”

Don’t get me wrong. Services as software isn’t just a buzzword. It’s a real transformation. But for most services, the reality is that the “services as software” businesses that deliver the most value to their customers will be ones that include people in the loop. And there are several companies out there — including Fixify — that are already leading the way.

Take Scale AI, currently valued at over $13 billion. They've masterfully combined human expertise with technology to deliver high-quality data labeling services. Their success comes not from eliminating humans, but from building technology that makes their people more efficient over time. Similarly, EvenUp Law, valued at over $1 billion, is transforming legal services by pairing the expertise of lawyers with technology, showing that even in highly specialized fields, the combination of human judgment and AI is what drives real transformation.

Notice a common denominator? At its heart, “services as software” is about combining advancing technology (LLMs, APIs, and automation tools) with human expertise to consistently deliver high-quality outcomes. This hybrid approach ensures companies benefit from technological advances without compromising service quality. And when technology doesn’t deliver as expected? Since there’s a human in the loop they can insert themselves as necessary to fix any wobble and ensure users get the outcomes they signed up for.

At Expel, where I was CTO, we embraced this vision early. Expel pioneered a new approach to managed detection and response (MDR) — a service that reviewed, triaged and investigated critical IT security threats across enterprise environments. We combined advanced technology, including AI, with hard-to-find human cybersecurity expertise. We quickly grew to over $100 million in revenue, achieved industry-leading margins, and were twice (2021 & 2023) recognized as a Leader in Forrester's Wave report. 

Our success at Expel came not from solely relying on AI to “do the thing,” but rather from creating a platform that enabled us to optimize the million intricate interactions that were required to deliver that result. This system was organic — the admixture of people and technology changed as we rolled out new features and discovered new ways to make processes more efficient. 

Five years ago we were already using technology to optimize the process of SOC analysts triaging alerts. In the early days at Expel, when analysts were reviewing alerts it was a fundamentally human moment. Then, over time we used technology to gather the most relevant information that we observed humans seeking out during security investigations. This enabled us to optimize the process — fetching the needed information automatically and displaying it to our users so they could make quick and accurate judgements. The impact of this work was a statistically significant shift in the time it took analysts to decide if an alert required action. It also led to a lower stress environment and higher quality outcomes.

This example and many more like it allowed us to capture the benefit of AI and to use the telemetry from the platform to continually tune it, adding just the right dash of technology in the right spot. 

At Fixify we’re big believers in the future of agentic AI. We also believe that services as software is just how you build a business to solve hard problems and deliver great outcomes to customers. In our view, the companies delivering the most valuable outcomes with services as software models will involve a blend of people and technology (including Agentic AI). Sure, technology will take a larger role over time but keeping people in the loop is critical to getting outsized outcomes.

The limitations of agentic AI in services

While agentic AI has come a long way, it falls short when it meets the messy realities of enterprise services. These limitations generally fall into two key areas: technical constraints (what AI struggles to do or simply can't yet) and human factors (what AI may never fully replace). 

Here’s the thing: AI (currently) shines when it's working on well-articulated problems in direct service of a human objective — not when it’s left unsupervised to mind the store. The probabilistic nature, which is its greatest strength, is also a weakness — in search of the determinism that most processes require, it needs supervision and quality control in a way that limits its scale. Enterprise services are never that tidy.

Let’s break it down:

1. Enterprise complexity & variability

The Challenge: Like roads, enterprise environments might seem uniform, but the differences lie in what happens within them. Workflows, tools, and processes vary widely. Just think about the last time you switched jobs. Did you submit an expense report the same way? How about submitting a help desk ticket to IT? Without guidance, agentic AI will obviously struggle to adapt to these nuances. Also, you’d be surprised by how many applications that IT teams manage lack APIs for critical functions like user management. It sounds like a small issue but it makes fully automated software solutions impractical. Without human adaptability, these systems can create more friction than they eliminate. 

The more specialized a service gets — think adventure travel recommendations vs. booking a flight from LA to San Francisco — the more variability there is.

At Fixify, we use humans to absorb this variance, and ensure we can adapt to the workflows that are unique to each organization. At the same time we’re using telemetry, to track each click and action our analysts take so we can identify friction points and engineer them out of the process. This preserves our peoples’ valuable time to do what people do best — make judgment calls and build relationships with other humans.

2. Deficient enterprise process telemetry

The Challenge: Automation thrives on data, and agentic AI needs vast amounts of it to perform. Most foundational models have trained on internet data. However that data, and the models that rely on it, don’t represent the internal realities of how enterprises operate. The result? AI systems can hallucinate or make incorrect recommendations. That’s particularly problematic in decision-support scenarios where accuracy directly impacts customer satisfaction and trust. As one IT Director put it to me “One of the things I love about AI is it's always looking for the most logical answer. Unfortunately, in business, the most logical answer sometimes — and maybe even a lot of the times — isn't the answer.”‍

At Fixify, we tackle this by keeping people in the loop. Our own teams generate data while performing the work. We use that data to train, and augment models as well as to refine our processes — building smarter tools that make our people even more efficient. It’s a virtuous loop. It isn’t just about feeding an algorithm — it’s about deeply understanding the work and making meaningful improvements. And it allows us to offer a deeper level of value that we pass on to our customers

3. The AI cost-benefit equation ‍

The Challenge: While AI solutions might be technically feasible, the cost of developing, implementing and maintaining them are still not well understood. Between quality control, integration and training, the cost to implement and maintain an AI solution often adds up to more than the cost of the problem it’s solving. This is very similar to the hidden and not-so-hidden costs that come from discussions like “should I move my servers from that closet to the cloud.” Organizations have to carefully weigh whether an AI-driven approach makes economic sense compared to human-driven or traditional software solutions.‍

At Fixify, we take a pragmatic step-by-step approach, implementing AI only where it demonstrably improves outcomes and efficiency. More important for our customers, we help bend the cost-benefit curve for implementing AI. Our customers get the benefits of AI — lower cost, faster and higher quality service — without having to devote their own resources to maintaining a collection of AI tools. Meanwhile, our focus remains on delivering value, whether through human expertise, traditional automation, or AI-enhanced solutions.

4. The comfort of human reassurance ‍

The challenge: While agentic AI can handle routine tasks with precision, it falls short in providing emotional intelligence and comfort during crucial moments. For instance, in wealth management, AI might effectively allocate investments, but it can’t offer reassurance and perspective when the stock market tanks by 10% overnight or help a child navigate sensitive life events like a death in the family or a parent with declining mental capacity. The human touch is essential in creating trust and confidence. It’s why even self-driving cars still have steering wheels and the option for human intervention. People want to know there’s someone to turn to when it matters most.‍

At Fixify, we embrace the irreplaceable value of human connection in service delivery. End users are always talking to a real person — not a bot. By integrating people into our processes — and enabling them with AI tools — we ensure that users feel supported, especially in high-stakes or emotionally charged scenarios. Our approach combines the efficiency of automation with the warmth and adaptability of human empathy.

Fixify’s vision: people-in-the-loop service models

At Fixify, we don’t see AI agents as replacements for people. We think their role is to amplify people’s potential by equipping them with the best possible tools and the most efficient processes to help them achieve more. 

As foundational technologies like LLMs (large language models) advance, our technology layer is designed to scale and evolve alongside them. But here’s the thing — progress doesn’t always move as fast or in the ways you expect. And that’s where our people-in-the-loop model changes the game. Even when the pace of technological improvement slows, human insight ensures we deliver the outcomes our customers rely on.

This is the real magic of “services as software”: not removing people from the equation, but thoughtfully integrating human expertise with technological capabilities. It’s this thoughtful combination that creates smarter, more adaptive systems — not just for today, but for the ecosystem we’re building for tomorrow. That’s the hard work we’re doing now, and it’s the foundation for the future of service excellence.

I’d love to hear your perspective. Connect with me on LinkedIn and let me know what you think.