ReadingDevRev’s Agentic AI: Powering human & AI connection
blog

DevRev’s Agentic AI: Powering human & AI connection

hero

AI agents are rapidly becoming the cornerstone of SaaS innovation, empowering companies and individuals to achieve more with less effort. They foster a deeper connection between humans and technology by anticipating needs, providing contextual assistance, and integrating smoothly into daily workflows. An agentic AI experience will allow users to seamlessly collaborate with intelligent agents that understand the user’s workflow and can couple personalization with task augmentation.

The past twenty years of SaaS required human inputs to deliver operational reports to managers, but the next decade will shift towards applications that listen to and work for the end user. Like the bot craze of the 2010s, driven by improvements in natural language understanding models, AI agents powered by large language models (LLMs) present a unique opportunity to redefine the software landscape's user experience.

What is an AI Agent?

At a high level, AI agents function like employees with specific skill sets. They are particularly valuable because you can create multiple agents, similar to a team, to tackle complex tasks by breaking them down into smaller components. Each agent handles a component independently, in the correct sequence, and can leverage other agents’ domain expertise to fully complete a task. The scale, complexity, and most importantly, utility of AI agents have created new excitement in their application.

As Andrew Ng points out:

“Today, we mostly use LLMs in zero-shot mode… This is akin to asking someone to compose an essay from start to finish…With an agent workflow, however, we can ask the LLM to iterate over a document many times. For example, it might take a sequence of steps to Plan an outline, Decide what, if any, web searches are needed to gather more information, Write a first draft, Read over the first draft to spot unjustified arguments or extraneous information, Revise the draft taking into account any weaknesses spotted, And so on…This iterative process is critical for most human writers to write good text. With AI, such an iterative workflow yields much better results than writing in a single pass.”

Building on Andrew’s research, I believe much of our knowledge work and team collaboration will be complemented by AI agents. At DevRev we are delivering on this hypothesis by creating a platform, AgentOS, to customize the discrete set of skills available to an AI agent. Skills define the specific capabilities an AI agent is trained to perform and the data authorization policies. These skills enable the agent to perform particular tasks such as data generation, data analysis, semantic search, and multi-step workflows. Each skill addresses a specific aspect of a larger problem, allowing the AI agent to execute tasks independently or in coordination with other agents to aid in the completion of a goal.

In summary, AI agents wrap around a large language model (LLM) to autonomously handle various tasks. Instead of manually prompting an LLM like ChatGPT, you can train an agent to perform a series of actions, enabling it to execute multiple tasks in a specific order to solve a problem.

The What and How of AI Agents.png

Empowering AI Agents

If AI agents will dramatically change our ability to perform complex tasks, what are the building blocks that empower AI agents to do great work? From an agent creation perspective, there will be many tools to define an agent’s capabilities. In the case of global tasks like web search, summarization, and writing a document outline, they will likely be very similar. However, the largest gains will come from their ability to work seamlessly with the applications we use daily and be securely fine-tuned on private data. Whether the agent is working as a personal assistant or as a senior-level employee within a specific domain (e.g., Senior PM, Developer, Support Engineer), successful implementations of AI agents will require native platform integration. Agents will move faster, generate more contextual answers, and understand relevance with higher accuracy when the APIs, data, workflows, and actions they take are centralized. This presents a complex problem, as modern compute companies are typically thin services, while legacy enterprise applications contain much of the business-specific data and processes required to guide an agent.

AgentOS: Streamlining Agent Adoption

To bridge modern compute with legacy platforms, DevRev developed AgentOS to connect humans with agents on one system. Agents are armed with the data and tools they need to perform a wide range of tasks efficiently, and humans are provided with a modern application interface to both define the skills of their agents effortlessly and collaborate with them in realtime. With over 30 integrations, AgentOS indexes disparate data into a knowledge graph that connects the work you do to its relationship with your product and customers. This is coupled with semantic search to support information lookups, a serverless workflow engine to drive real-time automation, and an in-browser analytics engine that empowers AI agents to query and join data for further analysis.

Your agents are only as good as the context you provide them. While the broader industry is offering “seamless integrations” with CRM platforms by pulling in data at runtime, we have taken a different approach by continuously syncing events from all your systems and indexing it during the initial setup. We also maintain a two-way sync to ensure that the indexed data is always up-to-date between legacy systems and your new modern human + AI application.

AgentOS_ Streamlining Agent Adoption.png

AI Agents in the Workplace

Let's look at an example of how this matters. Today, tons of internal content, conversations, meetings, and external documentation contain information on products that need to be disseminated across engineering, product, and sales teams. Teams may consolidate this information for easier discovery, or a well-constructed federated search may provide the ability to search across channels, but fetching the information is not the actual task. For instance, a customer success rep may need to consolidate information from the past, understand its relevance today, view previous account plans for how to position new product information to a customer, and ensure the proper information is being presented to a customer in a manner authorized by the company. With AI agents, these tasks can be consolidated to understand the access privileges of a particular role, collect the information, and calculate relevance based on time. Furthermore, they can build the right external material to present information to a customer that meets the internal policies of the organization.

Another example is incident management and bug resolution, where the AI agent is tasked with reactive or proactive discovery and remediation of problems. The first task of discovery entails turning signals into incidents or associating signals with existing incidents to avoid duplication. The agent may monitor signals from Datadog logs, user sessions, PagerDuty alerts, etc. In the case of remediating the incident, the agent would then create the PR, mention the appropriate team members, recall previous history, summarize progress, and analyze the blast radius of the incident across customers to identify severity and customer notifications.

These tasks are cumbersome and repetitive, making them ideal for AI agents to handle, saving countless hours while driving true business outcomes that lead to new revenue or saved costs across an organization.

AI Agents in the Workplace.png

AI Agents Driving Customer Experience

AI agents have the potential to reshape customer experiences within software applications for customer support and customer success/growth. These agents can seamlessly execute a series of tasks in a specific order, allowing them to address complex customer issues without the need for manual intervention at every step. By leveraging AI agents, companies can provide faster, more accurate, and personalized support, leading to increased customer satisfaction and retention. Furthermore, these agents can proactively identify growth opportunities and deliver tailored recommendations, driving customer success and fostering deeper, more meaningful engagement. This transformative capability not only enhances the efficiency of customer support operations but also empowers businesses to build stronger, more resilient customer relationships by connecting with their customers in more personalized and impactful ways.

AI Agents Driving Customer Experience.png

Parting Thoughts

Our future work experience will be AI integrated, where AI agents work autonomously on our behalf and will require a platform that provides context, data connectors, and seamless workflow integrations. AI agents' skills must be defined by their ability to work autonomously, perform analysis, and augment human efforts. Successful AI agents will be delivered through a modern user experience, collaborating with humans both within your work management application and directly with your customers, ensuring a consistent experience for internal team members and external customers.