Happy New Year! 🎆 

As we enter 2025, the AI agent landscape is evolving beyond mere task automation towards sophisticated ecosystems of specialized, interconnected agents. This shift presents both challenges and opportunities for designers and technologists.

The key developments this year

The Good 👍

1. AI Agents Develop Spatial Capabilities

2. Certified AI Agents Lead the Trust Revolution

3. AI Agents Learn to Collaborate for Results

4. Vertical Specialization Drives Results-Based Agents

The Bad 👎

5. Avoiding the Swamp of the Next Attention Economy Wave

The Ugly 😬 

6. Power-Hungry Data Centers Turbo-Charged with Unconventional Energy Sources

7. Standardized Safety Practices: The HTTPS Moment for Agent Deployment

1. AI Agents Develop Spatial Capabilities

This work marks a turning point in AI's ability to operate in 3D space. Large world models can now generate photorealistic virtual environments from text or imagery. Text-to-3D object models are also proliferating. This enables AI agents to rapidly advance their grasp of spatial relationships and geometry.

Fei Fei Li, founder of World Labs and Stanford HAI, sees spatial computing as a necessary leap for AI agents to operate alongside humans. We operate spatially by giving cues to objects and for agents to operate seamlessly with us they too must have spatial reasoning. 

Companies to watch:

• Google DeepMind: Genie 2

• Zoo 3D

• World Labs

3. Certified AI Agents Lead the Trust Revolution

Just as organic certification transformed food, agent certification is becoming a crucial trust signal. Consumers and technologists will seek transparency in the data used to train these systems. This movement will help push accuracy and avoid hallucinations from unequal data.

Users will seek transparency in the ‘ingredients’ of the data within AI systems. This movement to data transparency will also help push accuracy and avoid hallucinations when we don’t treat all data as equal, but make our agent builders responsible for the cleanliness of their agent training sources.

4. AI Agents Learn to Collaborate To Get Results

AI agent applications are evolving into interconnected ecosystems. Agent orchestration layers are emerging to manage complex workflows across multiple specialized AI agents.

The future is one where you can seamlessly interconnect AI agents into multidisciplinary project teams to tackle complex challenges, similar to hiring human contractors. The role of humans becomes more managerial, with AI teams providing 10-100x capacity boosts.

Companies to watch:

• Agent.ai

• Tmpt.me

• Langchain's Langsmith

• VertexAI

• CrewAI

5. Vertical Agents Begin To Deliver Results

Specialized agents focused on specific industry outcomes will continue to emerge in 2025 as the dominent form of production software applications. These will be evaluated based on concrete deliverables rather than as traditional SaaS. New "Results as a Service" (RaaS) business models will align customers and service providers.

This alignment of user and service provider is a positive development in the evolution of AI agents. Next, we will unpack the negative potential of agent business models we must avoid this year.

6. Avoiding The Swamp Of The Next Attention Economy Wave

The same bad actors who brought us the attention economy are entering the agent space, bringing manipulative monetization patterns. We must regulate companies that would take advantage of youth and at-risk adults by building artificial intimate relationships for profit. Business models built on creating faux intimacy must be outlawed to prevent societal harm.

As Yuval Noah Harari explores in his latest book, Nexus, we must be cautious of how different entities may attempt to wield AI to manipulate relationships and societal dynamics. In 2025 we will see the first companies monetizing relationships with AI agents. Harrari’s call-to-action is we must urge our politicians to regulate this technology and create better models like RaaS as builders.

Warning signs:

• Emotional manipulation

• Addiction-driven engagement

• Personal data privacy compromises

• AI-powered phishing schemes

6. Power-Hungry Data Centers Turbo-Charged With Unconventional Energy Sources

Data Centers Turbo-Charged with Nuclear Energy The rapid growth of AI is driving massive power-hungry data centers worldwide. While innovations in nuclear energy offer promise, we must remain vigilant about potential energy quality degradation. Prioritizing efficiency and sustainability is a core responsibility.

Areas with high data center centralization like Virginia are experiencing bad harmonics in their energy grid, which is resulting in power fluctuations that can cause brown outs and damanage to electrical appliances receiving power spikes. To combat these power supply issues Meta, Microsoft, Amazon to name a few are seeking nuclear power approvals for their data centers as they expand their incrasingly power hungry AI agent-based systems. For example, Meta requested approval for 4 GW of new capacity. This is enough energy to power 3.2 million homes for a year.

Opportunity areas:

• Designing opportunitie for agents to time shift compute intensive tasks away from peak energy use times

• A focus on energy efficient inference compute stack like Groq

7. Standardized Agent Communication Protocol

Just as HTTPS became the web standard, communication protocols for agents are emerging. Anthropic has recommended a standard practice for AI database access. In 2025, I expect this concept to expand to agent-to-agent communication.

It is these linkages across the architectural components of agents that will enable multi-agent systems to coordinate effectively. It is this work that will enable agents of different origins to coordinate just as the first users of the World Wide Web were able to communicate across the early internet.

Such protocols will need governing bodies and global regulations to ensure safe use and operation. This will be a hot topic in 2025 and the years to come. Multi-agent systems has much promise and many ethical and social concerns we must cautiously explore. Anthropic has taken a strong first step in laying a foundation for a standard protocol for connecting agent architectural components. This will make it easy for all the agent building tools to become cross compatible. A strong start to 2025!

How I Will Keeping A Sharp Lookout For These Trends In 2025

These trends point to a future of more trustworthy, specialized, and sustainable AI agents.

Designers and technologists must balance the power of these tools with responsible practices, emphasizing energy efficiency. We must also be vigilant against negative behaviors like the monetization of attention and artificial intimacy. AI Agents are early days, they still require a lot of time attention and infrastructure to support their conscientious development. Creating clear communication protocols along with efficient compute and energy management will go a long way in laying a strong foundation for multi-agent systems in 2025.

What trends are you seeing in your work with AI agents? Please share your thoughts and experiences.

💬 Suggestion box

A newsletter exploring the principles that will help us design AI agent experiences and startups that amplify human creativity.

Subscribe to join a community of designers and developers shaping purposeful AI agents.

Until next time, keep innovating and stay curious!

Jason