This is exactly what an AI lab should be focused on: solving real, urgent problems for teams and businesses, not wasting resources on pointless projects like Sora 2 or hyping up erotic chatbot features as something revolutionary. Anthropic’s approach with Claude Skills finally shows practical value and enterprise impact, unlike the sideshow we keep seeing elsewhere.
This is such a sharp breakdown. I also think this shift in Anthropic’s strategy has a lot to do with something I read recently - that a big portion of their recurring revenue comes from a relatively small number of enterprise clients. Compared to OpenAI’s lower-ticket, high-volume model, that structure carries more risk if one major client churns. It makes sense they’re doubling down on long-term utility and deeper enterprise integration.
The progressive disclosure bit is key. I've built a skill that delegates entire tasks to a different model. Claude Code learns about a fact-checker via SKILL.md, shells out to OpenCode headless with Kimi K2.5, and the search results never touch Claude's context window. Keeps the token budget separate, costs way less than Max. Covered the full setup here: https://reading.sh/how-i-use-opencode-as-a-headless-worker-inside-claude-code-fed04b8358f9
“Claude can now use Skills to improve how it performs specific tasks. Skills are folders that include instructions, scripts, and resources that Claude can load when needed.”
This being release was a lower key moment but high impact. It’s my favorite feature.
This is exactly what an AI lab should be focused on: solving real, urgent problems for teams and businesses, not wasting resources on pointless projects like Sora 2 or hyping up erotic chatbot features as something revolutionary. Anthropic’s approach with Claude Skills finally shows practical value and enterprise impact, unlike the sideshow we keep seeing elsewhere.
Couldn't agree more.
I tried a very simple one for presentations - I can see a lot of potential.
Now we're talking!
This is such a sharp breakdown. I also think this shift in Anthropic’s strategy has a lot to do with something I read recently - that a big portion of their recurring revenue comes from a relatively small number of enterprise clients. Compared to OpenAI’s lower-ticket, high-volume model, that structure carries more risk if one major client churns. It makes sense they’re doubling down on long-term utility and deeper enterprise integration.
The progressive disclosure bit is key. I've built a skill that delegates entire tasks to a different model. Claude Code learns about a fact-checker via SKILL.md, shells out to OpenCode headless with Kimi K2.5, and the search results never touch Claude's context window. Keeps the token budget separate, costs way less than Max. Covered the full setup here: https://reading.sh/how-i-use-opencode-as-a-headless-worker-inside-claude-code-fed04b8358f9
The framing of skills as "institutional knowledge made executable" is spot-on. Skills compound.
Knowledge-accumulating AI: https://thoughts.jock.pl/p/wiz-personal-ai-agent-claude-code-2026
“Claude can now use Skills to improve how it performs specific tasks. Skills are folders that include instructions, scripts, and resources that Claude can load when needed.”
This being release was a lower key moment but high impact. It’s my favorite feature.