Wordware (YC S24)’s Post

Transcript

What were enables anyone to build functional AI agents with natural language programming? But I know what you might be thinking. What the **** is an AI agent? Rag, hallucinations, multi agent collaboration, it's all pretty confusing. Right. Does it make you feel bored or stupid? Well, it's supposed to. Techies from Silicon Valley like to use confusing terms, make you think only they can do what they do. Or even better, for you to leave them the **** alone. So here is Margot Robbie in a bubble bath to explain. OK, we couldn't actually get Margot Robbie. So we got Robert, our CTO, instead. Basically, AI just mindlessly guesses the next word, like a freshman pulling all nighter to write an essay. No plan, no research, no review, just pure chaos. But AI agents? It's like training a group of these freshmen and giving them time to become graduate students. Little brighter, but still needing clear instructions and the right resources. For now, to make this group actually helpful, they need a manager. You need to give them the structure and point them to the right resources. That is prompt engineering. And here's the beautiful thing. There are only four tricks to make these students work their magic. First, make them plan their essays before diving in. No more freshman style panic writing. Second, get them to stop and reflect as they work. Is this making sense? Do we need more data? Third, multi agent collaboration. Get your econ grad to crunch the numbers, the comps eye wears to hand all the technical stuff and the English major to pull it all together. 4th. Give them the right tools, whatever they need to actually back up their work. That's it. That's an AI agent. 4 simple things. Planning, reflection, multi agent, collaboration and tools. Got it. Good. Now ****. Sounds simple enough, right? But until now, building a I meant assembling an entire engineering team. Mom spent coding and days needed for each small iteration. That changes today. We've created something entirely new. Now English is the programming language accessible for everyone. Get powerful enough to satisfy your CTO. Let me show you our editor. This is your canvas for orchestrating AI. Here is a prompt. And with Slush we can access all the features of Worldware. For now, let's use one generation. OK. Let's run it. That's pretty simple. Now let's add multiple generations. Change models. And add some tools. Run. We now get a different version of Robert script. Play. Alright, so you've heard of AI, but what about AIA? OK enough. From here it's simple. You can deploy it as an API or your back end, or turn it into a workflow. Oh, and we also support code execution within flow structure, generations, loops, conditional statements, type system for multimodal interactions, ROG, and everything else you might want from the ultimate AI operating system. I say yeah becomes more powerful. Communicating our creative vision to it becomes more nuanced, not less. And that's prompt engineering. We are enabling the next 500 million people become AI engineers. Build it, iterate on it and ship it. Or with water. Ohh I'm jacked. I'm jacked. Just as dead. Alright, we got that, Jesus.