Co-founder and CEO of Zapier, the no-code automation platform connecting thousands of web apps, which he launched in 2011.
Wade frames the AI automation space, now crowded with vertical and dev-centric players, as a trillion-dollar opportunity over the next decade.
“everybody's kind of trying to get a piece of this. What is, you know, going to be, you know, a trillion-dollar opportunity in 10 years?”
Asked how big Zapier gets in a decade, Wade predicts the company should be well past a billion in ARR if they execute.
“10 years? Well, shoot, I think we should be well past a billion in ARR if we do our jobs right.”
Wade feeds Claude (wired to HubSpot, ZoomInfo, and web search via Zapier's MCP server) a person's name and gets back a quick dossier: public details, whether they're a customer, and CRM notes. Like having Gary whisper in your ear in Veep.
“I usually just like feed Claude some details on the person and then Claude will just return like a quick little dossier that includes, you know, public details about the person, but also like what's going on. Like, are they a customer? Is there any details in our HubSpot account? Like, is there anything that ZoomInfo can tell me just to kind of just like get some quick hitters to be like, hey, is there something I can sort of talk to this person about?”
Steal thisConnect Claude to your CRM and ZoomInfo via Zapier's MCP server, then generate a one-paragraph dossier on everyone you're about to meet.
Wade explains Model Context Protocol as a layer that helps agents find and use tools. Watching Claude work, it visibly tries one endpoint, fails, and tries a more direct approach until the task succeeds.
“it's iterating over a bunch of different endpoints. So it's saying, hey, this didn't work, let me try a more direct approach. Okay, that didn't work. So this is Claude trying to figure out how to go do this task.”
Wade's go-to move when an AI summary feels generic: ask it to be 100x more specific. He stole it from Whoop's head of product, Hillary, and says it unlocks the nitty-gritty detail.
“I'm like, hmm, like, I don't know that I love this. So then I might just say like, hey, be 100x, you know, more specific. And I stole that from this woman who runs product at Whoop, Hillary. I saw a YouTube video where she did this and I found that to be like, okay, now you start to get like, all right, a lot of really good nitty gritty details inside of this.”
Steal thisWhen an AI answer is too generic, reply 'be 100x more specific' before doing anything else.
Wade's observation of power users: they don't fire off their question immediately. They feed warmup questions to load context and memory, then ask the real thing, like 'where are my blind spots?'
“they don't just like ask their question right away. They usually ask like these warmup questions where they're trying to like gather a bunch of context, get a bunch of specific information going. And then once they sort of have enough of like things in memory for ChatGPT, then they start asking like the real things they wanna know, which is like, where are my blind spots?”
Steal thisFront-load 2-3 context-building prompts before asking the AI the decision you actually care about.
Wade's co-founder talks back and forth out loud on long dog walks, then feeds the transcript to ChatGPT and asks it to write a strategy memo from all that context, rather than asking it to write the memo cold.
“my co-founder goes on like long walks with his dog and he will literally just talk back and forth, back and forth, back and forth. And then he'll take the transcript of that and then input that and say, hey, based on all of this, now I want you to go write a strategy memo on XYZ thing versus saying, hey, just write the strategy memo on this.”
Steal thisVoice-dump all your raw context first, then have the AI generate the deliverable from the transcript.
Wade builds a Zapier Agent live: triggered by a new Gmail email, it detects job-inquiry emails and drafts a polite, brief reply pointing to the careers page, saving as a draft for human review rather than auto-sending.
“let's make an agent that replies to your email. And actually I don't want it to reply to my email because I'm a little scared that it might reply in a way that is not up to my standards. So what I actually want you to do is make drafts for my emails.”
Steal thisWire a trigger-based agent to draft (never auto-send) replies to your most repetitive inbound emails.
Zapier's company value that primes employees to treat automation as a core primitive, framing AI tooling as something everyone builds rather than fears.
“We have a company value that's don't be a robot, build the robot. So we're literally trying to teach people that automation is a core primitive.”
Zapier got AI usage from zero to ~90% of employees daily via three moves: a one-week company-wide code-red hackathon, mandatory show-and-tell for accountability and knowledge sharing, and repeating shorter hackathons every 3-6 months.
“first, we called a code red and we did a hackathon where I stopped the company for an entire week and I said, I don't care what job you're in, you know, if you're in HR, accounting, or support, or sales, or engineering, we're all going to press pause for the whole week and we're going to go build stuff with AI.”
Steal thisPause the whole company for an AI hackathon, require show-and-tell, then repeat shorter versions every 3-6 months.
Zapier got AI usage from zero to ~90% of employees daily via three moves: a one-week company-wide code-red hackathon, mandatory show-and-tell for accountability and knowledge sharing, and repeating shorter hackathons every 3-6 months.
“first, we called a code red and we did a hackathon where I stopped the company for an entire week and I said, I don't care what job you're in, you know, if you're in HR, accounting, or support, or sales, or engineering, we're all going to press pause for the whole week and we're going to go build stuff with AI.”
Steal thisPause the whole company for an AI hackathon, require show-and-tell, then repeat shorter versions every 3-6 months.
A Zapier talent-team member (non-engineer) built a workflow hooking into Ashby, Slack, IP/phone APIs and more to score each applicant's fraud risk by cross-checking metadata against other applicants. Ten years ago this would have required ML engineers.
“it tries to score their risk on is this applicant potentially fraudulent. And so effectively, you know, you get an applicant that comes in, it takes the details of what came in from the applicant, and then it runs checks on the IP address, phone numbers, and a whole bunch of other metadata, and then it compares them to other applicants to try and spot mismatches or suspicious patterns or things like that.”
Steal thisBuild a workflow that cross-checks applicant IP, phone, and metadata against your other applicants to flag fraud risk.
Wade's parting advice: the biggest advantage is just trying stuff. People are paralyzed by fear of egg on their face, but usually when you fail nobody even noticed or cared, so try again.
“I think most people are so scared that they're going to have egg on their face. But usually what happens is that when you fail, nobody even noticed. Nobody even cares. Like, that's what usually happens. And so if you mess up, who cares? Nobody saw it. Like, try again.”