We want to shed light on a rapidly evolving development that has the potential to reshape the B2B AI landscape: Agentic Workflows and AI Agents.

While still in their infancy, AI Agents are rapidly climbing the capability ladder. Agents like Devin “the software developer”, Norm.ai, Sema4.ai and open-source alternatives along with established players like Zapier and UiPath, aggressively implementing agentic workflow capabilities into their offerings suggest that we are closer to a new paradigm of AI use than we thought a few months ago. Google's recent announcement of a suite of "AI Agents" designed as personal assistants further solidifies this trend as an industry-wide reality unfolding at an astonishing pace.

Google’s former boss Eric Schmidt expects that we’ll have autonomous agentic systems working for us in the background within the next five years as we get across three capability ladders: (1) an infinite context window, (2) enhanced chain-of-thought reasoning and agency, and (3) text-to-action programming.

(1)  An infinite context window: The context window is the prompt that you ask. That context window can have a million words in it. And this year, people are inventing a context window that is infinitely long. This is very important because it means that you can take the answer from the system and feed it back in and ask it another question in an almost infinite loop.

(2)  Enhanced chain-of-thought reasoning and agency: A technique that encourages large language models to breakdown complex problems into a series of intermediate steps, mimicking the intuitive process that humans use when solving multi-step reasoning tasks. Here's how it works: The model is prompted with a few examples that demonstrate a "chain of thought" - a sequence of logical steps that lead to the final answer. These examples serve as a way to instruct the model to generate its own chain of thought when attempting to solve new problems. It’s increasingly generalizing really well. In five years, we should be able to produce 1,000-step recipes to solve really important problems in medicine and material science or climate change. This will lead to enhanced agency. An agent can be understood as a large language model that can learn something new. An example would be that an agent can read all of chemistry, learn something about it, have a bunch of hypotheses about the chemistry, run some tests in a lab and then add that knowledge to what it knows.

(3)  “Text to Action” capacity for programming: The capacity to orchestrate software code via text prompting has developed considerably. Now you can simply say to an AI “Write me a piece of software todo X” and it does. AI’s capabilities continue to advance beyond the zero to one draft, but now can write python to tools and architect new applications.       

The evolution of these capabilities will advance in a cumulative manner, progressing every month, every six months and so forth. The reason being is that’s there is so much money being invested in this path and so many ways in which people are trying to accomplish this. While the big players have the large so-called frontier models at OpenAI, Microsoft, Google and Anthropic, you also have a very large number of players who are programming at one level lower at much lower costs, all iterating very quickly. At some point, these systems will get powerful enough that the agents will start to work together to solve problems.

 “In the future all human interaction with the digital world will be through AI agents.” – Yann LeCun

While we're still in the early wave of B2B AI applications, the ethos of mapping and owning the workflows and jobs to be done (JTBD), as explored in our Whitepaper, is already playing out paving the way for proactive automation and broader, more complex agentic workflows in the future.

Defined is working closely with our portfolio companies and supporting the product strategy and technology roadmaps to further agentic workflows. In particular, Arcanna AI is codifying agentic workflows into their AI copilot product to advance cybersecurity while Quandri is optimizing operations in insurance with software robots.

In Part 2 we will explore the Design Patterns for AI agents to accomplish specific and increasingly complex tasks and in Part 3we will examine Blueprints for AI-native approaches to building applications that reimagine workflows from first principles.

If you are a founder or know of people working to increase automation with agentic workflows and further the capability ladder of AI Agents, then please reach out to connect.


If you’re a founder shaping the frontier, we want to hear from you.

Get In Touch