Why OpenAI's Mark Manara thinks AI start-ups need new moats  

As foundation artificial intelligence (AI) models become more capable and increasingly affordable, the competitive edge for AI start-ups is shifting away from sophisticated prompting and engineering workarounds towards speed, execution and distribution, according to Mark Manara, Head of Startups at OpenAI.

Speaking on the sidelines of FutureLab, Manara talked about how AI product development has evolved over the past two years. Mark said start-ups initially spent significant effort building scaffolding around models to compensate for their limitations.

Models have since become far more reliable at reasoning, following instructions and using tools autonomously, fundamentally changing how AI products are built. Rather than relying on complex orchestration, founders are now increasingly trusting models to make decisions and execute tasks, with the challenge shifting to providing the right tools and evaluation systems.

The rapid improvement in model capabilities, however, also means businesses built around temporary model limitations could quickly lose their edge.

Manara echoed OpenAI CEO Sam Altman's oft-repeated advice that founders should be excited — not worried — when new models are released. "If you're upset that the model got smarter, that's probably a less defensible and harder business to build," he said, arguing that durable AI companies are those whose products improve as the underlying models become more capable.

At the same time, the economics of AI is changing. Manara said recent improvements in smaller, lower-cost models are unlocking applications that were previously too expensive to deploy at scale. Lower inference costs mean start-ups can support larger free tiers, run more agentic workflows and experiment with use cases that would have been commercially difficult just months ago.

The combination of stronger performance and lower prices, he said, is allowing companies to rethink product design and customer acquisition without significantly increasing compute costs.

As intelligence becomes more accessible, the real differentiation, according to Manara, lies elsewhere. Distribution, deep customer understanding, product quality, and execution will increasingly determine winners. He argued that while AI has dramatically reduced the time required to build software, it has not made building enduring companies any easier.

Production-grade AI systems still require robust evaluation frameworks, guardrails, latency optimisation, memory systems and rigorous testing — capabilities that go far beyond building an impressive demo.

The AI boom is also reshaping how start-ups are organised. Manara pointed to companies with fewer than 10 employees generating tens of millions of dollars in revenue by using AI across engineering, hiring, customer support, and internal operations. He also sees traditional roles converging, with engineers taking on product responsibilities and product managers increasingly contributing code.

At the same time, specialised AI engineering skills around model evaluation, orchestration and deployment are becoming increasingly valuable, alongside emerging roles such as Forward Deployment Engineers, who work closely with enterprises to implement AI systems.

For founders, the implication is that as AI models continue to improve and become cheaper, competitive advantage will depend less on access to frontier intelligence and more on how quickly companies can translate that intelligence into products customers are willing to pay for.