Robust Agents Are All We Need: Faktory Emerges from Stealth Mode with a Private Alpha

Autonomous agents for repetable business tasks.

Image Credit: Faktory

Last year, I had the unique opportunity to incubate a new project in the autonomous agents space, alongside a phenomenal team and outstanding investors. After months of tackling numerous challenges, Faktory is emerging from stealth mode with a private alpha version, loaded with super cool features. Before I delve into everything about Faktory, I invite you to check out this video; hopefully, it will excite you to continue reading.

You can immediately sign up for Faktory : https://faktory.com/ but please keep reading :)

Why Work on Autonomous Agents?

Autonomous agents are captivating the imaginations of technologists in the generative AI space. I often think of autonomous agents as the automation paradigm of the AI era. Just as workflows or apps dominated automation in previous computing trends, autonomous agents are now being heralded as the new unit of automation in the generative AI world. Despite their notoriety, autonomous agents remain an unsolved challenge in generative AI, filled with obstacles. This was part of the inspiration behind Faktory. We aimed to see if we could create a platform and experience that streamline the adoption of autonomous agents for knowledge workers everywhere.

Before we go too far, let me clarify what I mean by ‘autonomous agents’, given that definitions can vary widely.

Autonomous Agents in Five Capabilities

Avoiding a philosophical debate about what agents are and aren’t, I’d like to highlight a few practical points relevant to anyone considering building autonomous agents in the real world.

In machine learning theory, the distinction between agents and models is that the former can perform actions in a given environment. Extrapolating this definition into practice quickly leads us to a series of key capabilities of autonomous agents, specifically in B2B scenarios:

1. Planning: An essential capability of autonomous agents is their ability to start with a higher-level objective and formulate possible plans.
2. Short-Term Memory: Agents need to remember elements of interactions with users, which typically requires a short-term memory component.
3. Long-Term Memory: Short-term memory is largely insufficient for autonomous agent scenarios as it sort of implies that the agents will start from scratch on every task. To address these limitations, autonomous agents typically need long-term memory capabilities.
4. Tool Integration/Augmentation: Integration with external tools and data sources is a key capability of autonomous agents in business scenarios.
5. Guardrails: Given the stochastic nature of foundation models, autonomous agents often require guardrails to regulate their behavior.

There are many other capabilities of autonomous agents, but the aforementioned five are essential for this technology to work in real-world scenarios.

Autonomous Agents: Simple to Envision, Brutally Hard to Implement

Despite autonomous agents being a widely understood concept in generative AI, they are incredibly difficult to implement. This was the most humbling lesson of the last few months working at Faktory. Below, I’ve listed some non-obvious challenges of building autonomous agents that we faced while working with our design partners:

1. Stochastic Genies: The stochastic nature of foundation models translates into unpredictable behaviors in agents, which can be impossible to control.
2. Agents Need a New UX Paradigm: Autonomous agents represent a new automation paradigm that requires a different user experience. Plain chat interfaces are not enough when we talk about business automation.
3. Tool Integration is Harder Than You Think: We tend to trivialize the task of augmenting foundation models with external tools, even though experience shows otherwise. Efforts like the ChatGPT plugins have seen very little traction. Mapping language tasks to concrete sequences of API calls is incredibly difficult.
4. Planning and Reasoning are in Early Stages: Planning and reasoning are fundamental capabilities of autonomous agents. However, the techniques in these areas are in a nascent stage and haven’t been battle-tested at scale.
5. Broad Tasks are Impractical: The nascent state of autonomous agents means that they tend to be drastically more impactful in relatively simple workflows. Automating very complex business processes with autonomous agents at the moment is a recipe for failure. We will get there, though.

These challenges shaped our thinking and the capabilities of Faktory over the last 10–12 months.

Faktory: Using Agents for Boring, Repeatable Tasks

Every knowledge worker has to deal with boring, repeatable tasks. Imagine having a workforce of autonomous agents at your disposal that learn about your company, integrate with your systems, and automate those tasks. Think about an agent that can integrate with your CRM and keep you up to date on developments with your top accounts. Or another that helps your sales team prospect potential customers by integrating with tools like LinkedIn or Apollo. Or some that can organize your project tasks in JIRA or transcribe videos of interviews given by your competitors.

Faktory is a platform for creating those types of agents. There are many capabilities baked into the Faktory platform, but there are a few that we believe are truly differentiated.

1. Agent-First UX: The Faktory team spent months reimagining a user experience for autonomous agents. We still have a lot of work to do, but the initial results seem quite encouraging.
2. Multimodal: Faktory is not constrained to OpenAI or major models; the platform integrates over 40 of the top foundation models in the market for modalities such as language, image, video, audio, 3D, and many others.
3. Planning: Another area we have spent a tremendous amount of time on is incorporating cutting-edge research for agent planning and reasoning.
4. Tool Integration: Faktory includes dozens of connectors to line-of-business systems, and we have streamlined the tool integration process to incorporate new connectors in hours.
5. Automation: Faktory is about automation, not only conversation. Agents on the platform can execute tasks based on events, on a schedule, or on demand.

Image Credit: Faktory

Faktory is available today in a private alpha. Over the last few months, we have been fortunate to engage with a group of design partners that have really pushed the boundaries of the platform. If you are working on use cases for autonomous agents, feel free to join the waitlist and reach out. I would love to hear from you.