A System of Agents: Our view on how founders can jump on a $4.6T opportunity. 👇 When Foundation Capital partner Jaya Gupta and I first outlined the Service-as-Software framework months ago, we knew we were describing something transformative. Now, after engaging with hundreds of startups, the potential is even clearer: the way to put Service-as-Software into action is through a System of Agents. Unlike traditional software systems that passively wait for human input, a Systems of Agents approach reimagines how software operates. Groups of cooperating AI agents aren't just tools; they're autonomous workers—capable of capturing, processing, and acting on both structured and unstructured data with unprecedented intelligence. The agents understand context, make decisions, and continuously improve. The initial Service-as-Software shift was significant. But a System of Agents? This is the paradigm shift that will unlock new, previously unimaginable categories of work. Here are four tips for founders building with Systems of Agents: ✔️ Capture data at the source Traditional software relies on human input, but Systems of Agents thrive when they operate directly at the source of data creation. By owning the interface where data originates, you empower your system to orchestrate downstream actions seamlessly. ✔️ Reframe costs as workforce investments When a System of Agents handles an entire job function, think beyond software budgets—classify it as part of a company’s workforce spending. This positions your solution as something that aligns with personnel costs, rather than operational software expenses. ✔️ Expand workstreams with AI’s 24/7 capabilities AI agents never clock out. Builders can use this round-the-clock reliability and scalability to unlock new opportunities—think of offering constant support in healthcare or real-time operations in logistics and customer service. ✔️ Prepare for pricing and business model changes Embrace the shift from seat-based pricing to outcome-driven models. With Service-as-Software and Systems of Agents, your value proposition is tied not to the number of licenses issued but to the measurable outcomes and improvements your AI delivers. Read our full analysis in the comments 👓
This is not going to be easy. There is plenty to be learned from the RPA era. Somewhat counterintuitive, the biggest challenge is with the "what" and NOT the "how". The "what" being the intricate operational and technical details of business functions and processes. Understanding data and relationships. The devil is in the detail, and that detail is not SQL, APIs, Agent-Agent communication mechanisms. For AI and humans to effectively collaborate, they need to share a common language and understanding of data, relationships and associated detail…and memory.
"fire-and-forget" agents - 'cause most agents are just "packaged LLMs". A better kind of AI Agents are small ones who are doing only one task, repetitive, and can be killed if that task is done, whenever. Data leakage presentation, and data poisoning as examples. And a comm layer is missing, reliable and long-distance ready.
Your missing potentially a layer that we find is critical at raia and that's a workflow engine where you tie the 3rd party apps and business logic into the threads of the Agents. Also maybe not necessarily needed to be pictured here - but also a human in the loop block we find to be almost always present in some form (training, approvals, triggers). Great stuff!!!
SwarmZero is being built on the same thesis - swarms of AI agents that complete work autonomously. Right now, vertical AI agents may be first to market, but they will quickly become to difficult to upgrade/customize. A multi-agent platform like ours will not have this problem. A user should be able to create a swarm of agents that solve a wholistic solution end-to-end. It’s much easier when the AI agents are able to communicate with each other on the same platform. If an agent in a swarm is suboptimal, you can always swap out that agent with a better alternative that exists on our AI agent marketplace.
Good observations. Additionally, venture capitalists and founders might place greater emphasis on governance, particularly focusing on IAA: Identification, Authentication, and Authorization. Identity security with traceable audit trails will be crucial for addressing liability and governance concerns. Digital identity wallets for both individuals and legal entities are likely to play a significant role in this process. It's important to note that there's always a human and/or legal entity responsible for the actions of AI agents; accountability doesn't end with the bot itself.
Excellent points on Systems of Agents. At Engini we've built an orchestration platform enabling AI agents to seamlessly interact across complex enterprise systems, including on-premise ERPs like SAP and Oracle. Our focus is on making these traditionally siloed environments accessible for intelligent automation, driving measurable outcomes in efficiency and cost savings. Would love to hear your thoughts on applying this approach to legacy enterprise systems.
I am currently developing this kind of a system. The main challenges include managing the data ingress stream, which sometimes requires investment to make it real-time, and providing clear instructions to agents, which necessitates expertise. Additionally, there needs to be clarity regarding the goals and intent of the entire system. All of this requires individuals who can grasp the entire system end-to-end. If, like me, you can manage this complexity, "selling" the concept becomes another challenge altogether.
The two main challenges(which can and must be overcome) has to do with data quality and interpretation. Data that’s malformed, duplicate or ambiguously formed can cause incorrect outcomes. This can be overcome through rigorous testing and strict definitions. The 2nd problem will be the workforce that needs to do the above data vetting is the very workforce that may be streamlined. How to ensure that they’re bought into the outcome? I’m excited to see the next steps and solving thorny problems like this.
General Partner at Foundation Capital | Investing in early stage applied AI
3mohttps://foundationcapital.com/system-of-agents/