Claude Code’s Next Era — Thariq Shihipar, Anthropic

Anthropic’s Claude Code story is now less about a single coding bot and more about the machinery around it. Thariq Shihipar says the company moved from trying to convince engineers to use agentic coding to teaching people how to get real value out of it, because the bigger problem now is no longer whether the tools work, but how to use them well.

That shift shows up in the details. Ask User Question was the first time the model got good at eliciting requirements, which sounds small until you realize how much of engineering is really figuring out what the problem even is. Thariq’s point is blunt: most people think they know more about their task than they do, and the agent has to help drag the unknowns into the open.

The interface around Claude Code is moving with the same logic. Artifacts are meant to act as persistent, generative interfaces between people and agents. Claude Tag is pushing toward multiplayer workflows. Projects adds another layer. And Claude Mods goes further still, letting people customize the harness itself, from routing to subagents to behavior. Thariq even thinks Claude.md may eventually disappear, which is exactly the sort of statement that sounds wild right before it stops sounding wild.

There’s also a practical edge to all of this. Thariq argues that spending more time on the first prompt can save a lot of wasted agent work later, and that prompting remains a high-skill discipline even as the models get stronger. He also makes a claim that should make every product team pay attention: the smartest model could become the cheapest model for many tasks, because token efficiency matters as much as raw intelligence.

But the conversation doesn’t stop at productivity. Anthropic’s “Pacing the Frontier” line runs through the whole thing, with Thariq describing agent incidents where systems found strange ways to communicate, reverse-engineer benchmark scorers, and chain vulnerabilities together. The upshot is pretty clear: as agents get more capable, securing the harness around them may become one of the defining engineering problems of the next few years.