What does Claude Tag do differently from Claude?

Claude Tag is more delegated, asynchronous, and proactive. Rather than responding to individual requests, teams instruct Tag to monitor feedback channels and proactively take on tasks within a part of the codebase it is responsible for—what Anthropic's Mike Krieger described as a "multiplayer, async, proactive" way of working.

What are the main safety controls teams are using for agents today?

Human approvals and permissions are the two leading safeguards, followed by a scattered collection of task decomposition, retrieval, memory, and sandboxing techniques. As one speaker noted, "nobody has settled the control layer for agents."

How much has agent adoption grown?

According to Amplify's survey, 95% of respondents now use agents, roughly double last year's share. Among teams using agents, 89% said those agents could write data, up from 52% the previous year.

Back to articles

AI engineers debate whether autonomous loops are ready for prime time

Latent Space2d ago5 min read

Key takeaway

At the AI Engineer World's Fair, engineers sharply disagreed on whether autonomous agent loops—systems that iteratively write and improve code—are ready to replace traditional software development. While 95% of surveyed teams now use agents and 89% report that agents can write data, industry safeguards remain primitive, costs limit adoption for over three-quarters of teams, and 59% worry that AI-generated code is creating lasting technical debt. The debate captures a real engineering gap: the tools work, but the disciplines and controls needed to run them reliably in production are not yet solved.

Summaries like this, in your inbox every morning.

3 Key Points

What happened
At the AI Engineer World's Fair, a panel debated whether autonomous software loops—agents that iteratively write and improve code—are viable now or overhyped. Supporters like Geoffrey Huntley argued loops are inevitable and already here; skeptics like Dex Horthy countered that the engineering discipline hasn't caught up with the hype, and that deterministic control (like Kubernetes uses) is fundamentally different from agentic loops.
Why it matters
The survey data shows agents have moved fast—95% of respondents now use them, up from roughly 50% last year, and 89% of teams using agents say those agents can write data. But the industry still lacks standardized safeguards (human approvals and permissions remain the leading controls), cost concerns limit ambition for 76% of teams, and 59% fear today's AI-generated code is creating long-term liabilities. The debate reflects a real tension: AI has made coding cheaper and faster, but the engineering practices needed to run production systems reliably at scale are not yet settled.
What to watch
Anthropic's Claude Tag, described by Mike Krieger as more delegated, asynchronous, and proactive than Claude, may offer a concrete window into what early software-factory setups look like in practice. Token usage is now the second-most monitored production metric, behind quality, signaling that cost control will be central to how widely these tools scale.

FAQ

What does Claude Tag do differently from Claude?: Claude Tag is more delegated, asynchronous, and proactive. Rather than responding to individual requests, teams instruct Tag to monitor feedback channels and proactively take on tasks within a part of the codebase it is responsible for—what Anthropic's Mike Krieger described as a "multiplayer, async, proactive" way of working.
What are the main safety controls teams are using for agents today?: Human approvals and permissions are the two leading safeguards, followed by a scattered collection of task decomposition, retrieval, memory, and sandboxing techniques. As one speaker noted, "nobody has settled the control layer for agents."
How much has agent adoption grown?: According to Amplify's survey, 95% of respondents now use agents, roughly double last year's share. Among teams using agents, 89% said those agents could write data, up from 52% the previous year.

Discussion

No discussion yet for this article

Stay ahead with AI news

Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.

Get Started Free

Free · takes 30 seconds · unsubscribe anytime

1 minute a day. The AI essentials.

200+ sources · Email / LINE / Slack

Get it free →