Subscribe | Advertise
Welcome back to Building AI Agents, your biweekly guide to everything new in the AI agent field!
In this agent’s defense, it only happened because it had a targeted ad egging it on.
This week, I connected with CopilotKit to write a special feature on their tech, which allows developers to easily build agentic assistants to serve their applications’ users. Check it out!
In today’s issue…
Building agentic apps with CopilotKit
DeepSeek surpasses OpenAI with new reasoning model
The 10 best free agent courses
What NVIDIA’s CEO got wrong about agents
…and more
📣 SPECIAL FEATURE: COPILOT KIT
The problem
As LLM capabilities advance, businesses are rapidly developing and rolling out agentic applications to automate their operations, but customer-facing agents capable of reliably providing high-quality interactions have lagged behind. AI agents have the potential to revolutionize how users interact with websites and other software—if the pain point of user interfaces can be overcome.
The solution
Brothers Atai and Uli Barkai founded CopilotKit to provide human-in-the-loop agents for any web or desktop application, bringing them past the era of button clicks and into one in which any piece of software can intelligently interact with its users.
“You provide your users with a partner that's always there. This partner is better at some things (e.g. it can read a lot more and faster than you), and worse at other things (e.g. coming up with good high-level objectives),” Atai told me. “You work side-by-side, almost as if collaborating with a colleague. Together, you produce top-notch work in hours, that would have taken days or weeks prior.”
The tech
CopilotKit’s open-source library uses React to integrate copilots into apps, facilitating user-agent interaction, from agentic UIs to human-in-the-loop and more. These agents can both intelligently answer users’ questions and autonomously take actions, such as updating their account information, like in this example of an intelligent copilot for a banking application. I tried it out and it nailed every query I threw at it, successfully performed the actions I asked for—and was impressively fast as well.
In addition to its main library, CopilotKit recently partnered with LangChain to develop CoAgents, which allows users to build CopilotKit apps powered by LangChain’s widely-used agent framework LangGraph.
Have a look at CopilotKit’s tech if you’re interested in using AI agents to power your user-facing applications!
If you find Building AI Agents valuable, forward this email to a friend or colleague!
💡 ANALYSIS
Source: Created by the author using Dall-E 3
How to adapt to the age of agents
Based on a survey of over 300 AI practitioners, this piece examines which workers and businesses will be impacted first, and gives tips on how to stay ahead.
OpenAI product chief on AI agents
An interview with OpenAI’s Chief Product Officer Kevin Weil in which he discusses the rise of agents, describing 2025 as the year “when we go from…using ChatGPT mostly to answer questions, to ChatGPT actually doing things for you in the real world.”
Jensen Huang got it wrong: who should really manage AI agents
The author of this piece takes issue with NVIDIA CEO Jensen Huang’s claim that “the IT department of every company is going to be the HR department of AI agents in the future”, arguing that IT departments are poorly situated from this role, and entirely new centers of excellence within companies will be required.
Deloitte’s key findings on enterprise agents
A summary of a new report on enterprise generative AI by Deloitte, finding among other things that nearly ¾ of respondents’ genAI initiatives are meeting or exceeding ROI expectations, and more than half are pursing AI agents.
🧪 RESEARCH
System diagram of IntellAgent | Source: arXiv
A multi-agent framework for evaluating conversational AI systems
This paper’s authors introduce IntellAgent, an agentic platform for evaluating other LLM-based systems which converse with users, enabling richer assessments of their performance that are tailored to their specific task.
An LLM agent for academic paper searching
PaSa is an autonomous agent designed to answer complex scholarly queries using published literature, outperforming baselines such as Google Scholar and ChatGPT.
Thanks for reading! Until next time, keep learning and building!
If you have any specific feedback, just reply to this email—we’d love to hear from you