In our “Engineering Energizers” Q&A series, we explore the journeys of engineering leaders who are driving innovation in their fields. Today, we feature Ambika Sukla, Lead Enterprise AI Architect at Salesforce. Ambika manages the development of the Enterprise Architecture Agent (EA Agent), an AI tool designed to streamline enterprise architecture processes through intelligent automation.
Discover how his Enterprise Architecture team tackled the challenge of processing and normalizing vast amounts of documents for consistency, achieved high accuracy in multi-hop reasoning for complex queries, and embedded trust and security into rapid development timelines, among other achievements!
What is your team’s mission?
The team’s mission is to create scalable, AI-driven tools that empower architects to streamline complex processes and maintain high standards of compliance and security. We developed the EA Agent to support this mission by assisting architects in creating, reviewing, and optimizing system designs. By leveraging Data Cloud to process 100,000 documents, the EA Agent provides actionable insights, identifies compliance gaps, and helps architects align their work with Salesforce standards.
The EA Agent is built on Agentforce, Salesforce’s internal platform for developing AI-driven agents. Using Agentforce, the team streamlined the creation and deployment of the EA Agent, integrating advanced AI models into a scalable and adaptable framework. This ensures the agent is both efficient and responsive to evolving enterprise needs.
Ambika explains why innovation is everyone’s responsibility at Salesforce Engineering.
What challenges did your team face in preparing and managing data for the EA Agent?
Preparing and managing the EA Agent’s dataset involved significant challenges, primarily in processing and normalizing 100,000 documents from various sources, such as Salesforce manuals, architecture guidelines, and security policies. The team had to ensure data consistency and seamless indexing for effective retrieval, which was complicated by the inconsistent formats of these documents.
To address this, the team reformatted the documents to maintain structural integrity during ingestion into Data Cloud. They also applied metadata tagging to group related content and improve query performance, ensuring the dataset was not only structured but also retrievable under real-world conditions.
Another key challenge was integrating the dataset into the Data Cloud at scale. This required the development of indexing strategies capable of balancing rapid data retrieval with accuracy. Testing these strategies involved mock queries designed to simulate real-world usage, ensuring the EA Agent could provide architects with reliable and relevant insights in production.
What technical hurdles did you encounter while ensuring the EA Agent achieves high accuracy and reliability?
The EA Agent’s design had to support multi-hop reasoning, where queries require synthesizing information from multiple interconnected documents. This process demanded the development of algorithms capable of navigating up to 10 documents in a single query, extracting relevant details while filtering out noise. A key challenge was maintaining contextual consistency, especially when documents contained conflicting or outdated information.
Another hurdle was the complexity of real-world queries. To address this, the team created a robust test suite of over 1,000 questions, sourced directly from architects’ scenarios, to evaluate and refine the agent’s accuracy. These questions covered edge cases such as ambiguous phrasing or complex standards, which are common in enterprise architecture documents. Continuous iteration based on test results allowed the team to enhance the EA Agent’s ability to deliver precise and reliable insights.
Leveraging Agentforce and Data Cloud for architecture agents.
How did you balance the EA Agent’s rapid development timeline with Salesforce’s stringent trust and security standards?
Delivering the EA Agent within just three months required embedding trust and security into every stage of development. Instead of treating compliance as an afterthought, the team worked closely with the Enterprise Architecture Review Board (EARB) to align with Salesforce’s trust principles.
EARB reviews were conducted iteratively, allowing the team to address feedback in real time while maintaining momentum. These iterative reviews were paired with fast-track approval mechanisms, enabling key stakeholders to provide timely input on critical updates, such as data governance policies. This collaborative approach ensured that compliance issues were addressed early without delaying progress.
Early engagement with Salesforce’s AI Council further ensured that ethical AI considerations were integrated into the agent’s design. The combination of iterative feedback, rapid approvals, and close collaboration allowed the team to successfully meet the aggressive three-month timeline while adhering to Salesforce’s high standards for trust and security.
What strategies does your team use to ensure enhancements in one area of the EA Agent don’t compromise other areas?
Every enhancement to the EA Agent undergoes rigorous testing to ensure it aligns with the tool’s overall goals without creating regressions. For example, if retrieval algorithms are updated to improve speed, the team evaluates potential impacts on contextual accuracy and relevance.
New enhancements are first tested in isolated environments using mock queries to identify potential conflicts before deployment. This ensures that any unintended side effects, such as slower data retrieval or irrelevant results, are identified and resolved. The team’s structured testing framework also incorporates challenges encountered during real-world usage, which are added to the growing test suite to expand its coverage.
These testing environments are critical for maintaining system-wide integrity. By isolating each enhancement before rollout, the team ensures that improvements strengthen the EA Agent holistically without introducing risks to existing functionality.
What makes the EA Agent unique compared to other internal tools at Salesforce?
The EA Agent automates architecture processes while leveraging a vast knowledge base. Unlike traditional tools that require manual effort, the EA Agent provides real-time guidance and actionable feedback by integrating insights from Data Cloud.
Its pre-review functionality identifies compliance gaps and suggests improvements before architects submit their documents, saving significant time. By reducing the burden of manual reviews, the EA Agent allows architects to focus on innovation rather than administrative tasks.
What further sets the EA Agent apart is its ability to synthesize guidance from multiple sources. For example, architects no longer need to cross-reference standards, security policies, and best practices manually—the EA Agent consolidates all relevant insights into a unified response. Future phases of development aim to introduce generative AI capabilities, which will enable the agent to draft complete architecture documents tailored to Salesforce standards.
Ambika explores how Salesforce Engineering innovation thrives when everyone contributes and inspires others.
Can you explain the three-phase development plan for the EA Agent?
The EA Agent’s development is structured into three strategic phases:
- Phase 1: Focuses on data ingestion and providing architects with guidance for document creation. This phase includes building foundational capabilities for referencing documents, applying metadata tags, and ensuring consistency in retrieved insights. This phase concludes in February 2025.
- Phase 2: Introduces automated document reviews, leveraging historical review data to identify gaps and suggest improvements. This phase integrates machine learning models trained on past review patterns to flag common issues architects encounter.
- Phase 3: Adds generative AI capabilities, enabling the agent to draft architecture documents that align with Salesforce’s standards.
Each phase builds on the last, enhancing the agent’s functionality while maintaining a focus on usability and trust. These planned milestones ensure that the EA Agent evolves efficiently to meet architects’ changing needs.
What ongoing research and enhancements are planned for the EA Agent?
Ongoing research is focused on fine-tuning the EA Agent’s ability to handle complex queries and improve multi-hop reasoning. Future advancements aim to enhance the agent’s contextual accuracy, allowing it to retrieve and synthesize information seamlessly.
The team is also researching new ways to optimize data ingestion pipelines to reduce latency when handling real-time queries. This includes exploring more efficient indexing algorithms and testing next-generation machine learning models.
Additionally, preparations for the generative capabilities outlined in Phase 3 are underway. Researchers are prototyping methods for generating architecture documents while ensuring strict adherence to enterprise standards. These efforts aim to expand the EA Agent’s value as an indispensable tool for Salesforce’s enterprise architecture workflows.
Learn more
- Stay connected — join our Talent Community!
- Check out our Technology and Product teams to learn how you can get involved.
