This is Hamamoto from TIMEWELL.
AI is transforming organizations across every sector — enabling new approaches to complex problems, accelerating growth, and opening business opportunities that didn't exist before. This is particularly pronounced for startups, which can move quickly to capture new market opportunities as AI capability expands.
Google Cloud compiled perspectives from 23 industry leaders and investors for a report titled "Future of AI: Perspectives for Startups 2025." This article covers the key themes.
AI Agents: The Central Evolution
AI agents dominated the discussion. The progression: from AI that responds to queries → to AI that executes tasks → to AI that operates continuously and proactively.
Current State: Multi-Modal, Web-Connected Agents
David Thacker, VP of Product at Google DeepMind, describes where Gemini-based agents are now:
"Gemini can natively use tools like Google Search to access real-time information, and DeepMind's Project Mariner demonstrates that agents built on Gemini models can complete tasks using a web browser. With the Gemini Multimodal Live API accepting voice and video streaming input, it's now possible to build conversational experiences. Combining these capabilities enables a new class of agent experiences — I'm excited to see what startups build with Gemini in 2025."
The Emotional Dimension
Matthieu Rouif, Co-founder and CEO of Photoroom, emphasizes AI's growing capability to recognize and respond to human emotion:
AI systems are becoming increasingly capable of recognizing emotional states and adapting content and experience to individual emotional responses — creating connections that feel more meaningful than prior AI interactions.
Jia Li, Co-founder and Chief AI Officer at LiveX AI, sees a similar evolution: AI agents moving from simple task completion to genuine understanding of customer intent and emotional state, providing guidance that feels human rather than mechanical.
Ambient Agents
Harrison Chase, CEO and Co-founder of LangChain, articulates the most forward-looking formulation:
"To truly harness agentic systems, I want them to be 'ambient agents' — always running in the background, monitoring streams of events, alerting me only when something interesting happens or when help is needed."
This is the shift from AI you query to AI that monitors. The agent doesn't wait to be asked — it surfaces what matters when it matters.
Looking for AI training and consulting?
Learn about WARP training programs and consulting services in our materials.
AI Infrastructure: The Enabling Layer
Almost every advance in AI capability depends on appropriate infrastructure. This was the second major theme.
The Computational Challenge
Amin Vahdat, VP & GM of Systems and Cloud AI at Google Cloud:
"With tight synchronization and large-scale computational requirements, infrastructure is being pushed to unprecedented levels of computational density and capability."
Modular Architecture as the Target
The emerging infrastructure design pattern: modular architectures combining smaller specialized models across different data modalities, supported by sophisticated orchestration and observability layers.
Mayada Gonimah, CTO and Co-founder of Thread AI:
"Integration and infrastructure for observable workflow management will become an increasingly critical part of the AI stack. Enterprises need interfaces that let them insert AI into existing workflows and test it in parallel with existing processes. Building these practical AI-native workflows requires being more intentional about where AI is embedded, and having a toolset for observability."
Infrastructure Efficiency as Competitive Advantage
Glean CEO Arvind Jain emphasizes designing infrastructure to be model-agnostic:
Organizations that achieve 2x efficiency improvements will hold significant market advantages. Anticipating what's coming and enabling plug-and-play integration — being able to leverage updates without major redesigns or disruption — is critical.
Cloud providers will continue to play a central role, but successful organizations will build flexible systems that can integrate new models and technologies seamlessly.
What Investors Are Looking For in 2025
AI investors in the report have moved past the initial excitement to focus on companies demonstrating real-world problem-solving.
Real Problems, Specific Verticals
Salim Teja, Partner at Radical Ventures, is focusing on startups using AI to address concrete problems:
Healthcare improvement, disease treatment, climate change solutions, and addressing the affordable housing crisis through construction robotics — these are the application areas I'm watching.
Strategic Data Use
Having data access isn't sufficient. Investors want to see high-quality, strategically deployed data that improves AI performance in specific applications — not just data volume.
Deep Workflow Integration
GV General Partner Crystal Huang:
"If a product is easy to implement, it's just as easy to uninstall. Creating lasting value requires stickiness — being both essential and deeply integrated into users' workflows."
The Winning Formula
The pattern across investor perspectives: clear competitive advantage + visible path to profitability + deep integration into users' workflows.
"Product-Algorithm Fit"
AI21 Labs Co-founder Yoav Shoham emphasizes understanding and leveraging current AI capabilities rather than betting on future ones:
Rather than waiting for the perfect model, organizations must continuously invest in evolution and adaptation — integrating AI solutions that solve actual business problems and deliver measurable outcomes.
The Commoditization Warning
Ohalo Genetics CEO David Friedberg:
"If you're just an LLM wrapper, it will be very hard to build a sustainable business — you're likely to be commoditized. You need a durable value creation engine with initial product advantage and continuous improvement. This typically comes from proprietary data generation that continuously improves model performance, or from network effects that lock in data or customer access."
The Broader Context
Google Cloud positions itself as a partner for startups navigating rapid AI evolution — providing infrastructure, partnerships, and guidance for responsible AI deployment that meets the needs of employees, customers, and the communities organizations serve.
The report's intention: to illuminate the direction of one of the fastest-moving technologies the industry has seen, and provide actionable guidance for founders building the next wave of innovative AI startups.
Summary
Five themes emerge from the 23-perspective report:
- Ambient AI agents are the next frontier — AI that monitors continuously and surfaces insights proactively, not only when queried
- Modular, observable infrastructure with model-agnostic flexibility is the target architecture
- Investors want specificity: real problems in defined verticals, strategic data, deep workflow integration — not just "AI-powered"
- Deep integration creates defensibility: easy-to-implement products are easy to replace; sticky products become essential
- Proprietary data loops and network effects are the most durable moats — LLM wrappers face commoditization
For startups: the question isn't whether to incorporate AI but how to build around it in a way that creates compounding rather than replaceable value.
Reference: https://cloud.google.com/transform/future-of-ai-for-startups-23-industry-leaders-survey
