From Systems of Intelligence to Systems of Agents: The New Moats in Enterprise Software

In 2017, Jerry Chen coined the term Systems of Intelligence . Today, we are suggesting that this changes to systems of agents. The term "agents" has been thrown around for quite a bit, but if we take a step back to truly understand what the new technologies fundamentally change and what is the "why now" to new moats to exist - it's the difference between collecting garbage data to now not only collecting and unlocking better data, but in particular, collecting interactions and actions of humans - what we like to call learning from action spaces.

The Evolution of Enterprise Systems & Their Shortcomings

To understand where we're headed, we need to look at where we've been. Over a decade ago, the enterprise software world was grappling with a shift from "systems of record" to "systems of engagement." This transition exposed critical limitations that persisted even as we moved towards systems of intelligence. Systems of Intelligence, while promising, fell short of their transformative potential due to three key limitations:

1. Incomplete Data: Time-pressed employees often neglected data entry, leading to gaps in systems of record. Shockingly, this problem was so pervasive that in 2012, a TowersWatson survey revealed that 58% of HR organizations would consider switching to Workday primarily for its user interface. Think about that - companies were willing to undergo massive, multi-year migrations just to make their systems more user-friendly and increase data input!
2. Quality Issues: Even with more engaging interfaces, the fundamental issue of data quality remained. Manual entry resulted in errors and inconsistencies, embodying the "garbage in, garbage out" principle. The promise of systems of intelligence was limited by the very data they relied on.
3. Untapped Unstructured Data: Perhaps most critically, valuable insights in emails, calls, and meeting notes remained inaccessible. Traditional systems, even as they evolved, struggled to tap into this wealth of unstructured data where most valuable insights often reside.

These constraints, rooted in the very systems they aimed to enhance, hindered the true potential of intelligent business operations.

From Intelligence to Agents

Why is there a shift? We have a few better elements:

1. From system of record silos to knowledge ecosystems: Unstructured data is no longer the dark matter of business intelligence. LLMs are unlocking vast amounts of insights previously hidden in emails, documents, and conversations. This shift from system of record dependence to holistic information synthesis allows us to capture the entire ground truth of business operations and customer interactions.?
2. Breaking free from the chains of the system of records: While systems of record remain valuable, they're no longer the sole source of truth. The ability to process unstructured data liberates us from the constraints of rigid data entry into these systems. We're moving from "input everything or know nothing" to "understand everything, input what matters." This flexibility reduces the impact of incomplete or inaccurate data entry in systems of record, allowing for a more nuanced and comprehensive view of business realities. It also enables us to extract more from the systems of engagement. It's a far cry from the days when 58% of companies would switch systems just for a better input interface!
3. Workflows that adapt, not dictate: With reduced dependence on traditional systems of record, we're no longer forced to fit square pegs into round holes. The newfound flexibility allows us to craft workflows that mirror actual business processes, not just systems of record structures. It's a shift from "process designed for data entry" to "data molded to process," enabling more efficient and intuitive operations. For example, today sales and revenue operations teams download data from Salesforce, manipulate it and cleanse it to create sales forecasts for executives, but there are many unstructured data inputs from the systems of engagement we can now ingest to have real time sales forecasting.?
4. Harnessing the collective intelligence of the workforce: The new frontier of operational insight lies in understanding both individual actions and collaborative dynamics. Every click, decision, and interaction – whether with systems or colleagues – becomes a learning opportunity. By observing how employees navigate various systems and collaborate with each other, we gain insights into the unwritten reasoning behind their actions and the synergies that emerge from teamwork. This shift from prescriptive to adaptive systems allows us to create workflows that don't just accommodate individual work and teamwork – they actively enhance both. Whether it's human-to-human or human-to-agent collaboration, every interaction enriches our understanding, capturing the tacit knowledge and collective wisdom that often eludes traditional process documentation.
5. Work begins where reality unfolds, not where data is stored: Agent systems are redefining the starting point of business processes. By operating across multiple platforms and data sources, they're shifting the paradigm from "record first, act later" to "act now, record seamlessly." This approach allows businesses to initiate and complete tasks in real-time, tapping into the rich, dynamic ecosystem of information that exists outside traditional systems of record. It's not about reimagining workflows; it's about embedding intelligence directly into the fabric of daily operations, turning every interaction into an opportunity for insight and action.

This is why we believe that we are now at systems of Agents: autonomous AI entities that interact with systems of record, systems of engagement, and even other agents to perform complex tasks and workflows.

The New Moats: Vertical-Specific Agent Systems followed by Downstream Automation?

We believe the most successful agent systems are those tailored to specific industries or business functions. Across industries – from financial services to logistics, from healthcare to legal – we're seeing AI-driven systems earn the right to automate and optimize entire workflows by solving the initial data synthesis challenge.?These vertical-specific agents are trained on industry data, understand domain-specific rules and workflows, and integrate seamlessly with existing systems. For example:

Healthcare Administration: Companies like Tennr are revolutionizing healthcare administration by addressing the overload of communication from patients. They've trained LLMs on millions of medical records to accurately extract patient information from various sources like faxes, emails, and call transcripts. This allows them to automate patient intake, insurance verification, and scheduling processes, significantly reducing administrative overhead for medical practices.

Financial Services /Insurance: In the lending industry, companies are developing AI agents that can synthesize loan files from brokers, streamlining the underwriting process for business lenders. These agents can quickly extract relevant information from various documents, speeding up decision-making and reducing costs. Companies like Fulcrum are tackling some of the use cases here with proposal generation, policy comparison and policy checking.

Supply Chain : AI agents in supply chain management can analyze vast amounts of data from various sources - including historical sales data, market trends, weather patterns, and social media sentiment - to provide highly accurate demand forecasts. These agents can then optimize inventory levels across multiple locations, balancing the need to meet customer demand with the desire to minimize carrying costs. Companies like Ikigai are building here.

Logistics and Transportation : In the freight industry, AI agents are being employed to automate quoting and pricing processes for brokers, shippers, and carriers. These systems can analyze market conditions, historical data, and current demand to provide accurate and competitive pricing in real-time. Companies like Vooma and Levity are building here.

The true power of this shift lies in its ability to unlock downstream automation. By synthesizing unstructured data at the outset, we're unlocking a cascade of automated workflows downstream. Consider sales forecasting: traditionally, sales teams would manually input data into CRM systems, managers would spend hours in pipeline reviews, and executives would struggle to get an accurate picture of future revenue. Now, AI can extract and synthesize information from a variety of sources – emails, calendar invites, chat logs, and even recorded sales calls. This initial synthesis triggers a chain reaction of efficiency throughout the sales process. By capturing and understanding the full context of business operations from unstructured data, these systems can initiate a series of downstream actions that were previously unimaginable.?

Path forward?

As we look at this evolution, from systems of record through systems of engagement to today's emerging systems of agents, we see a clear trajectory. Each step has brought us closer to software that truly serves the needs of businesses and their employees. The future of enterprise software isn't just about better interfaces or more comprehensive data collection. It's about creating systems that can think, learn, and act - systems that don't just record or engage, but actively contribute to the success of the business by automating entire chains of workflows.

Reach out to [email protected] if you are building in this space!