Latest / The Series A Show with Fexingo: Early-Stage Funding, Pitch Decks, and Startup Milestones / How AI Coding Startups Like Cognition Pitch VCs
Transcript
- Lucas: So Cognition's Scott Wu was at TechCrunch Disrupt this week, and he said something that I think is going to cause a lot of hand-wringing in boardrooms over the next few months. Luna: I caught that headline. He said AI coding agents shouldn't replace humans, right? Lucas: Exactly. And look, on the surface that sounds like the right thing to say — responsible AI, human-in-the-loop, all that. But if you're a founder pitching your Series A right now, that kind of framing is actually really tricky. Luna: Because investors want to hear that your product is indispensable, that it's going to change workflows dramatically. 'Not replace humans' can sound like 'mild productivity boost.' Lucas: Right. And we've got real market data showing the market is hungry for disruption. I mean, Palantir is up 14 percent in the last five days. ARKK, the innovation ETF, up over 7 percent. The appetite for ai native companies is clearly there. Luna: And if these conversations are useful for what you're building or running, you know, listener support is what keeps this show ad-free and independent. Lucas: Yeah, it's a small thing that makes a big difference for us. You can find us at buy me a coffee dot com slash fexingo. Luna: And back to Scott Wu's point — he's essentially trying to thread a needle. On one hand, you need to promise enough automation to justify a massive valuation. On the other, you can't scare off developers who are your end users. Lucas: That's the core tension. And it's not just Cognition. We saw this week that Nvidia did what is being called a '20 billion dollar not aqui hire' — essentially acquiring an AI chip startup's talent without calling it an acquisition. And then Groq, another AI chip company, is reportedly raising 650 million. The whole ecosystem is frothy. Luna: So for a startup like Cognition, which builds an AI coding agent called Devin, the pitch has to walk this line. How do you actually frame that in a deck? Lucas: Yeah, let's get into the specifics. I've been looking at how these companies are positioning themselves. There are basically two camps: the 'co-pilot' camp and the 'autopilot' camp. Co-pilot means you assist the developer — you're like GitHub Copilot on steroids. Autopilot means the AI takes the wheel and the developer just reviews. Luna: And Cognition is squarely in the autopilot camp — Devin is marketed as an AI software engineer that can take a ticket and ship code. But Scott Wu is now saying, 'hey, don't replace humans.' That feels like a pivot. Lucas: I think it's more nuanced. The product demo shows Devin doing independent work, but the messaging has to account for the real-world deployment. No enterprise is going to let an AI agent push code to production unsupervised. So the pitch becomes: we automate the boring stuff, but humans stay in the loop for the critical decisions. Luna: That's a classic enterprise sales move — reduce the perceived risk. But does it work for a Series A pitch? Investors want to see a path to massive margins. If you're saying 'we need a human reviewer for every task,' that sounds labor-intensive. Lucas: Exactly. So the smart founders I've seen pitch this space use a different framing. They say: 'Our agent handles 80 percent of the work automatically. The developer's role shifts from writing code to reviewing and architecting. That still delivers a 5x productivity gain, which means you need fewer developers, or you can ship faster with the same team.' Luna: That's stronger. It's still a 'replace' story — just replace the bottom 80 percent of tasks, not the whole job. Lucas: Right. And then you back it with metrics. I saw one deck from a similar startup that claimed their users saw a 40 percent reduction in time to merge for pull requests. That's concrete. That's what VCs want to see. Luna: But here's the thing — a lot of those metrics are early, from beta users who are already tech-forward. How do you convince a VC that this scales beyond the early adopter crowd? Lucas: That's the million-dollar question. One approach I've seen work is to show the improvement curve. Like, in month one, the agent only handled 30 percent of tasks autonomously. By month six, after retraining on your codebase, it's at 70 percent. That shows the product gets better with usage — it's a network effect with your own data. Luna: That's a good pitch because it creates a switching cost. Once a company has invested in training the agent on their codebase, they can't easily leave. Lucas: Exactly. And the best founders I've talked to are also very clear about what the agent can't do yet. They'll say: 'Our agent is terrible at greenfield architecture decisions. It's great at implementing well-specified tickets.' That honesty actually builds credibility. Luna: So it's similar to the 'co-pilot vs autopilot' framing, but instead of promising full autonomy, you're promising excellence in a specific slice. That's more believable. Lucas: Yeah. And it maps to how VCs think about total addressable market. If your agent can only handle bug fixes and small features, that's still a massive market — every company has a backlog of tickets. You don't need to claim you're replacing all developers to build a billion-dollar company. Luna: Right. So for founders listening: when you're putting together your Series A deck, avoid the 'AI will replace everyone' pitch. Instead, focus on the specific job to be done, show the productivity gain with real numbers, and be honest about where the technology is limited. Lucas: And one more thing — the market backdrop really matters right now. With stocks like Shopify up nearly 15 percent in five days, and Rivian up 14 percent, there's clearly risk-on sentiment. AI startups are in a sweet spot where VCs are allocating heavily. But the bar for revenue projections is getting higher. We're seeing more VCs ask for unit economics even at the pre-revenue stage. Luna: So it's not enough to say 'AI is the future.' You need to show a path to gross margins above 70 percent, preferably with a services angle that gets you to 80 percent. Lucas: Yeah. And one clever thing I've seen is startups bundling their AI agent with a managed service. So you pay a subscription for the agent, but also a per-seat fee for human oversight from the startup's own engineers. That adds a services revenue stream that VCs love because it's recurring and high-margin. Luna: But doesn't that make you a services company, not a software company? VCs typically apply lower multiples to services revenue. Lucas: That's the tension. The best founders I've seen pitch this actually separate the two. They say: 'Our software subscription is 80 percent gross margin. Our services layer is 50 percent gross margin, but it's a gateway to get enterprises to adopt the software.' Then they show that over time, the services percentage shrinks as the agent gets better. Luna: That's a compelling narrative — you're using services to build a moat and collect data, then the software improves and you phase out the services. It's like how Salesforce started with implementation consultants. Lucas: Exactly. And the data backs it up. Look at the ARKK ETF — it's up over 7 percent this week, signaling that the market is betting on innovation stocks again. That's the tide that lifts all boats, but you need a strong pitch to stand out. Luna: So to wrap this up: if you're a founder building an AI coding agent, the playbook for your Series A pitch is: define a specific job the agent does well, show real productivity metrics with a learning curve, separate software from services revenue, and be honest about limitations. That's how you thread the needle between ambition and credibility. Lucas: And don't forget to check the market window. With Nvidia's massive talent move and Groq's upcoming raise, the space is getting crowded. But the winners will be the ones who can articulate a clear, defensible path to adoption — not just a vision of full automation. Luna: Right. And Scott Wu's comment, while cautious, might actually be the smartest positioning in the long run. Because if you promise replacement and fail, you lose trust. If you promise augmentation and deliver, you build a long-term relationship. Lucas: That's the thing about this market — the hype is real, but the scrutiny is just as real. Founders who can navigate that with a grounded pitch are the ones who will close their rounds.