Product Strategy in the Age of AI

The AI-Driven Transformation of Software

The landscape of product strategy is undergoing a significant transformation driven by advancements in Artificial Intelligence. Traditionally, software systems primarily functioned as data repositories or tools requiring human input for task execution. However, AI is enabling software to perform work autonomously and proactively [5]. This fundamental shift means that systems like accounting software can now accomplish tasks independently, rather than merely serving as filing cabinets for human retrieval [5]. This evolution is prompting businesses to re-evaluate their processes, distinguishing between "input-constrained" and "output-constrained" work, although this distinction may oversimplify complex business realities [5].

AI as Foundational Infrastructure

By 2026, AI is projected to transition from an experimental technology to a foundational business infrastructure, essential for maintaining competitiveness [8]. This urgency is fueled by AI's accelerating ability to solve core business problems and the need to keep pace with competitors already leveraging AI [8]. A significant increase in AI adoption is already evident, with 88% of organizations expected to use AI in at least one business function by 2025 [8]. True competitive advantage in this new era will stem not just from AI adoption, but from the deep integration of proprietary data with customized AI models to create unique value [8].

New Paradigms in Product Development and User Experience

AI is fostering significant software extensibility, allowing for highly tailored applications that leverage existing platforms, thereby making core software 'stickier' and more valuable [5]. This includes "vibe coding" approaches, where Large Language Models (LLMs) can generate functional applications from minimal prompts, enabling rapid prototyping even for non-developers [11]. For instance, LLMs like Claude Opus 4.6 and GPT-5.4 have demonstrated the ability to independently generate complete SwiftUI macOS applications, acting as design agents to suggest features and handle UI/UX elements [11]. However, LLM-generated applications may lack accuracy and credibility without expert validation, and the long-term maintainability of highly tailored, AI-generated extensions remains a concern [11].

User experience design in AI-powered software requires a strong focus on building trust, managing context, and facilitating human-agent collaboration, rather than solely optimizing model quality [5]. Novel approaches to user input, such as using an HTML range input for phone number formatting, demonstrate how client-side scripting can provide interactive formatting [1]. While this method showcases dynamic formatting, its practical utility for phone numbers is limited due to accessibility, validation, and internationalization challenges [1].

The Rise of Agentic Workflows and AI Clouds

The concept of a "token factory" is emerging, where companies produce intelligence at scale rather than static web experiences [2]. This involves the development of AI Clouds and agentic workloads, which are autonomous, self-healing infrastructures purpose-built for AI agents [2]. Vercel's V0, for example, has seen rapid adoption, surpassing 100 million app generations and significantly expanding Vercel's user base, with 7 app generations occurring every second [2]. These figures, while impressive, are self-reported and the definition of an "app generation" may inflate actual product usage [2].

Agentic AI, capable of reasoning and executing multi-step workflows autonomously, is creating a significant productivity gap between AI-native firms and their slower peers [8]. This shift is also leading to new protocols, such as Vercel's MCP (Model Context Protocol), envisioned as the "HTTP of the agentic web" to enable agent-to-agent interoperability at scale [2]. Proponents suggest MCP could accelerate business development by enabling agent-to-agent integrations at 100x the speed of human-to-human interactions, though critics argue this overstates AI's role in complex human-centric processes like trust-building and negotiation [2].

Cloud billing models are also adapting, with platforms like Vercel's Fluid Compute shifting from charging for allocated idle time to actual CPU cycles consumed, which is particularly suited for long-running LLM workloads [2]. While this consumption-based billing is not entirely novel, its application to the specific demands of LLM inference jobs is a key development [2].

Knowledge Management and AI Search

Traditional Retrieval-Augmented Generation (RAG) systems, which re-derive knowledge from scratch with each query, are being challenged by new approaches. One alternative proposes LLMs incrementally building and maintaining a persistent, structured wiki [4]. In this model, the LLM acts as an active knowledge base curator, continuously integrating new information, updating knowledge graphs, and flagging contradictions, thereby reducing maintenance overhead [4]. This system involves immutable raw sources, an LLM-managed wiki, and a schema configuring the LLM's behavior, with humans curating sources and directing analysis while the LLM handles maintenance [4]. However, this approach is still a form of RAG and introduces new complexities and potential points of failure, requiring significant human oversight to prevent hallucinations and ensure accuracy [4].

In the realm of AI search, companies like Perplexity are exploiting gaps left by incumbents. Perplexity's CEO, Aravind Srinivas, argues that Google's reliance on ad revenue prevents it from fully embracing an AI-native search product, despite its superior infrastructure [9]. Perplexity aims to be an "accuracy layer" for AI decision-making and is expanding distribution through hardware partnerships and an AI-native browser called "Comet" [9]. Comet is envisioned as a "cognitive operating system" that blends navigation, information retrieval, and agentic action, operating asynchronously to perform tasks without explicit user prompts [9, 10]. It builds a personalized model of user preferences using local browser history, avoiding server-side data centralization for privacy [10]. Perplexity also proposes a publisher monetization model based on query-level revenue sharing rather than traffic referrals [9]. Google, however, is reportedly actively intervening to block Perplexity's OEM distribution deals [9].

Strategic Shifts for AI Companies

OpenAI, after achieving consumer success with ChatGPT, is strategically pivoting towards the enterprise sector, with enterprise growth now outpacing consumer growth [6]. ChatGPT maintains a dominant position in the chatbot market, a lead OpenAI expects to grow [6]. OpenAI views 'code reds' as proactive measures to address competitive threats and identify product weaknesses, as seen with Deepseek earlier this year [6]. The company anticipates significant model improvements in the first quarter of next year, with specialized advancements for both consumers and enterprises [6]. AI models are nearing or exceeding expert-level performance in a significant portion of knowledge work tasks, with GPT 5.2 and GPT 5.2 Pro demonstrating expert-level capabilities in over 60% of such tasks [6].

Google Labs is also democratizing AI agent development through experiments that allow users to build consumer-facing AI agents without coding knowledge, receiving positive public reaction [7]. This indicates a broader trend towards making AI development more accessible.

Business Models and Growth

The "SaaS apocalypse" fear, suggesting widespread failure of SaaS companies due to AI, is considered overblown. However, not all SaaS companies will thrive, particularly those with traditional per-seat pricing for outcomes that AI can now automate [5]. The shift towards outcome-based pricing is a potential solution, though it faces practical challenges [5]. Vercel, for instance, has seen substantial growth, with its ARR increasing from approximately $100M to $180M within a year, largely attributed to V0 adoption and AI infrastructure demand [2]. V0 itself is reported to have positive and improving gross margins, driven by retention among enterprise and dev-adjacent users [2]. These growth figures and margin claims, while positive, are based on self-reported data and require independent verification [2]. OpenAI's Codex, a code-generation AI, has also seen rapid user growth, with weekly active users more than tripling since the beginning of the year [3]. While this indicates accelerated adoption of AI-powered developer tools, the lack of absolute numbers makes it difficult to assess the full significance of this growth [3].