Capably

Over the past few months, we at Prosus Ventures together with Prosus AI have extensively explored the Agent and AgentOps realm, and created two accompanying mappings as part of this: 1) An Agent Landscape and 2) AgentOps Ecosystem. Please check out our full blog post for more insights: https://lnkd.in/e_NmBr9Z What’s in these mappings?   💡Building useful AI agents can be challenging due to factors like tech readiness, scalability, and access to tooling. However, once an agent works well, it can become an invaluable asset to a given team. While previously, automation happened via a rule-based / pre-programmed approach, agents today can execute each task individually, accounting for small nuances between tasks.   🤖 Agent landscape. Here we differentiate between 3 agent categories: -- General agents often address the individual as a personal assistant (such as Adept, MultiOn or Cognosys). We already see lots of players building in this category, competing head on with OpenAI -- Function-specific agents are fine-tuned for a specific task (such as Ema Unlimited, 11x.ai or PolyAI) -- Vertical agents are fine-tuned for an industry (such as Defog.ai (YC W23) or Hippocratic AI) In our experience, agents get better if built for a specific domain or a narrower set of tasks. This is an area we are closely watching for potential investments.    🔧 AgentOps Ecosystem. These are the tools that remove technical barriers, enabling Agent development and scalability to ultimately foster advanced and efficient systems. Investing in this category offers exposure to diverse agent opportunities without the need to focus on a single use case. -- Agent-specific tools: Frameworks such as LangChain or Hugging Face Agents allow developers to get off the ground faster. Multi-agent systems such as AutogenAI enable agents to work together to achieve goals beyond the capabilities of individual agents. Honu aims to take this a step further by proactively initiating requests to fulfill strategic business objectives, mimicking a manager or even CEO -- General ML Ops: these tools find application across genAI more broadly. LLM routers such as Martian allow developers to integrate various models and redirect each task to the best LLM for a specific use case. This is important for AI agents, where an error in the action plan can lead to an incorrect output. When an LLM is involved in executing steps of an action plan, these errors can propagate through each step The space is still in its infancy and rapidly evolving, but we are excited about the disruptive potential of this technology. 🚀 If you are a founder building in the space, please do reach out to myself and Sandeep B.! Thanks Javier Valverde, Euro Beinat, Paul van der Boor, Ioannis Zempekakis, Nishikant Dhanuka, Ahmed Mohamed, and many others for your contributions! Thanks to Sifted and Tim Smith for covering our agent scan!

Over the past few months, we at Prosus Ventures together with Prosus AI have extensively explored the Agent and AgentOps realm, and created two accompanying mappings as part of this: 1) An Agent Landscape and 2) AgentOps Ecosystem. Please check out our full blog post for more insights: https://lnkd.in/e_NmBr9Z What’s in these mappings?   💡Building useful AI agents can be challenging due to factors like tech readiness, scalability, and access to tooling. However, once an agent works well, it can become an invaluable asset to a given team. While previously, automation happened via a rule-based / pre-programmed approach, agents today can execute each task individually, accounting for small nuances between tasks.   🤖 Agent landscape. Here we differentiate between 3 agent categories: -- General agents often address the individual as a personal assistant (such as Adept, MultiOn or Cognosys). We already see lots of players building in this category, competing head on with OpenAI -- Function-specific agents are fine-tuned for a specific task (such as Ema Unlimited, 11x.ai or PolyAI) -- Vertical agents are fine-tuned for an industry (such as Defog.ai (YC W23) or Hippocratic AI) In our experience, agents get better if built for a specific domain or a narrower set of tasks. This is an area we are closely watching for potential investments.    🔧 AgentOps Ecosystem. These are the tools that remove technical barriers, enabling Agent development and scalability to ultimately foster advanced and efficient systems. Investing in this category offers exposure to diverse agent opportunities without the need to focus on a single use case. -- Agent-specific tools: Frameworks such as LangChain or Hugging Face Agents allow developers to get off the ground faster. Multi-agent systems such as AutogenAI enable agents to work together to achieve goals beyond the capabilities of individual agents. Honu aims to take this a step further by proactively initiating requests to fulfill strategic business objectives, mimicking a manager or even CEO -- General ML Ops: these tools find application across genAI more broadly. LLM routers such as Martian allow developers to integrate various models and redirect each task to the best LLM for a specific use case. This is important for AI agents, where an error in the action plan can lead to an incorrect output. When an LLM is involved in executing steps of an action plan, these errors can propagate through each step The space is still in its infancy and rapidly evolving, but we are excited about the disruptive potential of this technology. 🚀 If you are a founder building in the space, please do reach out to myself and Sandeep B.! Thanks Javier Valverde, Euro Beinat, Paul van der Boor, Ioannis Zempekakis, Nishikant Dhanuka, Ahmed Mohamed, and many others for your contributions! Thanks to Sifted and Tim Smith for covering our agent scan!