You should have built X yesterday and showed me a demo.
What happens when your customer can just ask Claude Code to do your startup’s job? What if the moat you spent two years building gets bypassed by someone texting an AI agent on their phone? And what does a VC actually want to see when a founder walks in pitching an AI company in 2026?
These are the questions we tackled in this week’s virtual event, “How To Help Your Startup Survive (And Thrive) In The Age of Agentic AI.”
We brought on Itamar Novick (founder of Recursive Ventures, a pre-seed/seed fund investing in AI-first companies, serial entrepreneur with two decacorns and six unicorns in his portfolio — and a fund investment of Venture5’s founder) and Sid Bharath (founder of Refound, an AI consultancy that helps businesses redesign themselves as AI-native companies, with roots going back to building GPT-3 powered companies in 2020).
TL;DR: The old playbook for startup defensibility is getting rewritten. Proprietary data and complex business logic still matter, but the real threat isn’t someone vibe coding a competitor app. It’s the customer bypassing your product entirely by asking an AI agent to do the job instead. The founders who win will be building “service as software” businesses in complex, regulated verticals where AI can deliver the outcome directly, not just make humans 30% more productive.
This virtual event was presented by Refound
Refound helps mid-sized companies become AI-native through hands-on audits, implementation, and training, delivering measurable ROI in time saved and revenue generated. We’re practitioners first, having built and worked inside some of the biggest AI companies in the world.
The Real Threat to Your Startup (It’s Not What You Think)
There’s a narrative floating around that anyone can vibe code a competitor to your product overnight. Sid doesn’t buy it. The actual threat is more subtle and more dangerous.
“I don’t think the real threat is someone vibe coding a competitor app. The real threat is when the customer is just using an AI agent, like Claude Code, to do the job that your startup is gonna do for them instead.”
He used a concrete example: You’re not going to build a Webflow clone, but you could just tell Claude Code to spin up a page for you. You bypass the need for Webflow entirely by telling an AI agent to do the task directly.
This reframe matters. It changes the question from “can someone copy my product?” to “can someone skip my product altogether?”
Sid’s litmus test for whether a business is safe:
“If you stripped away your UI entirely, deleted all the screens and buttons, does your business still have value? Because at the end of the day, you’re then just a database holding data.”
The Moats That Still Work (And the New Ones AI Created)
Itamar laid out a framework for what actually keeps startups defensible in the age of AI. The classic moats around marketplaces, network effects, and customer lock-in still hold. But AI has introduced new types of defensibility that founders should be thinking about.
The obvious one is proprietary data. Your AI is only as good as the underlying data, and if you have data that others don’t, you’re off to the races. But the less obvious moat is the customer data flywheel married with reinforcement learning. Users giving thumbs up, thumbs down, and your AI learning from that to deliver better service. The moat? When a new competitor enters, they don’t have that backlog of post-training data.
Sid reinforced this a specific example: Ahrefs has been collecting SEO backlink data for a dozen years. You can’t just tell Claude Code to do a full SEO analysis because you don’t have that data. You still need Ahrefs, even if you access it through an API via an AI agent.
Other moats that both agreed on:
- multi-party trust and coordination (think DocuSign’s network of regulators and signatures)
- regulated industries where certifications and compliance create real barriers
- and deep infrastructure plays like Shopify that handle payment processing and logistics that an agent simply can’t replicate
Itamar’s advice for founders looking to build something defensible:
“Focus on smaller markets, very complex market, highly regulated, lots of IT systems, very complicated business logic. This is, I think, where companies can shine.”
His reasoning is practical. OpenAI and Anthropic are going to prioritize trillion-dollar markets like search and CRM. They’re not going to chase a $20 billion vertical that requires deeply specialized AI. That’s where startups thrive.
From “Software as a Service” to “Service as Software”
The most forward-looking part of the conversation was around what Itamar calls “service as software,” a concept he’s been championing and investing behind at Recursive Ventures.
The shift: instead of selling seats and helping people get 30% more productive, you’re actually delivering the end result with AI agents. You’re doing the job with and for the customer.
Itamar posed a provocation to Sid live on the call: what if Refound took its consulting business, which runs at 10-20% operating margins, and turned it into a service-as-software business running at 60-80% margins by having AI agents deliver AI consulting automatically?
Sid’s answer? That’s already what they’re doing. They built an internal agent called Jarvis that has access to all company data, including Google, LinkedIn, calendars, and drives. When Sid has a conversation with a client, Jarvis picks it up, identifies the pain points, drafts a proposal in their format, and starts planning the project phases.
“Our team is small, but we’re able to work with companies like (our client who is a) multi-hundred-million-dollar cookware company, because we’re utilizing all this internal proprietary agents that we have built to deliver services at a much higher scale.”
Itamar shared two principles he’s seen from his best service-as-software portfolio companies. First: employees are measured by their ability to automate what they do. The internal culture is “what did you automate this week?” Second: they teach their customers that self-service with AI is the future. When end customers understand they can solve problems through AI tools instead of picking up the phone, that’s where the high margins come in.
But there’s a cautionary note. Itamar shared a story about a portfolio company called Symphony AI that tried to deliver legal services as software for startups. It didn’t work. Not because the AI wasn’t good enough, but because legal is a multi-party process. You have to convince the counsel on the other side to work with the AI, and firms like Orrick aren’t interested in redlining contracts with a machine. The lesson: service-as-software works best in two-party interactions, not multi-party negotiations where human trust is still essential.
The New Bar for Founders Pitching AI Companies
Itamar was blunt about what he expects from founders walking into a pitch meeting in 2026.
“A founder shows up to a pitch session and they’re like, ‘we wanna raise money to build X.’ I’m like, what? You should have built X yesterday with Claude Code and showed me a demo. If you’re not doing that, then we’re already out of sync here.”
The bar has shifted. You don’t need to hire three engineers to put out a demo. If your CTO hasn’t built something with AI coding tools before the pitch call, Itamar says don’t even come on the call.
This extends to hiring too. Both Sid and Itamar agreed that when you’re interviewing people at an AI-native company, the question should be “show me what you vibe-coded yesterday.” If candidates aren’t actively experimenting with the latest tools, they’re not adapting fast enough.
The full conversation covered even more ground, including the token cost bubble and what happens to margins if AI infrastructure subsidies dry up, Sid’s framework for running AI audits (which he did for Venture5), and the concept of “bring your own agent” as a new product strategy. Check it out above.
