Mike’s Agent Insights #5

Welcome to the 5th edition of Mike’s Agent Insights coming to you from Zürich. I have appreciated all of the conversations which the last couple of posts have started

In a world where AI increasingly shapes our lives and AI agents will increasingly transform even more aspects of our day to day  lives, it's important to stay informed and understand the implications of this technology. That's why I'm launching Mike's Agent Insights, a newsletter dedicated to exploring the latest research, investment, advancements and applications of AI agents. 

This week I want to explore (it was a light week in agent announcements); 

1) Agents in the news

2) How an average Product Manager looks at AI and AI Agents today

Meta AI /imagine a crochet llama in Zurich

Agents in the news

AI agents are poised to revolutionize the way we interact with our devices, potentially replacing traditional app-based interfaces and transforming the digital landscape - (CNet | Katie Collins)

The emergence of AI agents could fundamentally change how we use our devices, making interactions more seamless, intuitive, and personalized, but also raises concerns about data privacy, security, and the future of apps. Qualcomm's Snapdragon Summit highlighted the potential of AI agents to become the central interface for our digital lives, with CEO Cristiano Amon envisioning a future where AI agents replace traditional app-based interfaces. Industry experts, including Geoff Blaber, CEO of CCS Insight, agree that AI agents could transform the paradigm of interacting with technology.

The rise of AI agents could have far-reaching implications for the tech industry, including:

• A shift away from traditional app-based interfaces

• New opportunities for developers to create AI-powered experiences

• Potential disruption to the dominance of major tech companies

• However, experts caution that the transition to AI agents will be complex and may involve significant challenges, including regulatory hurdles and concerns about data ownership.

Key takeaways from the Snapdragon Summit include:

• AI agents will be powered by advanced neural processing units (NPUs) and will tap into the capabilities of next-generation chipsets

• Qualcomm is providing developers with tools and resources to integrate AI capabilities into their apps

• On-device AI is seen as a key factor in ensuring data privacy and security

• Industry experts predict a slow fade for traditional apps, with AI agents becoming increasingly prominent in the coming years

My thoughts: Your favorite apps on your phone are not going to be replaced tomorrow by AI agents, but in the near future a number of features will be.  Soon after that, how you interact with those apps will matter less than all of the other ways you will interact with AI and replace a lot of the things you do today, from work to entertainment to doom scrolling content. I see a bunch of promise in some of the early open source and research projects for generating code or even UI today.  There are a few things which will need to happen before we see these changes. If we are living our lives waiting for a series of inferences, those inferences will need to be much more performant and much more efficient.  We will also likely need to see more distributed inferences for the most basic things on device or the edge, and more complicated things in the cloud. This will also mean that how we approach building products and features will need to change as well.  

Deloitte's latest report “How AI agents are reshaping the future of work” - (Download the report)

Deloitte believes that AI agents are revolutionizing the future of work by expanding automation capabilities and enterprise impact through GenAI. AI agents have the potential to significantly enhance enterprise productivity and program delivery by automating end-to-end processes, particularly those requiring sophisticated reasoning, planning, and execution, thereby driving business growth and efficiency.

Early GenAI use cases have been limited to standalone applications, such as generating personalized ads, reviewing contracts, and predicting molecular behavior. However, AI agents are designed to address these limitations by leveraging domain- and task-specific digital tools to complete more complicated tasks effectively.  The emergence of AI agents is opening new possibilities for business process automation, enabling use cases that were once thought too complicated for GenAI to be executed at scale, securely and efficiently. This has broader implications for the future of work, as AI agents can reason and act on behalf of users, transforming the way businesses operate. Key distinctions between AI agents and traditional LLMs/GenAI applications include:

• Ability to understand context and plan workflows

• Capacity to connect to external tools and data

• Execution of actions to achieve defined goals

• Long-term memory to remember customer interactions

• Ability to automate end-to-end processes

• Sophisticated reasoning, planning, and execution capabilities

They presented their thoughts on how they view AI agents as a new paradigm.

My thought: If you are reading this you are already likely thinking about AI agents in some way today. The truth is that a lot of the world is not yet.  While this is a self-serving report for Deloitte’s AI practice, it is also a very good introduction to the role AI agents can play in transforming businesses.  I like the 6 things they focus on from use case scope through accuracy and how product approaches shift between them. Not much controversial in this report, but still a good read. 

How an average Product Manager looks at AI and AI Agents today  

Denver Product Summit attendees conducted a SWOT analysis of AI, highlighting its potential to increase productivity, drive innovation, and create new opportunities - (Denver Product Summit post)

A few weeks ago, I got to attend the Denver Product Summit, and of course AI was a constant theme through the day. This was the first year of this summit and it sold out very quickly.  There was a great set of speakers and workshops.  I definitely appreciate the Denver tech scene. One of the highlights was Eric Marcoullier giving his classic “Your product doesn’t f**king matter!” speeches. 

One of the exercises in the afternoon broke the crowd of a few hundred product managers and designers into groups to go through and do a classic SWOT analysis of AI, which is a strategic planning exercise used to identify and evaluate Strengths, Weaknesses, Opportunities, and Threats.  Who doesn’t love a good post it note exercise. 

The high level themes of what came out of the exercise;

Strengths (S)

• AI can increase productivity (e.g., automation, faster learning, efficient natural language search).

• Saves time and allows for a quicker cycle in tasks (e.g., summarizing data, producing results fast).

• AI represents the future of products and offers the potential for breakthroughs in various sectors.

• New jobs and industries can emerge as AI technology advances.

• Encourages focus on core tasks and allows for more innovation.

Weaknesses (W)

• Lack of trust in AI and concerns about hallucinations or inaccuracies in results.

• AI often lacks empathy and doesn't always understand the nuance in information.

• Bias and reliability issues with AI outputs are ongoing challenges.

• Not accessible to everyone, particularly with gaps in education and resources.

• AI sometimes loses the essence of the original goal when processing or summarizing data.

Opportunities (O)

• AI has the potential to become an assistant that helps with rapid prototyping  and other tasks, encouraging more reading and learning.

• It can improve autonomy, creating new use cases and better engagement with users.

• The ability to solve problems faster, develop new solutions in industries, and accelerate innovation.

• Encourages creativity and opens opportunities for more flexible, intuitive solutions to problems.

• Helps reduce the burden of repetitive tasks and enhance work/life balance.

Threats (T)

• There are risks of foreign interference, particularly regarding misinformation.

• AI systems may exacerbate inequalities, as they are not equally accessible.

• Threats include loss of control or misuse of AI, possibly impacting autonomy and individual decision-making.

• Concerns over privacy and how AI handles personal information.

• The legal and regulatory environment around AI is uncertain and could lead to litigation or slow down progress.

My thoughts: The reason I wanted to drop this in an agents newsletter is that agents are not yet top of mind for people building and developing products.  Broadly people are just beginning to understand LLMs alone and are not yet in a place to internalize the additional dimensions which come from reasoning and tools combined with an LLM.  I think this is also important to call out as we must rethink and revolutionize our approaches to design patterns, application building, and service interconnectivity to harness their full potential.  Reusing a prior design pattern will not help take the cognitive lift off of a consumer's mind for completing a transaction.  One of the goals of putting agents in front of a consumer is to remove complexity where possible, especially on actions which are low risk.   The SWOT exercise was still great and helped me understand how other product managers are thinking about this space.