It’s fun to watch folks without engineering backgrounds go DIY with AI taking the lead. My friend Austin Johnsen leads CorpDev at Zapier and in recent months I noticed his LinkedIn posts detailing the tools he built for the job. So I figured it would be a good opportunity to learn more about what he’s discovering. And thus, Five Questions.
Hunter Walk: How do you explain Zapier to people not in tech, how’d you end up there, and what’s your role there?
Austin Johnsen: I usually tell people that Zapier connects all the apps you use so they can talk to each other and do work for you automatically. If you’ve ever wished that when something happens in one tool, something else would just happen in another tool without you doing it manually, that’s what Zapier does. We have millions of users and integrate with thousands of apps. Obviously, AI is changing a lot of this, so we’re also pushing hard into AI-powered automation, so that it’s not just “if this then that” but agents and AI steps that can reason through complicated workflows on your behalf. The biggest benefit Zapier has here is that all of the things people wanted to automate with Zapier pre-AI are also all the things people want their agents to take over now, so our wealth of connections, triggers, and actions perfectly position us to connect agents to workflows.
I lead Corporate Development, which means I’m responsible for acquisitions and investments. We do 2-3 acquisitions a year, plus we have a unique, angel-scale corporate VC fund, the Zapier Fund, that does 10-15 investments a year. Before Zapier, I led corporate development at Twitch and Patreon, and I started my corp dev career way back when Twitter was still Twitter. Prior to that, I did stints in investment banking and private equity.
HW: I’m sure you get asked for career advice from a bunch of new grads or other folks who want a career in tech on the ‘business side.’ What advice do you give them?
AJ: I have two versions of this conversation. First, on the corp dev side, I try to be honest with people. These roles are pretty unique and the math is brutal. Most companies, outside the giants like Google, Amazon, etc. have a few people in corp dev, if that. Even those giants, the entire corp dev teams are probably under fifty people. Zapier is just me. We look at tons of deals, but actual deal volume is low. That means there are very few roles, they rarely open up, and when they do, there’s no time to train someone. You have to hit the ground running. If you don’t already have the reps from banking, PE, or one of the larger tech giants, it’s going to be really hard to break in directly.
On the broader business side, the conversation is a bit easier. Companies will always need smart, ambitious people for things like business development (which I think of as partnerships vs. corp dev which is M&A) or finance. However, my one big tip for any of these business side roles is you still need to get close to the product. Nothing bothers me more than people that work at companies and never use their own products. The most valuable business people in tech actually understand what the company builds, how customers use it, and what the competitive landscape looks like at a technical level. You don’t need to be an engineer, but you need enough fluency that engineers trust your judgment. And, while it’s probably still worth going to somewhere like Google and getting that stamp of approval on your resume, if you can find a smaller, high-growth company that will let you operate above your title, that’s going to be more valuable over the brand name every time.
HW: It’s clear you’re AI-pilled – at least in the sense that I see you sharing lots of content around the tooling you’ve built to help you do your job and live your life. What’s one thing about AI that people can’t really understand until they are building with it themselves?
AJ: How fast you can go from “I have no idea how to do this” to “it’s done” and how that changes what you’re willing to attempt.
I’m aggressively not an engineer. I don’t have a CS degree. I pretended to operate a web design business in high school, but I only built like 3 sites in Dreamweaver. I almost failed a project in college that required me to learn the barest minimum of Python. I try all the AI tools and am an early adopter, but I stuck to chat and I avoided touching code. We’ve had people in Zapier pushing this “Cursor for non-devs” idea since last fall and I tried and I was like there’s no way I’m using an IDE. Hard pass.
This January though, I finally gave in and tried Claude Code once Anthropic added it as a tab to Claude Desktop. On day one, Claude Code told me to connect it to my GitHub account and I was like “I don’t have a GitHub account”, so it had me create one. That was my starting point. But the thing about Claude Code is that it’s like LEGOs, you just need to follow instructions, and I’m very good at following instructions. Every time I was confused, I asked it what to do, and it’d walk me through it, step by step: what to type, why I’m typing it, what each piece does. A couple of months later, and I’m now running a full corp dev automation server on my laptop that’s already way more ambitious than anything I first envisioned.
But it’s not just Claude Code. Last week, I used Claude’s Excel tool to build a full liquidation waterfall for an acquisition, which included modeling out proceeds across multiple investor classes, preference stacks, conversion thresholds, the whole thing. That’s several days of work for an analyst. It took less than an hour. I used Claude Chat to do a deep dive on churn and retention data, pulling threads across Slack conversations, internal dashboards, and financial data to build a picture of what was actually happening with customer disengagement. That kind of cross-source analysis would normally be a multi-week project for a data team. I did it in a few hours. And when I tell people this, they always immediately go, “what about hallucinations”, and honestly, people are way too anchored on that. Are the AI responses perfect? No. But are your human-produced analyses perfect? Also no. You still need to check them. But the AI can correct and iterate a million times faster.
None of this was planned from the start. It started with one enrichment script that pulled company data into my deal tracker (replacing and augmenting an existing series of Zaps). Then I realized I could automate triaging the flood of inbound opportunity emails I get each week. Then a calibration system that injects years of deal history into every evaluation so the AI’s judgment is grounded in my past decisions and improves with each deal. Then a data room search engine across thousands of files for a live acquisition. Then a Slack bot anyone at the company can use to get feedback on acquisition ideas that’s a lite version of my full triage workflow. Each piece solved one problem, and solving it revealed the next one.
The thing that finally made it all truly click though didn’t happen until last week when I finally integrated Claude Code with Zapier’s own SDK (shameless plug here, but it’s still in closed beta, so you can’t try it yet). The Zapier SDK allowed me to give Claude Code access to literally everything Zapier connects to, all 8,000+ app integrations. Before, I was always limited by what I could realistically connect (either via native connectors or flaky MCPs). The SDK solves that and also offloads auth management to Zapier (which is truly a nightmare, especially for someone non-technical). Finally, with the SDK, all my context could live in one place for the first time. Everything I use (Gmail, Google Calendar, Google Suite, Slack, Airtable, Granola, Coda, Notion, Glean, Zoom, and on and on) is finally connected and fully accessible and usable by Claude. It’s an absolute game changer.
You can’t understand that compounding effect by reading about it. You have to start building, ship something small, and let the next problem find you.
HW: And how do you think AI will change the nature of CorpDev/Strategy work at companies long term?
AJ: I think a lot about this, and I’m genuinely worried about it.
The analytical work (reviewing opportunities, building models, scrubbing data rooms, writing memos) is getting compressed dramatically. That’s already happening. I use AI daily for deal sourcing, company research, and enriching my pipeline. What used to require a junior analyst pulling an all-nighter is now something I can do myself in a fraction of the time. And the build-vs-buy calculus is shifting fast, because engineering teams can now ship in weeks what used to take months, which means the bar for “we should just acquire this” keeps getting higher. Corp dev teams are going to need to be more tightly integrated with product and engineering leadership, because the window in which an acquisition makes strategic sense is getting shorter. And it’s going to shift back to being more of a GTM decision than a tech decision because customers and traction are going to become more valuable than “I built something unique” in a world where Claude Code can recreate it overnight.
But here’s what keeps me up at night: how do the next generation of corp dev professionals actually get good? When I started in investment banking, I had a crazy boss who had me churn out what he called “2-pagers” (brief company overviews with financials, comps, the works). I was doing maybe 20 a week, each taking a couple hours, all by hand. That work eventually got replaced by CapitalIQ and PitchBook, and now AI is replacing those too. But the reps I got doing it manually are the reason I can fly through financial statements today. I built an intuition for what to look for and what looks right and what looks off that only comes from doing the work hundreds of times. Nobody does things like this by hand anymore, and I honestly don’t know how you replicate that learning. AI is incredible for people like me who already have the expertise and judgment. It’s a massive force multiplier. But for someone just starting out who’s never had the reps? I think it’s going to be an absolute disaster. You end up with people who can generate beautiful outputs but can’t tell you if the answer is right.
Wade, our CEO, has this framework of “drivers vs passengers” – the strong performers are the ones generating ideas and taking action, not just evaluating what AI produces. I think that’s right, but it assumes you’ve already built the foundation to know what a good idea looks like. We haven’t figured out how to build that foundation when AI is doing all the reps for you.
HW: You and I both have MBAs – if you were the Dean of HBS (where you attended), what – if any – changes would you make to the program in order to maintain its relevance going forward.
AJ: One of the things HBS was always clear about is that they were training leaders, not analysts. We weren’t down in the trenches preparing financial statements – we were learning how to read and analyze them, how to ask the right questions, how to make decisions with incomplete information. I actually think a lot of that philosophy carries over into the AI world better than people might expect. The skill that matters most is judgment, and HBS was always focused on building judgment.
Where I’d push the program is on making sure that judgment doesn’t become disconnected from the underlying work. The risk with AI is that you fully outsource your thinking. You ask it to analyze a company and it gives you a beautiful answer, and if you’ve never done the analysis yourself, you have no way to know if that answer is wrong. So I’d pair the case method, which is still the best tool I’ve seen for developing business judgment, with a requirement that students continually build things with AI before they graduate. Not just a single case study about AI, not a strategy deck about AI. Build working products, automations, or tools. Get your hands dirty. The most valuable skill in business right now is the ability to go from zero to one on an idea quickly, and AI makes that accessible to people who can’t write code. HBS should be producing graduates who have that muscle memory.
I’d also push the school to require students to do hard analytical work by hand before they’re allowed to use AI for it. Build a DCF from scratch. Read 50 10-Ks and write up what you found. The value of AI is only as good as your ability to evaluate its output, and you can’t evaluate what you’ve never done yourself. Think about it the way math classes think about calculators, yes, you’ll use them eventually, but you need to understand the underlying math first or you’re just pushing buttons.
Thanks Austin!
Tip DON’T LET AIRLINE RIP YOU OFF ESPECIALLY AS OIL PRICES CLIMB: I’m using Junova to track purchased airline tickets I’ve taking and auto-reclaim credit if the price drops. It was started by a friend and so far has recouped $2000+ of American and United credits for me. Their business model is: service is no cost, but if they successfully get you credit, they charge 20% of the value to your credit card. If you use this referral link, your first $25 of fees (ie $125 of flight credit) is free. Let me know how it works for you!
My Actual Savings This Year!!!