5 Claude Code Updates That Just Changed What 'AI in Your Business' Actually Means

I just read Lenny Rachitsky's recap of Anthropic's Code with Claude event, and a few of these updates are the kind of thing that quietly change how a one-person business operates.

Anthropic announced five new Claude Code features. The headline ones — Routines, multi-agent orchestration, and a memory system called Dreams — are basically about letting AI do work in the background while you go live your life.

Let me walk through what stood out to me, and where I think it actually moves the needle for someone running a marketing agency or a SaaS side project.

1. Claude Code Routines

This is the one I'm most excited about. Rachitsky describes Routines as a way to automate recurring workflows on a schedule or via webhooks. So instead of opening Claude every morning to run the same five tasks, you set them up once, and they fire on their own.

For me, this is the difference between "AI that helps me" and "AI that does jobs." My agency has a daily content workflow that I currently start manually every morning. If that becomes a Routine that runs at 6am while my coffee is brewing, that's an hour back every day.

2. Outcomes/Rubric-Based Agent Grading

You define what success looks like, and the agent gets evaluated against that. Rachitsky's framing: it's the difference between "did the agent run" and "did the agent do the job."

This is the unsexy update that actually matters. Most people running AI in their business don't have any feedback loop on whether their prompts are working. A rubric layer means you can finally answer "is this agent making me money or wasting my time."

3. Multi-Agent Orchestration

Deploy specialized agents that work together — different roles, different tools, one orchestrator to manage them.

I've been doing a janky version of this for months. I'll have one Claude conversation that researches a topic, another that writes the draft, another that edits. It works, but it's a babysitting job. Built-in orchestration means I can spin up a small "team" of agents and let them hand work off without me playing project manager.

4. Dreams (Long-Term Memory)

A memory system that lets agents stay consistent over time. Rachitsky calls this "a bigger deal than it sounds."

I agree. The biggest weakness of every AI assistant I've used is that it forgets me every conversation. I have to re-explain my brand, my voice, my clients, my style — every time. A real memory layer means the assistant gets better the longer I work with it, instead of resetting to zero.

For business owners, this is what "my AI" actually means. Not a chatbot. An assistant that remembers I am.

5. Increased Usage Limits

More capacity. Less hitting the wall mid-task.

This is the boring update that's secretly huge. The thing I tell every business owner trying to use Claude or any LLM in their workflow: the moment you hit a limit, your usage drops permanently. Bigger ceilings means more room to actually build the workflows you've been wanting to build.

What I'd add from my own experience

The pattern I keep noticing in my agency clients: most of them are still in the "chatbot" phase. They use AI like a search engine. They don't have routines, they don't have evaluation, they don't have memory. So they're getting maybe 10% of what's possible.

This update is what makes the next layer reachable for non-engineers. Routines don't require code. Multi-agent orchestration is becoming a config conversation, not a Python project.

If you're a small business owner reading this and thinking it's not for you — it is. The barrier to AI being a real teammate just dropped again.

The Bottom Line

The Code with Claude announcements aren't just developer news. They're the moment AI in business shifts from "I asked it a question" to "it does jobs while I sleep." Routines and Dreams are the two updates I'd implement first. The rest of it follows once those two are running.