OpenAI crosses the Zapier Barrier
Generative AI is moving and shaking so fast. So I’m going to make this brief, more newsy, and with just a side dish of analysis to go with it so that I can get back to my job hunt.
But it’s a really substantive and opinionated side dish, like mashed potatoes and gravy.
ChatGPT — heard of it?! — now has plugins and can talk to other products. If you want to go deeper on this, I recommend swyx’s instant reaction:
This app store model will be stupendously impactful for folks who want to deploy a chatbot as an all-purpose, multi-modal interface. If you’re a developer who wants to keep their platform options open (source) take a look at Fixie.
But ChatGPT is still a 1:1 product experience. You can be incredibly productive from that perch, but you can’t (yet) create a repeatable process that doesn’t start from a natural language trigger. Even Zapier’s entry on this side of the equation runs one action at a time.
That’s why I think the OpenAI integration on Zapier's side is even more important for deployment at scale. It has been available in beta for at least a few months, but the underlying models (GPT-3 and DALL-E, both likely to be obsolete soon!) have been tested quite heavily.
Non-technical folks can prompt these models in a programmatic manner and use their outputs anywhere. Remember “personalize the pass”? You no longer need to go through the rigamarole of specifying a column for the reason you passed or even writing out your reasoning at all. You’ll just feed the prompt with your notes and an instruction to write a pass email in your chosen tone.
I’m not making a moral judgement. But everyone should expect more of this in the future and less clarity on whether that personalized note was written by a person or by a bot. If you want to know if they’re actually cheering you on from the sidelines, you’ll have to pick up the phone.
That’s just one use case. Do yourself a favor and scroll down the list of Zapier’s OpenAI templates.
We’re still a step removed from using a chatbot to build new recurring Zaps (and thank goodness because we all know that editor interface has always been a little difficult). We’re maybe two or three steps from large scale deployment of systems that can generate new processes in response to fresh data:
When that hits… look out.
Great observations Jeremy. The amount of automation here will push up the volume of sourcing and screening of new deal flow everywhere. Hopefully, this will make things more data-driven too.