Why Saying No to Everything Is How AI Startups Survive

Dara Ladjevardian made it his New Year’s resolution: say no to everything.

This isn’t normal founder advice. Most founders obsess over optionality—keeping doors open, exploring use cases, building features that might appeal to various customer segments. But Dara realized early that the AI space is different. The technology is so new that people project infinite possibilities onto it. Everyone thinks their custom use case is the obvious next feature, and many have the status or resources to demand it.

For traditional software companies, feature creep is a slow death. For AI startups, it’s instantaneous.

“As an AI company, you can be pulled in a lot of directions. People want to use your product for all these different things,” Dara says. “And my resolution is to just say no to everything and just continue to only do what we’re doing. The product works the way it is. Either you’re happy with it or you’re not.”

The Temptation of Infinite Customization

The problem is most acute when status is involved. A famous person wants a custom workflow. A celebrity wants their Delphi to generate videos and call users through a phone number. Build-this-one-custom-feature requests come from people who could be tremendous for your brand. The stakes feel real.

“People want to use your product for use cases that it’s not intended for. It can be tempting—especially if it’s someone famous—to be like, okay, I’m gonna do this,” Dara admits. But this is the trap.

When you start building custom features for high-status customers, a few things happen simultaneously. Your roadmap becomes fragmented. Your engineering team splits focus. Most critically, your product becomes less good at its core thing because you’re maintaining multiple feature branches and integration points. The user who’s not famous gets the degraded experience. The custom feature for the celebrity barely works. Everyone loses.

Why AI Products Are Uniquely Vulnerable to This

Traditional software has clearer category boundaries. If you say no to a CRM requesting accounting features, they understand. Accounting is different from CRM. But AI is novel. “People have not conceptualized this technology before. So it’s causing people to have more ideas than, you know, pre-AI,” Dara explains. “If just a regular software company starts, you know what a software company is. You’re not going to project your ideas onto them.”

With AI, every person who touches the product has a different vision of what’s possible. That ambient creativity is a feature of the technology, but it’s poison for product focus if you listen to all of it.

There’s also the sunk-cost dynamic. You’ve built a system that can do X. Building X+Y feels like an incremental extension. The cost is hidden—it’s in the debugging, the edge cases, the security implications, the maintenance burden. But the temptation to build it anyway is enormous because you already have 80% of the infrastructure.

Delphi’s position is simple: the product is a knowledge graph representation of a human mind, made accessible through conversation. The person owns their data. The platform is human-verified. It’s not a video generator. It’s not a phone system. It’s not an app builder. If you need those things, there’s probably a company better suited to building them than us.

The Muscle of Saying No

Dara acknowledges this wasn’t natural at first. “It was initially hard to say no, but I think I’ve grown the muscle now,” he says.

Growing that muscle required repeating the same response until it became reflexive. When someone pitches a feature, the first instinct is to consider it. The second instinct is to want to help. The third is to realize: helping by building the wrong thing is worse than not helping at all.

The response Dara settled on doesn’t change: “Hey, this is what we’re focused on. You can connect all your data. We create a high-fidelity version of your mind that understands what you know and how you think. And you can allow other people to learn from you and talk to you, whether for free or you want to monetize it.”

If that doesn’t fit the customer’s need, they need a different product. And that’s okay. As Dara puts it, “We’ll get around to that eventually. Though there are some things we’re just like, we are human only, human verified, no fake characters. That’s never going to be us. You should go to another platform.”

The Hard Boundary: Human-Only, Verified

The focus isn’t just about saying no to features. It’s also about saying no to use cases that violate the core mission. Some companies want to use Delphi to create anime characters or fictional personas. Delphi won’t. Some want to create a digital version of a historical figure (not the person themselves, but a recreation).

Again: no. Human-verified only. This isn’t a negotiable constraint—it’s the entire product thesis. If you allow fake versions of people, you’ve destroyed the trust that makes the real versions valuable. One fake Elon Musk profile and suddenly all Elon Musk profiles are suspect.

This kind of boundary-setting looks like rejecting revenue. In the early days, it might actually be. But it’s the price of coherence. You can’t be a platform for trusted human knowledge and also allow deepfakes of famous people. You just can’t. The two missions are incompatible.

Why Focus Becomes Your Advantage

Here’s the paradox: by saying no, you become more defensible.

If you try to be the platform for all AI use cases—video generation, code synthesis, character creation, knowledge representation—you’re competing on features with OpenAI, Anthropic, and companies with 100x your resources. You’ll lose.

But if you commit to one thing—representing how human minds think—and you get that incredibly, deeply right, you’ve built something that’s hard to replicate. A general-purpose AI company can bolt on a knowledge graph feature. But they’ll never have the same obsession with human accuracy, temporal reasoning, and explainability that Delphi has because it’s not the main character in their story.

Dara’s framing: “Let’s obsess over like, what does it mean to have a human mind? Let’s make that useful and then let’s build a big enough network such that it’s like a distribution mode where, you know, LinkedIn has not been able to be disrupted because they built such a strong network.”

Delphi isn’t competing with OpenAI or Anthropic. It’s competing with how people currently access expertise—through books, courses, in-person mentorship, Twitter follows. The alternative platform is not an AI company. It’s the status quo of asynchronous, passive learning.

By focusing relentlessly on that one use case and saying no to everything else, Delphi creates a situation where the company becomes less like a feature in someone else’s platform and more like a new category. And new categories are worth defending.

FAQ

Doesn’t saying no leave money on the table?

Yes, in the short term. A custom video-generation feature for a celebrity could generate revenue. But it distracts the team, degrades the core product, and trains customers to ask for more customization. You’re trading long-term cohesion for short-term revenue, which is a losing deal in startups.

How do you know what the “right” focus is?

You test your thesis against use cases. Delphi’s thesis is: “Always-on availability for handling repetitive questions from people who want to access you.” Every feature decision filters through this. If it doesn’t enable that use case, it’s a distraction.

What if the market wants something different than what you’re building?

Then you’re building for the wrong market, and that’s important to discover early. But markets don’t need all features immediately. They need coherence. A product that does one thing well beats a product that does ten things badly.

Isn’t market feedback important? What if customers are asking for something that would increase adoption?

Customer feedback is valuable, but it’s not always right. Customers ask for what they can imagine within their current context. They don’t ask for things that require paradigm shifts. If you listen only to customer requests, you’ll never innovate. You’ll optimize the status quo.

How do you handle the pressure from investors or board members to expand?

By showing them the data. If saying no to features resulted in a more coherent product, more word-of-mouth adoption, and faster team velocity, investors eventually get it. But this requires consistent results, not just a philosophy.

What’s the difference between focus and missing a market opportunity?

This is the real risk. Saying no to something that could have been huge. The answer is: you can’t avoid this risk. Every company that says no misses something. The question is whether you miss the right things—distractions that would have weakened you—versus the wrong things—the core insight that drives your category.

If OpenAI decides to build this knowledge graph thing, don’t you lose?

Maybe. But OpenAI’s thesis is “general-purpose AI.” If they add a knowledge graph feature, it will be general-purpose. Delphi’s thesis is “human-verified knowledge representation.” These are different bets. OpenAI’s version might be technically superior. Delphi’s version might be more trusted. The focus makes that difference possible.

How do you communicate the “no” to people who make the request?

Dara’s approach is direct: state what you’re focused on, acknowledge that they have a legitimate need, and suggest they look elsewhere. “We’ll get around to that eventually” is honest—maybe in five years when it makes sense. “You should go to another platform” is kinder than pretending you’ll do it someday.

Can you change your focus later if the market shifts?

Yes, but the cost is high. You have to rebuild trust from scratch, retrain the team, realign the product. This is why focus is so valuable in the early years—it buys you time to get the core thing right before you expand.

What if you’re wrong about what customers want?

You’ll know. If your thesis is wrong, you’ll see it in adoption rates, churn, or customer acquisition cost. But you’ll see it quickly because you’re not wasting time on side bets. Wrong focus + clarity = quick feedback. Wrong focus + sprawl = slow death.