Roundtable #7: Spotify, Adobe and Linkedin on How AI Changes The Future of Product & Design | E1097

Strategic Shift to "AI-First" Architecture

• Product leaders (Adobe, Spotify, LinkedIn) agree that AI is now the core product, while the UI serves to capture signal and facilitate the AI's output.
• The paradigm is shifting from deterministic, control-heavy interfaces to probabilistic, "AI-first" experiences where the algorithm dictates the flow.
• This requires leaders to let go of total control over user outcomes, accepting that models generate non-deterministic results rather than fixed responses.

Model Ecosystem: Specialization vs. Aggregation

• Contrary to the belief that a few "mega models" will dominate, the consensus is a future of many specialized, "long-tail" models (open, closed, local, and cloud-based).
• Innovation is expected in the "dispatcher" or "router" layer that selects the most cost-effective and specialized model for a specific task.
• Spotify's Stance: Does not plan to build a general-purpose LLM like GPT-4; instead, they focus on proprietary, specialized models (e.g., for audio) and partner with providers for generic capabilities to optimize cost efficiency for massive scale.
• Adobe's Stance: Builds proprietary imaging models (e.g., Firefly) to protect IP and maintain competitive advantage in their domain, while partnering for large-scale LLMs.
• LinkedIn's Stance: Focuses on building an "integration model" that leverages their unique graph data to act as a router, connecting users to specialized agents (e.g., job seeker, seller, knowledge agent).

Data as the Primary Competitive Moat

• High-quality, proprietary user data is increasingly viewed as more critical than raw model parameter size for long-term success.
• LinkedIn: Historically spent significant effort filtering data to ensure "good diet" for algorithms; currently emphasizes that data is the "oxygen" feeding AI.
• Adobe: Enforced strict data policies to train models only on non-copyrighted, licensed content, turning IP safety into a product feature (e.g., Firefly).
• Future Trend: The "Chinchilla" paper suggests model scaling is predictable, but long-term value will accrue to companies with deep, high-fidelity user understanding rather than just massive general models.

Product Development & Design Evolution

• Design Role Transformation: Designers must now understand model capabilities (e.g., GPT-4 performance, diffusion mechanics) as deeply as user psychology to create fault-tolerant UIs.
• UI as Persona: Interface design is evolving into "persona design," where the tone, inflections, and voice of the AI (e.g., Spotify's DJ) constitute the primary user experience.
• Prompt Engineering: Prompt creation has become a core design discipline, with teams required to submit and critique prompts as rigorously as code or wireframes.
• Efficiency: AI tools like GitHub Copilot and generative UI testing are accelerating development cycles, allowing teams to explore more design variations and identify bugs faster.

Cost, Infrastructure, and Hardware

• Cost Pressure: Cost efficiency is a primary driver of innovation; generating two minutes of voice for half a billion users daily creates significant financial pressure, necessitating intelligent model routing.
• Hardware Trends: The industry is moving toward verticalization with dedicated chips (e.g., Microsoft, Apple) to reduce power consumption and improve efficiency, mirroring historical semiconductor trends.
• Moore's Law Analogy: While traditional transistor scaling slows, neural hardware progress and model compression (smaller models performing better than previous large ones) are expected to continue driving cost reductions.

Business Model Implications

• Disruption Potential: The current wave of AI is largely "sustaining innovation" for incumbents; a true business model disruption (e.g., shifting from "seats" to "task completion") has not yet fully materialized.
• Selling Value, Not Seats: Leaders anticipate a shift away from per-seat licensing toward value-based pricing as AI increases organizational efficiency and reduces the need for headcount.
• Enterprise Adoption: Adoption is predicted to follow an S-curve with a sharper inflection point than cloud/mobile transitions, driven by small-team pilots rather than top-down mandates.

Career Advice for Product Leaders

• Skill Shift: Job requirements are changing rapidly; skillsets are expected to change by 25% every 5-6 years, reaching 65% by 2030.
• T-Shaped Growth: Future leaders need broad skills across domains while maintaining deep expertise in one specific vertical.
• Soft Skills: Interpersonal dynamics, imagination, and empathy are becoming more valuable as technical tasks become automated.
• Actionable Advice: Early-career professionals should lead by example, using AI tools personally to pilot new workflows and challenge organizational inertia.

Key Quotes & Decisions

• Scott Belsky (Adobe): "The need for centralization... I re-centralized design in one organization... sometimes you make a completely opposite decision at different times in the same business."
• Tomer Shalev (LinkedIn): "I would not hire [a product leader] to my org if [they] don't have the willingness and the aptitude [to go] deep [into AI]... I need you to understand... the underlying principles that come with it."
• Gustav Söderström (Spotify): "We are back to that kind of change [seen during mobile shift]... I'm secretly hoping for that business model challenge... I'm kind of back to like more excited than maybe in the last seven, eight years."
• Tomer Shalev: "The conservation of complexity... the more complex your product, the more impact AI could have in terms of simplifying it for your user base."
• Scott Belsky: "Novelty precedes utility... embrace these tools... you can be the person on your team that introduces new practices."