Tech • IA • Crypto
Building AI-native across industries with NTT, Mizuho and Mercari
7/10TL;DR
AI-native startups are rapidly reshaping how products are built, priced, and used, with agents, automation, and non-technical builders driving a new wave of software development.
KEY POINTS
AI unlocks new product capabilities
Advances in models from companies like Anthropic have enabled startups to move beyond simple automation into decision-making and action-taking systems. Tools can now analyze unstructured data, generate insights, and execute tasks such as personalized outreach or customer targeting in real time. This shift allows businesses to perform workflows that were previously impractical or impossible.
Creative real-world applications emerge
Companies are applying AI in unconventional ways, such as analyzing Google Street View imagery to identify potential customers based on visual cues like dumpster colors linked to competitors. These insights can trigger automated marketing campaigns, illustrating how AI blends data extraction with execution to drive revenue.
Non-technical users gain power
Platforms like Emergent report millions of users building production-ready applications without coding experience. With over 10 million apps created and roughly $100 million in annualized revenue, the trend highlights how AI tools enable small business owners to develop sophisticated software, sometimes rivaling traditional engineering teams.
Rise of autonomous and “ambient” agents
Startups are increasingly deploying agents that operate continuously in the background, monitoring data and proactively notifying users. These “ambient agents” reduce the need for active engagement, delivering insights or actions automatically, particularly in consumer finance and productivity tools.
Cost challenges and lessons learned
Early implementations of always-on agents led to significant expenses, including reported monthly AI costs exceeding $500,000. Companies have since shifted toward more controlled systems, allowing user behavior to guide automation and reduce unnecessary token usage.
Evolving pricing models
Traditional cost-plus pricing has re-emerged in AI due to token-based expenses, but companies are experimenting with hybrid approaches. Some separate pricing into two layers: one covering infrastructure costs and another tied to customer value, reflecting a broader industry struggle to move toward outcome-based pricing.
Architecture shifts toward flexibility
Builders are reducing rigid infrastructure and allowing models greater autonomy. By designing systems where agents operate like developers with access to tools, logs, and environments, companies can adapt more easily as models improve, avoiding frequent reengineering.
Improved model performance changes economics
Newer models are not only more capable but can also be more efficient, reducing overall costs despite higher intelligence. This trend challenges assumptions that better performance necessarily leads to higher expenses.
Persistent need for human oversight
Despite improvements, enterprise users עדיין demand strict quality control, particularly in brand-sensitive areas like marketing. Human review remains common, indicating that trust in fully autonomous systems is still developing.
Memory and personalization as next frontier
Features like persistent memory and file-system-based context are enabling deeper personalization. These systems store user preferences and history, allowing AI to deliver more tailored and continuous experiences over time.
Constant rebuilding becomes the norm
Startups report rearchitecting systems multiple times within months due to rapid model evolution. Adaptability and internal evaluation systems are critical, with companies continuously testing and replacing components to stay competitive.
Opportunities remain wide open
Despite crowded markets, many existing products fail to meet user needs. Founders are encouraged to focus on real customer pain points, as gaps persist even in heavily funded sectors.
CONCLUSION
AI is transforming both the creation and operation of software, favoring adaptable systems, autonomous agents, and user-centric design, while leaving ample সুযোগ for new entrants to challenge existing solutions.