Service as Software Part 3: How Systems of Agents will collapse the enterprise stack

Cooperative AI agents are eating four decades’ worth of tech systems (Systems of Intelligence are the first course).

Enterprise leaders have for decades been playing a game of Jenga—adding complex, precarious layers to the tech stack, hoping nothing topples. They envision each layer boosting productivity, but each addition falls short. Systems of Record force humans to squeeze their workflows into rigid database fields. Systems of Engagement aim to facilitate collaboration but generate sprawling, unusable data. Systems of Intelligence deliver illusory analytics.?

The approach that will collapse all three into one cohesive framework: a System of Agents, an AI-driven paradigm that doesn’t just improve existing software, but completely dismantles it. These agents, autonomous digital workers, do more than optimize; they reimagine how work gets done.

Two software systems led to digital dysfunction

Let’s take a walk down enterprise software’s memory lane. Starting in the 1980s, most software was designed with a singular focus: storing and processing data. These programs, known as Systems of Record, were built for utility rather than usability. They managed three critical data areas: employee information (Human Capital Management), customer relationships (CRM), and company assets (Enterprise Resource Planning). Because all that data was (and still is) core, software Systems of Record spawned giants like SAP, Oracle, Workday, and Salesforce.

The problems with Systems of Record, however, are many: early on, users struggled with clunky interfaces, from primitive client applications to early web browsers (which weren’t necessarily more user-friendly). The usability problems gave rise to Systems of Engagement, designed with intuitive, user-friendly interfaces that could connect to an underlying System of Record.

But Systems of Engagement created new problems. The programs range from simple email and chat to more sophisticated platforms for audio, voice, and video (think Slack, Microsoft Teams, or Zoom). Because Systems of Engagement are so much more intuitive and collaborative, their user-friendliness creates vast amounts of unstructured data, including multi-dimensional content that traditional databases can’t capture.

The promise (and failure) of Systems of Intelligence?

So new systems sprung up to fill that need. In 2017, Greylock’s Jerry Chen captured the industry’s imagination, coining the phrase Systems of Intelligence, which lived between Systems of Record and Engagement and incorporated basic predictive analytics. The thesis was compelling: these new systems would connect Systems of Record and Engagement, analyzing data to drive insights and actions. Investors and founders rushed to build the new intelligent layers, seeing them as key to unlocking value from enterprise data.

But looking back, Systems of Intelligence were flawed from the start. Such systems are:

1. Misaligned with human work: Systems of Intelligence were architected around data storage paradigms, not human workflows. They force knowledge workers to decompose their rich, contextual work into structured database fields. The result? A profound misalignment between how humans create value and how systems capture it.

2. Dependent on human data entry: Systems of Intelligence require comprehensive, high-quality data from Systems of Record, but employees often neglect to enter critical information, leaving incomplete datasets. This reliance on human input hinders the accuracy and effectiveness of the systems. Even the most sophisticated analytics engine can’t overcome its dependence on manual input. Humans are brilliant at pattern recognition, relationship building, and complex decision-making, but they make terrible data entry clerks. Therefore the data intelligent systems are built on is unreliable.

Software like Outreach exemplifies this problem. Ostensibly Outreach uses AI to design optimal sales sequences and timing for tools like email, LinkedIn, and texts. Under the hood, however, the software depends heavily on sales reps logging activities and keeping CRM data clean. Its “intelligence layer” can only be as good as its underlying data quality, meaning it cannot overcome the “garbage in, garbage out” problem. In addition, its predictions struggle with the sparse data problem; there simply isn’t much historical data to go on. Furthermore, Outreach depends on humans (mostly consultants) to generate the text of its emails; as a result, these email templates only get refreshed a few times a year. By many measures, Outreach is considered a tech-world success, achieving unicorn status in 2019. But its uneven performance since then shows the limitations of Systems of Intelligence, which promise advanced analytics but deliver a glorified dashboard plus workflow tool.

3. Not built to handle unstructured data: As the enterprise data landscape has grown more complex, rigid Systems of Intelligence have failed to adapt. Business value now flows through fluid, multimodal communication—text, audio, video, and voice. But Systems of Intelligence remain imprisoned by structured data. They can’t access, let alone understand, the primary channels where modern work actually happens.?

This is true of Clari, which tries to solve the unpredictable revenue problem using AI to analyze sales data. To function, Clari requires extensive integration with existing systems such as CRMs, meaning it too is vulnerable to inaccuracy if underlying data is incomplete. Clari gives some visibility into pipeline health but struggles to adapt to changing customer behavior, sales territory redesigns, hiring/firing changes for sales reps, and macroeconomic conditions.?

Seven years after Systems of Intelligence emerged, the fact that there is no billion-dollar-revenue company in the space suggests the approach is fundamentally flawed.??

The end of interface: Systems of Agents?

Enterprise companies today have been building layered solutions on top of flawed foundations, but what they need isn’t another patch—it’s a fundamental shift. A System of Agents is that shift. Systems of AI agents don’t just passively wait for human input but actively capture, understand, and act on real-time business activities across all communication channels (emails, calls, documents, video). This creates a system of work that makes decisions and executes tasks like a highly-trained human team would. Think of Systems of Agents as autonomous AI entities that will collapse traditional software categories into one cohesive system; they’re digital workers that understand and act on natural business communication.

Case Study: Regie reimagines sales prospecting

Consider how traditional sales development works: SDRs spend countless hours manually writing emails, tracking responses, and trying to personalize outreach at scale. They juggle multiple tools—one for intent data, another for email sequences, another for tracking engagement, and yet another for task management. Each system has its own interface and data silo; the whole operation requires constant human attention to maintain.

Our portfolio company Regie‘s System of Agents completely transforms this workflow. It starts at the true beginning of the sales process—the moment any prospect shows buying intent. The system automatically synthesizes data from multiple sources, including company priorities (like public financial documents), buyer personas, and previous content interactions. Instead of requiring SDRs to manually craft emails and monitor responses, the system generates highly personalized outreach, understands prospect engagement (like email opens and web visits), and autonomously determines the next best action.

What makes this a System of Agents is the way it collapses traditional software boundaries. It doesn’t just provide better email templates (System of Engagement) or analyze prospect behavior (System of Intelligence); it actively orchestrates the entire prospecting workflow. When a prospect shows high intent, it can operate in “Copilot” mode, drafting personalized content for SDR review. For lower-intent prospects, it runs in full “Autopilot,” autonomously managing thousands of accounts while creating tasks for human SDRs only when their intervention is valuable. The results are already there. One company achieved 3x more prospect touches and 2.5x more meetings with just one-third of their previous SDR headcount.?

Systems of Agents revolutionize the entry point, replacing human processes at their source and capturing data where it’s born—not where it’s recorded. The interface is reimagined entirely and fades in importance: no more forms or fields, just an organic workflow where data capture happens automatically. For instance, when a sales team discusses a new opportunity over Zoom, the System of Agents automatically creates and populates a deal record without anyone touching the CRM. Most critically, such systems result in superior data, generating richer, more comprehensive information including both structured and unstructured content.

One of our portfolio companies, Oliv AI, tackles that data portion of the sales process. Its System of Agents captures and records sales data for account executives across any platform where they’re engaging with prospects, automating hours of tedious work. Sales managers then have a clearer picture of what’s going on in each deal, focusing less on chasing information and more on high-leverage tasks like closing deals or coaching their teams.?

Collapsing the stack into one cohesive system

What makes a System of Agents so revolutionary is not only its ability to perform tasks once requiring highly skilled humans (a concept we call Service-as-Software), but also its capacity to absorb and replace every layer of a traditional enterprise software stack.?

Unlike previous approaches that add complexity by adding layers of tech, Systems of Agents streamline and unify. They eliminate the need for siloed tools like CRMs, marketing platforms, or analytics dashboards by collapsing these functions into a single cohesive system. This isn’t just a new category of software; it’s the dismantling of enterprise software as we know it. Below we’ll detail how they will pull off this task.

Replacing Systems of Intelligence

Systems of Intelligence are the first to be dismantled. They were built on a foundation of manually-entered data and basic pattern matching. Systems of Agents transcend these limitations by understanding the full context of business communication in real-time. They don’t just analyze data—they understand intent, make decisions, and take action.

This consolidation fundamentally reimagines the interface. Software interfaces you know today—forms, buttons, dashboards—are built around how computers need to receive information, not how humans naturally work. AI agents will shatter these constraints.

Think about humans’ core work—they do data entry in Systems of Record, engage with customers through Systems of Engagement channels, and receive analysis via Systems of Intelligence tools. These are artificial software boundaries created from the way software has evolved; software evolved to make humans more productive, not to carry out human tasks.?

With Systems of Agents, these traditional boundaries between data entry, engagement, and analysis disappear. Work flows naturally, and software adapts to humans, not the other way around. The disruption will be particularly severe for systems that have resisted modern user interfaces or API connectivity. These systems, often deeply embedded in enterprise workflows, have survived because of switching costs and network effects. But when Systems of Agents can bypass them, capturing data at the source and providing better insights without manual input, their moats will quickly dry up.

Consider sales forecasting: instead of waiting for reps to update opportunities, an AI agent can spot when the CFO gets cc’d on an email thread, when legal starts redlining documents, or when technical requirements start circulating, automatically updating deal probability and next steps without any manual input.

Absorbing Systems of Engagement

The original Systems of Engagement emerged to make rigid systems more user-friendly. But as communication channels expanded, from email to chat to video calls, the programs became their own data silos. How many times have you found yourself hunting for something that lived only in Slack? Systems of Agents eliminate this problem by becoming the engagement layer itself, capturing and synthesizing information across all modes of communication without requiring human translation.

Consider how ConverzAI transforms seasonal hiring. Traditionally, staffing agencies face a paradox during peak seasons: they need to hire more blue-collar workers precisely when they lack the recruitment capacity to do so. The old solution? Hire more recruiters temporarily—creating a recursive staffing problem. ConverzAI’s System of Agents not only eliminates the need to hire double recruiters but also eliminates the need for existing recruiters to use half a dozen software programs. It does so with a System of Agents that handles the full communication flow, from initial outreach to phone screens, texts, and follow-up emails, knocking out software like Greenhouse, Workday, and Zoom. Rather than just providing a better interface for recruiters to manage candidates, the system becomes the recruiter, understanding and engaging with candidates directly.?

Consuming Systems of Record

Last in line for dismantling are Systems of Record, which can prove the stickiest. Think about your typical enterprise CRM or ERP system. It is essentially a glorified database with forms on top. When Systems of Agents can understand natural communication and automatically extract relevant information, these rigid, form-based systems become unnecessary intermediate steps. Their core value proposition—being the structured source of truth—evaporates when AI can create structure from natural human workflows. While System of Records are deeply embedded in enterprises, they are also expensive. And when buyers can stop paying for what is essentially a commoditized system, they will.??

An enterprise-grade EMR (Electronic Medical Record) or EHR (Electronic Health Record) system requires medical staff to manually translate every patient interaction into structured fields. When a patient calls or a fax arrives, staff must parse the information, enter it into the EHR, verify insurance in another system, and schedule appointments in yet another, a fragmented process that creates delays and errors.

Tennr‘s System of Agents reimagines this workflow. The moment patient information enters the practice—through faxes, emails, calls, or forms—Tennr uses AI trained on millions of medical records to instantly process information, without waiting for staff to enter data into the EHR system. This is not only faster but generates richer, more accurate data because it understands the full context of patient interactions.

Our portfolio company Eightfold is challenging traditional HR Systems of Record, such as Human Resources Information Systems (HRIS) and Applicant Tracking Systems (ATS). These systems are basically fancy digital filing cabinets, documenting and processing talent activity. Eightfold, on the other hand, employs a proactive, AI-driven System of Agents to match individuals with job openings more accurately. It can predict career trajectories, suggest learning and development paths, and aid organizations in making data-driven decisions about hiring, developing, and growing talent. Eightfold cuts hiring time and costs, reduces employee attrition, and creates a more agile workforce, ready to adapt to changes.

As agents become the de facto point of data entry, traditional Systems of Record are reduced to commoditized storage solutions. Instead of merely storing structured data, Systems of Agents will evolve into dynamic, AI-integrated repositories that capture both structured and unstructured data, with intelligence built in, not bolted on.

The age of siloed systems is over

Systems of Agents will begin by automating human workflows, initially coexisting with all existing systems. As they capture and act on information at its source, understand the full context of business communication, and continuously learn from every interaction, they’ll generate richer, more accurate data than traditional systems ever could. The old boundaries between data entry, engagement, and analysis won’t just blur—they’ll become irrelevant. The winners in this new era won’t be those who build better interfaces or smarter analytics. They’ll be those who create systems that think, learn, and act with the fluidity of human teams while operating at machine scale. They’ll be those who recognize that the future of enterprise software isn’t about making better tools; it’s about creating digital workers that truly understand and enhance how business gets done.

The era of separated systems is ending. The age of Systems of Agents has begun. SAP, Oracle, and Salesforce, beware.?

If you’re building something groundbreaking with Systems of Agents, email us at [email protected] and [email protected].?

Published on December 3, 2024. Written by Foundation Capital

• Facebook
• Twitter