Michael Cohen

Notion 🤝 Claude Managed Agents Learn more next week 👀

today at code w/ claude, we shipped major upgrades to Managed Agents to make your agentic systems even more powerful dreaming: agents learn from past sessions outcomes: rubric + grader, iterates till the work is done multiagent orchestration: work split across specialists

I'm so excited about this batch of releases. When we launched CMA in April it was super painful for us to not include these on day 1, but we had to let them bake a little bit longer. I'll do a separate writeup on outcomes, webhooks, and dreaming at some other point BUT what I'm really excited for is our multiagent feature. we knew from the get-go that the "one shot subagent" approach was good, but not nearly as powerful as long lived multiagent threads could be. In our testing, we found it to be a significant improvement for the other agents in a session to be long lived. It gives the main "coordinator Claude" the ability to steer and follow up with it's partner agents. It also helps with recalling much older pieces of context those agents may have already discovered. Claude's gotten REALLY good at acting as a delegator, which means the main agent you're talking to directly can keep its attention on high level important details without a flurry of tool calls / older context to worry about. the future is decisively multiagent, but it's not trivial to implement. the vast majority of developers are going to benefit from letting us worry about "getting it right". go try it out and see for yourself and give me some feedback if there's anything we can improve.

Founder friends ask me: why Claude Managed Agents vs. X, Y, Z? Beyond feature fit, every integration has a cost. Picking a product is picking a long-term partner. You're betting on who listens, who supports you, who ships faster than your roadmap grows. A few weeks after we announced CMA, today at Code with Claude we launched multi-agent, outcomes, webhooks, and Dreaming. So much more coming. platform.claude.com/docs/en/manage…

@banksathapon just followed you back, DM away!

@_hi_mc Looks like your DM is closed 😅 Any other way to reach you?

we're taking Claude Managed Agents international. come join the wave

@banksathapon lets do it! I'll be around all day, shoot me a DM or just look for the dude w/ the mustache

@_hi_mc We are building HumanZero using Claude Managed Agent to enable zero-human companies. Will be at Code w/ Claude SF. would love to connect for a quick chat and share what we’ve learned if you’re around!

version controlled, ready out of the box, and intended for you to use in your own applications. clone it, remix it, take what works for you, and ditch the rest. we want to make it easy and delightful for you to focus on your own products. this is the power of Claude Managed Agents.

change my handle no one freak out everyone STAY CALM I'm trying to be cool and fun

don't stay too attached to a certain way of using AI tools. things are changing so quickly and we're all still figuring out how to use LLMs together. be nimble and embrace the fact that what "good" looked like in Jan'26 might be inefficient today. learn from your peers, especially those who are pushing the boundary of model capabilities. they might run into clunky behavior and rough edges right now, but when the next generation of models comes around they'll be running laps around everyone else. remain flexible.

@kevinwhinnery this is great. giving me lots of ideas. thank you!

@mc_anthropic I would love to hear more about patterns you expect to emerge and best practices for clients of the API! How should I version control config for my agents / prompts? How should I approach staging and production deployments of agents? Do I manage memory per agent, per end user?

i'm gonna be hosting a workshop at the Code with Claude conferences over the next few weeks! first up is SF, see ya'll soon. i'll focus on building a production-grade agent using Claude Managed Agents, showing how easily and quickly you can get started. going to go into technical details, show off how to build in minutes with the new APIs, answering questions, etc. what would you like me to go deep on?

I'm really happy to share that Medkit won 1st place 🥇 at the global Built with Opus 4.7 hackathon by @claudeai, @claudedevs, and @cerebral_valley. Medkit is a voice-first AI clinic for medical students and junior doctors. You consult with AI patients in real time — take the history, order labs, read imaging, diagnose, prescribe — and get a structured, cited debrief on every case. Let's see where the journey goes, but I'm truly excited to be part of shaping the future of medical education! 🎥 Full demo: youtube.com/watch?v=6bN6hn… 🔗 Try it: medkit-app.vercel.app 🖼️ Project gallery: cerebralvalley.ai/e/built-with-4… Thank you to @claudeai, @claudedevs, and @cerebral_valley for the week, and to every builder I met along the way. #BuiltWithClaude #ClaudeCode #BuiltWithOpus

Jess is the one of the best product leaders I've ever worked with. She's constantly pushing the boundary of her role and revisiting what great looks like in the era of AI. Read this excellent blogpost.

Another Claude Code hackathon comes to an end. Thank you to everyone who spent a week building with Opus 4.7, and to @cerebral_valley for co-hosting. Introducing the winners:

@MylesLX shoutout to @felixfbecker and the rest of the DevX team they really cooked on that UI

@mc_anthropic Sub agent traces looks like the browser dev tools network tab, but instead of loading clients you are loading agent traces, well done!

our new Memory suite of APIs works out of the box in Claude Managed Agents. they're also portable by design and can be used in any other AI product. the building blocks are coming together. mix and match what you need. go to production in minutes. scale your usage along with your business. then go touch grass while Claude does the rest.

@GilCaspi1 @Anthropic_ai hope my talk was helpful! let us know if you have any questions or have any feedback :)

As part of the "Built with Opus 4.7" Hackathon by @Anthropic_AI, just wrapped an amazing workshop with @mc_anthropic about managed agents.

@dsp_ ...I like it

I have this old habit of swearing and people rightly fully complain that I shouldn’t do that during my talks. Sorry everyone. I’ll stop doing it.

The Claude Code hackathon is back for Opus 4.7. Join builders from around the world for a week with the Claude Code team in the room, with a prize pool of $100K in API credits. Apply by Sunday: cerebralvalley.ai/e/built-with-4…