Cool convo with Naza Metghalchi last week when we discussed what are the types of automation that could be opportunities for challengers 🦸♀️🦸♂️ ( 👋 Qevlar AI | We're Hiring! Parloa ) vs incumbents 🐘 The key for me was to really think around what agents can actually do that was not possible in the past (and not just make easier something that we could somewhat do with effort already). Agents are changing the game if you think about a non sequence-friendly automation: like with potentially « infinite » branches and loops. This unlocks goal oriented automation when in the past we had « steps/actions flow » oriented automation. Exciting 🚀
💡 I was discussing with Julien Hobeika about agents the other day. We had a typical VC question. What is gonna be grabbed by incumbents and what could be opportunities for newcomers. This conversation is endless but we started slicing the problem with 2 axis: Where automation sits in the enterprise: 🅰️ -"Top-down" sponsored automation. These are high volume areas and if not met properly can have business critical risk 🅱️ - Long tail user productivity, often underserved and sponsored by individual users or small teams, requiring a more DIY approach. I have around 10 automation ideas daily, but as a non technical user, I find the barrier to entry for using Zapier effectively to be non-trivial The complexity of automation: 1️⃣ - Sequencing friendly workflows with straightforward chains and finite branches 2️⃣ - Complex, non-sequencing friendly workflows that require a goal-oriented approach with potentially infinite branch and loop combinations. ➡️ Our initial hypotheses point towards: 🐘 There is an opportunity for incumbents (Workato, UIPath) to use genAI to add even clever 'steps' in their tool AND potentially ease the creation of workflow using "prompt to workflow" mechanism. This is already happening. I've been impressed by how aggressively they have been implementing more capabilities into their offerings 🦸🏻♀️🦸 This is a new paradigm. Existing sequencing friendly automation tools are failing at this. A new agentic native approach can build those automations. We’re excited about founders building here and have already partnered with teams like Qevlar AI | We're Hiring! (in security operations), and Parloa (in customer service). Who’s next? 😉 🪄 There is an opportunity to bring automation to the masses. We believe that delivering prompt-to-sequence automation to every individual holds a lot of promise. We expect teams to create a magical product experience enabling anyone in the organization to become a software creator 🤖 This is the more futuristic box. Is that basically automating yourself? Get your digital twin to work for you and you earn the money on its behalf? We will keep ideating on this framework. What do you think? Doreen Huber Tom Mendoza Sai Sriramagiri Ted Persson EQT Ventures Kaushik Subramanian
Cyber Insight - Measure, Manage, Mitigate Risks
1moYes, we agree, but with current models it needs to be handled with care. Complex workflows require many thought and action processes, as well as a larger context, which can increase the likelihood of hallucination effects in LLM-based agents. We reduce hallucination effects by combining agents with redundant analytic / ML methods through a solution space constraint. This leads to a different quality of results for our AI architecture than what we currently see in the market. If you want to learn more, feel free to drop me a line.