Artifacts in Claude Code are now also available on Pro and Max plans.
Ask for an artifact, Claude writes the code, publishes it live to claude.aโi, and updates it in real time while it keeps working. Pages are private to your account and fully self-contained.
@Gamingtronium I believe engineer B also generated the code but as he spent time on unit test, integration test, documentation and reviews so.
but think like, engineer A just vibe coded the feature but on the otherhand now engineer A becomes engineer B, and let the feature done.
@claudeai
I just got weekly limit today suddenly in my account yet I didn't used that much tokens and in my current session, I have usage of 27%. Today, morning when I was working with Claude Code, I checked my weekly limit around 60%, so my question is within few hours, it is reached to 100%, is it possible?
I am on my MAX plan and so my team is also in MAX plan, everyone faced the same issue.
@narendramodi I am in India and building AI development agency which build and trains LLMs, ML models and Deep Learning. I wish and I hope, we could help India and Govt to do AI Enablement. I tried a lot to contact to Govt. to do AI Enablement within it but never got an opportunity due to others.
An AI agent that calls tools one by one is not an agent.
It's an expensive middleman.
Here's what we changed, and what it actually looks like in production.
The engineering moat is not in how many tools your agent can access.
It's in whether your agent is designed to think, or designed to execute.
The best systems know exactly when to do which.
We read the AWS blog on PTC, tried it the same week.
We were already building agents with the Anthropic SDK on Bedrock, so we used the SDK-compatible proxy approach, same codebase, different execution layer.
No full rebuild. Just a smarter design.
https://t.co/9b1RBMv46p
We maintain a Customer Account Health Review Agent for a customer on AWS Bedrock.
Before โ 15โ20 sequential tool calls, ~128k tokens, ~40s per review, NL aggregation errors.
After โ parallel sandbox execution, ~13k tokens, ~8s per review, deterministic Python logic.
Programmatic Tool Calling flips the design.
The model writes Python code that orchestrates all the tool calls inside a sandbox.
asyncio.gather() fires all sources in parallel. Python filters and aggregates. Only the final summary returns to the model.
Model is sampled exactly twice.
In a standard agentic workflow, the model calls a tool โ result floods the context window โ model calls the next tool โ repeat.
For a single account review: 15โ20 tool calls. Every raw record passing through context. ~128,000 tokens. ~40 seconds.