AI Agents in Practice • Henrik Kniberg • GOTO 2025

Item: AI Agents in Practice • Henrik Kniberg • GOTO 2025
Rating: 4
Author: PostSingularity Institute

by Henrik Kniberg

Applied Sciences & Engineering AI — News & Applications UYComputer science UYQArtificial intelligence (AI)

🎙 Henrik Kniberg 👥 1.1M 📅 March 25, 2026 ⏱ 39 min 👁 3K 📄 expert opinion 🧭 2026-08-02

Available in: English (current) Français

Keywords

AI agentautonomytoolshuman-in-the-loopagent platform

Summary

Henrik Kniberg, co-founder of Abundly.AI, shares practical insights from building AI agents over 2.5 years. He defines an AI agent as an autonomous digital entity using an LLM as its brain, with a mission and tools to interact with the world. He demonstrates creating an agent via a no-code platform, showing how it can check GitHub, post to Slack, make phone calls, and handle email. He emphasizes the importance of combining tools and allowing agents to interact via human channels like phone and email. He discusses the spectrum between code and humans, positioning agents as a middle ground. He highlights the need for a platform with instructions, documents, capabilities, and logging. He stresses that content (data and documents) is crucial for agent effectiveness. He provides design tips: start simple, use human-in-the-loop, give clear instructions, and monitor agent behavior. He showcases agents creating their own tools, a powerful capability. He concludes that agents will become like colleagues, and the future involves easy creation of agents by non-programmers.

166 words

Critical Evaluation

The talk provides a valuable, hands-on perspective on AI agents, grounded in real-world experience. Kniberg’s definition of an agent, emphasizing autonomy and tool use, is practical and aligns with industry trends. The live demos effectively illustrate key concepts, such as human-in-the-loop authorization and agents writing their own instructions. The discussion of the code-agent-human spectrum offers a useful framework for understanding agent capabilities and limitations. However, the talk lacks formal citations or references to academic work, relying on anecdotal evidence. The speaker’s role as a vendor of an agent platform introduces potential bias, though he does not overtly promote his product. The technical depth is moderate, suitable for a general technical audience but not for those seeking deep architectural details. The advice on content being king and the importance of data is well-taken, but could be elaborated. Overall, the talk is informative and practical, but its rigor is limited by its anecdotal nature and lack of external validation.

157 words

Title / Content Match

Title accurately reflects the content: a practical talk on AI agents.

Quality & Reliability

8/10

Speaker is a recognized expert with hands-on experience; practical demos and concrete examples; but no formal citations or rigorous methodology.

Chapters

Cited Sources

Henrik Kniberg's blog — Speaker's blog with related articles.
GOTO Copenhagen 2025 session page — Session details and slides.
GOTO Conferences on Bluesky — Conference social media.
GOTO Conferences website — Conference homepage.
GOTOpia — GOTO's content platform.
GOTOpia newsletter — Newsletter signup.
GOTO Conferences LinkedIn — Conference LinkedIn page.
Henrik Kniberg's LinkedIn — Speaker's LinkedIn profile.
GOTO Conferences YouTube channel — Channel subscription link.

Concurring Sources

ReAct: Synergizing Reasoning and Acting in Language Models — Supports the idea of agents combining reasoning and tool use.
Toolformer: Language Models Can Teach Themselves to Use Tools — Aligns with the concept of agents using tools.

Dissenting Sources

On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜 — Raises concerns about the reliability and biases of LLMs, which underpin agents.

Contribution & Novelties

The talk offers practical, experience-based insights into building and deploying AI agents, emphasizing autonomy, tool integration, and human-in-the-loop design. It highlights the importance of content and data for agent effectiveness, and showcases agents creating their own tools, a relatively novel capability.

Pour aller plus loin :

AI agent - Wikipedia — Provides a broader definition and context for intelligent agents.
ReAct: Synergizing Reasoning and Acting in Language Models — A foundational paper on combining reasoning and acting in LLM-based agents.
Toolformer: Language Models Can Teach Themselves to Use Tools — Explores how LLMs can learn to use tools, relevant to agent tool use.
AutoGPT — An open-source project demonstrating autonomous agents.
LangChain — A framework for building agents with LLMs.

120 words

Radar Profile

The radar profile shows strong scores in information quality and technical level, with moderate quantity and reliability. This indicates a well-articulated, expert-driven talk with practical value, but limited breadth and formal rigor.

Reliability 7/10