Antje Barth

Antje Barth develops and teaches practical AI education and explains how agents connect models, tools, and cloud services. She co-developed DeepLearning.AI and AWS’s Generative AI with Large Language Models course and coauthored Generative AI on AWS. AWS’s author biography identifies her as a Principal Developer Advocate for generative AI.[5][4][2]

At AI Engineer World’s Fair 2025, Barth presented Building Agents at Cloud Scale. She advocated a model-driven approach: developers specify the task and available tools, while the model handles planning and action selection. Describing the approach behind the open-source Strands Agents SDK, she urged teams to “let the developers focus on what their agent should do rather than telling it exactly how to do it.”[7]

Her Amazon Q Developer CLI demonstration made that division concrete. Asked about Amazon Bedrock, the agent selected a tool exposed by the AWS documentation team’s Model Context Protocol (MCP) server, requested permission, and returned an answer grounded in official documentation. Barth reported that AWS teams had built and shipped the agentic service within three weeks, treating delivery speed as an example of execution’s competitive value.[7]

Barth also explained how retrieval can help an agent choose among a large tool catalog. She reported an internal AWS agent managing more than 6,000 tools: the team stored tool descriptions in a knowledge base, retrieved candidates relevant to the task, and loaded only that subset into the model’s context. This separates finding plausible capabilities from deciding which action to take.[7]

A second demonstration moved a dice-rolling MCP tool from a local process to AWS Lambda, using Streamable HTTP, an API Gateway endpoint, an authorizer, and DynamoDB session storage. A Strands client discovered the remote tools and answered a request to roll a 20-sided die; the result was seven. Barth credited Mike Chambers with building the demonstration and described its bearer-token authorization as a simple concept demo that could be developed into a more robust integration.[7]

Before these agent demonstrations, Barth worked on education covering the broader LLM development lifecycle. In June 2023, she developed and delivered Generative AI with Large Language Models with Chris Fregly, Shelbee Eigenbrode, and Mike Chambers. The team developed the content under Andrew Ng’s guidance, with input from experts and applied scientists at Amazon, AWS, and Hugging Face.[5]

The course paired practical application with the science behind LLMs. Its labs progressed from prompting for conversation summarization to parameter-efficient fine-tuning, then reinforcement learning from human feedback. Comparing fine-tuning results with the earlier prompting lab gave learners a concrete way to examine the qualitative and quantitative effects of adapting a model. The final lab used a reward model and proximal policy optimization to train for less harmful responses, then evaluated harmlessness before and after training.[5]

Barth’s book work complemented her teaching. O’Reilly credits Chris Fregly, Barth, and Shelbee Eigenbrode as the coauthors of Generative AI on AWS, dated November 2023. She also coauthored Data Science on AWS.[4][2]

Earlier in her career, Barth worked in technical evangelism and solutions engineering at Cisco and MapR, focusing on data center technologies, big data, and AI applications. Those roles provide technical context for her subsequent cloud AI education. She also co-founded the Düsseldorf chapter of Women in Big Data.[3][2]