Agentic Engineering & ToolingDeep Dive180min
Specialized AI Agents: from concept to production
The proposal shows how to build cheaper, specialized AI agents with minimal code using Docker Agent. It covers creating disposable agents quickly, iterating to solve real business problems, adding tools, evaluating and optimizing performance, monitoring costs, and connecting multiple agents for production collaboration.
talk.summaryAiDisclaimer
David GageotDocker
talkDetail.whenAndWhere
Tuesday, October 6, 13:30-16:30
TBA 2
talks.roomOccupancytalks.noOccupancyInfo
More and more of us use **Coding Assistants** that excel at **writing code**.
They are **good** and they are **expensive**!
For our business needs, we need **more specialised AI agents** that are as good
or ever better, but for a much **cheaper** price.
The same underlying models can be used to write
more **specialized, cheaper, more predictable AI agents**.
We'll start by writing very disposable agents in literally seconds, **without code**.
And then we'll demonstrate how to grow such agent into solving a real business case.
We'll iterate on the agent, we'll give it tools and then we'll evaluate it and optimize it with **solid evals**.
We'll monitor its costs and see how changing models help or degrade.
Finally, we'll discover how multiple agents can connect and collaborate in production
with **MCP, ACP and A2A**.
All of that can be done with almost no code, leveraging [Docker Agent](https://docker.github.io/docker-agent/),
an open-source project and a real Swiss Army knife for AI agent creation.
They are **good** and they are **expensive**!
For our business needs, we need **more specialised AI agents** that are as good
or ever better, but for a much **cheaper** price.
The same underlying models can be used to write
more **specialized, cheaper, more predictable AI agents**.
We'll start by writing very disposable agents in literally seconds, **without code**.
And then we'll demonstrate how to grow such agent into solving a real business case.
We'll iterate on the agent, we'll give it tools and then we'll evaluate it and optimize it with **solid evals**.
We'll monitor its costs and see how changing models help or degrade.
Finally, we'll discover how multiple agents can connect and collaborate in production
with **MCP, ACP and A2A**.
All of that can be done with almost no code, leveraging [Docker Agent](https://docker.github.io/docker-agent/),
an open-source project and a real Swiss Army knife for AI agent creation.