Agent "Skills": The discussion is lively, but actual usage is almost zero.

SnapshotBot · 2026-03-29T11:25:01+00:00

AI engineer Dax Raad criticizes the current trend of modularizing agent skill modules, emphasizing low actual usage rates, and calls for the industry to remain calm and avoid passionate but unsubstantiated spending. He points out that building infrastructure does not align with user needs, leading to information gaps, and recommends that teams focus on real data to guide decision-making.

SnapshotBot

2026-03-29 11:25:01

Abstract generation in progress

Headline

Agent Skills: AI Engineer Calls Out the Recent Hype

Summary

Dax Raad, an AI engineer from OpenCode and SST, recently posted an awkward fact: the modular “skills” designed to enhance agent capabilities, while giving rise to various standards, supporting tools, and endless discussions, are not installed by the vast majority of users at all. He urges the industry to calm down: many ideas ultimately fall flat, and enthusiasm without evidence is just burning money. His words directly hit the pain point of the AI tools community—those who create and those who use are completely on different wavelengths.

Analysis

Background: Raad has worked on AI coding tools like OpenCode and has consistently spoken out about the issue of “overbuilding” in the cloud and AI fields.
The “skills” ecosystem he refers to includes:
- The discoverable endpoint proposal by Cloudflare (/.well-known/skills)
- The .agents/skills directory structure agreed upon in frameworks like Claude Code and LangChain
The original intention behind these items is good: to make agent capabilities modular and interoperable. But the actual adoption rate is very low, and the reasons are not complicated:
- Implementation is troublesome and the landing costs are high;
- There is overlap with existing custom command functionalities, and it’s unclear how much new value it really brings.
This is actually an old problem in the AI tools community: first lay the infrastructure, then discuss needs later. There are indeed voices in the community claiming that “skills” can improve task success rates, but very few are actually using them.
Raad has over 120,000 followers on social media, making such questioning easily amplified. Perhaps it can push some teams to return to reality, first checking if anyone really needs this before deciding whether to continue investing.

Expectations vs Reality

Dimension	Initial Concept	Actual Situation
Modular/Interoperable	Skills plug-and-play, ecosystem interoperability	Many standards, complex implementation, poor interoperability
User Value	Increase agent success rates and scalability	Overlap with existing command systems, users lack motivation to switch
Adoption Path	Built-in standards in frameworks to promote adoption	Most users have not installed any skills

To put it bluntly:

Low adoption rate is a fact, not noise, and we can’t cover up the problem of nonexistent demand by creating more standards;
Without real usage data, massive investment in infrastructure is a poor cost-benefit ratio;
There is a serious information gap between those who create and those who use, and we need real usage data and retention data to calibrate direction.

Impact Assessment

Importance: Medium
Category: Industry Trends, Developer Tools, Technology Observation

Verdict: If you are following the “skills standard” line, it’s not too early to get involved now—this concept is currently in an awkward stage of weak validation. Those who can truly benefit are the teams willing to hit the brakes and use actual usage data to guide decisions, rather than those who continue to pile on standards.

View Original

This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.