Product Strategy in the AI Era

If we define AI as a kind of technology, then we need to rethink what “technology” really means. In this context, technology refers to the process by which an organization turns labor, capital, users or clients, and the implementation of specific tools into higher-value products and services.

The stories of the past don't necessarily predict the future. Product developers, technologists, and business managers all need to shift their mindset. It's not enough to just expand on existing services. Technology evolves rapidly. In the early stages of disruptive technologies, the return on investment often doesn't meet the annual profit growth targets of large companies. Similarly, within an organization, the execution from top to bottom often can't keep up with the market growth—or new markets—created by these disruptive technologies, causing the company to lose future market share.

Over the past decades, we've repeatedly seen companies fail when faced with disruptive technological change and shifts in market structure. Only by deeply understanding “how the world works” and aligning product creation and innovation with these forces can a company adapt to the times and secure future profits.

“Only by deeply understanding ‘how the world works’ and aligning product creation and innovation with these forces can a company adapt to the times and secure future profits.”

With this in mind, we asked ourselves several questions:

• What are the core metrics of large language models (or AGI)?
• What new things can we do with them?
• Which processes can be replaced?
• Where might new markets or budgets emerge?

At AtomGradient, we've tried to answer these questions through our own research and development.

Core Metrics of Large Language Models

• Understanding the intentions of collaborators
• Providing reasonable, common-sense responses that reflect some physical reality
• Drawing on subconscious or tacit knowledge
• Flexibly using external tools or creating new tools on their own

Although current models are far from fully achieving these capabilities, we can already see signs of progress.

What new things can we do with them?

To answer this, we first ask: what can humans not do—or what are humans limited in doing? Large language models extend human capabilities in new ways: creating art or music, writing creatively, automating knowledge work, generating insights from data, programming, or assisting in collective decision-making. They make people more versatile and creative, reduce repetitive labor, and improve efficiency. Over time, this can lead to new careers and more efficient use of resources. Essentially, one person's capabilities can be amplified tenfold.

Which processes can be replaced?

As human efficiency and work quality improve, roles that require repetitive effort or specialized knowledge are increasingly replaceable. Examples include:

• Highly repetitive, rule-based work, such as customer support, quality inspection, data entry, or general consulting
• Knowledge-intensive or digital work, like bank risk assessment, weather forecasting, recommendation systems, data analysis, and programming
• High-risk or harsh-environment work (as models integrate more with the physical world), such as mining, deep-sea oil exploration, disaster rescue, or exploration

We've already seen examples in autonomous driving, automated logistics, and creative fields like music, art, and writing.

Where might new markets (or budgets) emerge?

The shape of future markets is unpredictable and dynamic, but based on what we know:

• Because models require knowledge to function effectively, industry-specific data represents a vast, untapped opportunity. Platforms supporting professional or user-generated content will likely continue to have advantages.
• Current models are still difficult to use, whether for enterprises or individual users. Making AI services easier, cheaper, and more accessible will likely attract significant investment for years to come.
• With increasing regulation and enforcement of data privacy and sovereignty, AI applications in safety, finance, data protection, contracts, payments, identity verification, and fraud detection will remain critical throughout every stage of a service's lifecycle.