AI Agents & New Age of Software
How AI Agents are changing the software landscape and proliferating across every industry
Just over a year ago, we wrote a post titled “The Dawn of Agents”. In that piece, we predicted that autonomous AI agents would revolutionize how we interact with AI, ushering in an era where we delegate entire tasks rather than just prompting systems step-by-step. This was right around the time projects like AutoGPT and BabyAGI first emerged, shocking the world with their ability to independently accomplish tasks without human intervention. Back then, the very idea of an AI agent autonomously carrying out objectives seemed like science fiction.
Fast forward to today, and autonomous agents have proliferated across nearly every industry, vertical and layer of the software stack. Numerous applications and services now incorporate agent functionality, albeit not always fully autonomous, enabling natural language interaction and unleashing the power of generative AI across their products.
We've witnessed innovative new companies like Ema.co, Continua.ai, Norm.ai, and Sema4.ai burst onto the scene. Incumbents like Zapier and UiPath have also aggressively implemented more agent-like capabilities into their workflow offerings. In fact, just this week, tech titan Google announced a suite of "AI Agents" designed to serve as personal assistants to consumers across a variety of tasks. What once seemed like a bold vision has become an industry-wide reality and it's unfolding at a pace that continues to astonish.
In this post, we provide some refreshed thinking around AI Agents and where we think there is room for continued innovation. We’re excited to co-author this post with our colleague Maria Gilfoyle. Also a special shoutout to our colleague Ted Kummert for being an incredible thought partner and contributor! Let’s dig in!
What are AI Agents (Reminder and Setting the Stage):
We view AI agents as one of the most significant shifts in UX since the introduction of the GUI, albeit more gradual than the transformation from character mode applications. AI agents can automate work and tasks traditionally performed by humans, creating better experiences for employees within enterprises and SMBs, as well as improving end consumer experiences.
Agents are systems that can perceive their environment & context, make decisions, take actions, and achieve specific objectives. With LLMs, systems will be able to work more independently with greatly enhanced context and decision-making capabilities. This opens a much wider range of applications for agents to work with humans in their daily lives as consumers, employees, and teammates. In addition, these capabilities will enable a whole new class of automation for business processes – enabling more agility, faster development, and automation of many more processes & tasks.
Simple Framework For Thinking About Agents:
We expect agents will work across a variety of verticals (i.e., personal finance, insurance, healthcare) and use cases (i.e., book a trip for me, do my taxes, or clean up messy data). Agents will be able to automate repetitive tasks and ad hoc tasks. Agents will help users in their personal lives and in their lives as employees or coworkers. Agents will also enable automation of a whole new range of processes & tasks in the enterprise. Essentially, agents will touch all aspects of the software and technology stack as we know it!
Below is a very simple taxonomy for how we think about AI agents and their functionality:
Who: Who are the agents targeting individuals or businesses?
What: What types of agents are being deployed? Are the agents solving ad-hoc tasks or repetitive tasks?
How: How much complexity is involved in the tasks? Do they require multi-step prioritization? Do they require multi-modality (vision, behavioral preferences, sensory data)? Do they work across heterogenous applications and data sources?
Where: Where are the agents being deployed in Enterprises, SMBs, or Consumer settings?
How We See AI Agents Impacting The New Age of Software:
Agent applications can or will be able to complete human-like tasks: There will be a class of standalone expert agents that complement humans or completely perform specific tasks or services (repeatable tasks). Most applications and services will incorporate these agents, and many legacy apps will introduce new capabilities in this manner, creating ripe competition for new companies.
The best new applications will be reimagining UX: We are moving towards a UX that goes from “point and click” first to natural language first. Companies like Glean and Perplexity have excelled in unlocking how users engage with agents. Successful companies will have compelling, reimagined UI/UX that is seamless to use.
The true problem unlock moment for agents is identifying use cases where there is an urgency for automation. There are different "types of work" where tasks are performed either once or frequently, and they can range from simple to complex, often requiring users to work across multiple applications and sites. We believe one of the keys lies in automating tasks that are performed frequently, are time-consuming, and have not been automated before. For many of these use cases, founders' domain expertise will be crucial.
There’s an emerging software infrastructure layer. Platforms designed for developers, such as Tiny Fish and Browserbase, are necessary to build agents at the application layer. The development of these platforms is still emerging and will play a crucial role in the proliferation of AI agents.
Business models and monetization strategies are evolving. Agents will ultimately face users and be utilized by apps and other agents. Monetization will likely occur through user subscriptions and consumption. There may also be potential for cost reduction and performance improvement in some use cases, making them economically viable at scale.
Use Cases Are Emerging Across Every Sector:
We are seeing opportunities for agent-based companies across consumer, enterprise, and SMB sectors, and in various verticals such as personal finance, insurance, and healthcare. We won’t dig into all the different use cases and opportunities we see in this post, but highlight a few below:
Consumer: Consumer AI agents are poised to provide seamless, low-friction utility that noticeably improves users' lives, especially for common mobile/web tasks. There is a major opportunity for consumer AI agents that can solve complex daily tasks, but they must be extremely easy to use without requiring users to teach the system. Consumers demand a quick "wow factor" and are unwilling to spend significant time learning new technology products. The key is delivering reliability and personalization with minimal setup effort.
Enterprise: Variable processes can now be automated across the enterprise. We believe there are many processes today that haven't been automated by applications due to their variability. LLMs can help automate variable processes across enterprises by providing customization and semantic understanding capabilities. For example, in cybersecurity, LLM-based agents can replicate analyst workflows, respond quickly to alerts, and allow human teams to focus on complex issues. For infrastructure management, LLM agents can continuously monitor and optimize cloud resources, acting as 24/7 cloud architects. LLMs enable data extraction from unstructured documents, automating end-to-end workflows in areas like supply chain management.
SMBs: Just like enterprises, small and medium businesses (SMBs) have many manual processes and ad-hoc tasks across multiple apps but lack resources to automate them. AI agents that can perform tasks autonomously can be incredibly beneficial for resource constrained SMBs. SMBs represent an underserved market segment that is open to adopting new technologies early, providing an opportunity for AI agent startups to gain initial traction before potentially moving upmarket.
In Conclusion:
We think there is an opportunity for AI agent applications to be vertical or horizontal, and net new multi-billion dollar companies will be created. Horizontal agent-based companies, like Perplexity and ChatGPT for consumers and Glean and MSFT 365 CoPilot for enterprises, have already gained significant traction. These companies are re-imagining workflows and interacting with information stored across various data sources to enhance user productivity.
Ultimately, the best AI agents will be able to automate both repetitive and ad-hoc tasks, assisting users in their personal lives and in their roles as employees or coworkers and enabling automation of a whole new range of processes and tasks in enterprises. We are excited to see how the space continues to evolve!
Please subscribe, and share the love with your friends and colleagues who want to stay up to date on the latest in artificial intelligence and generative apps!🙏🤖
 | A guest post by
|
Just wrote a piece on the fusion of LLM architecture and the emergence of JEPA. Would love your thoughts
https://open.substack.com/pub/matthewharris/p/what-comes-after-llms?r=298d1j&utm_medium=ios