CrewAI

Today we launch CrewAI's Agent Control Plane! Between the first car and the first seatbelt there was 63 years, agents need it faster! With a growing number of agents being built and orchestrated, between both code and no code setup, the need for control is growing fast! Federation of the building to accelerate adoption fast tracks this even further. The faster you build the more a need control, because the thing you are governing is the degree of agency you delegated to the system, and that has nothing to do with how it was built. That belief is why, starting today, the Agent Control Plane is generally available on CrewAI AMP! We launched Crew Studio in July with a promise that what business users and engineers build would run on centrally governed infrastructure. The control plane (ACP) is the other half of that sentence. Governance lives in the execution runtime, not in agent code and not only at the network perimeter. If Put guardrails in code and every policy update becomes a software release, if you put them in an external gateway and you get useful routing and API reuse, but an HTTP proxy cannot see inside the agent loop. It cannot cap reasoning iterations, pause a run mid-loop, or enforce business-hours windows. What is live today: 1. Visibility across every deployed automation 2. Cost control, with clear understand of models and agents 3. 13 out-of-the-box policies covering data handling (PII detect-and-mask), allowed models and tools, execution windows and frequency, content filters, and cost limits, plus custom policies that evaluate at runtime without a redeploy. Excited to finally have it out there! Link in the comments

Approved Monday, but still not funded Friday. Credit decisions are starting to be a focus for agentic implementations. Similar to other processes that have been incorporated into agentic system, credit decision making can drive efficiency gains but the true win is scale and impact to the top line. The credit decision itself can be fairly fast but the rest of the week disappears into a missing stip, a title sent to the wrong address, or the funder who could fix it being out on a random Thursday. Meanwhile, the system said "approved, pending." Most of the existing "automated" processes are lacking, status are somewhat inferred, people do all of the work through email, spreadsheets, and some follow-ups can go unnoticed. And the problems increases drastically as the volume goes up. It doesn't need to be that way though! That why I'm excited to invite you to join us on Nov 9 at 1 pm ET, Arindam Choudhury from Capgemini and I are going through one of these deals live, from approval to funding. We’ll look at the actual work and decide what an agent could own, what still needs a person, and what shouldn’t be automated at all. If you have a deal stuck somewhere between "approved" and money in the bank, bring it. Save a seat, link in the comments!

Today we launch CrewAI's Agent Control Plane! Between the first car and the first seatbelt there was 63 years, agents need it faster! With a growing number of agents being built and orchestrated, between both code and no code setup, the need for control is growing fast! Federation of the building to accelerate adoption fast tracks this even further. The faster you build the more a need control, because the thing you are governing is the degree of agency you delegated to the system, and that has nothing to do with how it was built. That belief is why, starting today, the Agent Control Plane is generally available on CrewAI AMP! We launched Crew Studio in July with a promise that what business users and engineers build would run on centrally governed infrastructure. The control plane (ACP) is the other half of that sentence. Governance lives in the execution runtime, not in agent code and not only at the network perimeter. If Put guardrails in code and every policy update becomes a software release, if you put them in an external gateway and you get useful routing and API reuse, but an HTTP proxy cannot see inside the agent loop. It cannot cap reasoning iterations, pause a run mid-loop, or enforce business-hours windows. What is live today: 1. Visibility across every deployed automation 2. Cost control, with clear understand of models and agents 3. 13 out-of-the-box policies covering data handling (PII detect-and-mask), allowed models and tools, execution windows and frequency, content filters, and cost limits, plus custom policies that evaluate at runtime without a redeploy. Excited to finally have it out there! Link in the comments

Developer tools should make any coding agent able to easily know how to use their tool. At CrewAI, we've made it so easy for coding agents like cursor, claude code, codex, opencode, etc... build with the crewai library. Every crew project comes with an agent.md file as well but we've also published to skills.sh to make it easy to access and update as we continue making improvements. If you're product is a developer tool make it so easy for coding agents to interact with your open source tool as well as your platform product if you have any. It starts with coding agents building, but continue with how they should iterate and deploy.

CrewAI is expanding globally, and this one is specially meaningful to me. We are partnering with DXC Technology to help companies across Latin America move agentic AI from experimentation into real business operations. DXC is massive. Around 115,000 people. Operations in more than 70 countries. More importantly, they understand what it takes to deploy technology inside complex enterprises. That combination matters. A lot of companies can build an AI prototype now. The harder part is identifying the right business process, integrating it with the systems already in place, getting it through security and governance, putting it into production, and operating it after launch. That is the gap this partnership is designed to close. As a Brazilian founder building a global company, I have always believed Latin America would be a major part of the agentic shift. We are already seeing companies move beyond isolated pilots. They are building AI-first programs, training their leadership teams and putting agents into real workflows across financial services, retail, manufacturing and consumer goods. This partnership gives us even more local depth to help companies go from the first use case to production systems running at scale. Global expansion cannot mean selling from far away. It means local people, trusted partners, the right language, and teams who understand how companies in each market actually operate. This partnership with DXC is a big step in that direction. Let’s go!! 🔥

I gave my agent a phone number and now it 𝐭𝐞𝐱𝐭𝐬 𝐦𝐞 𝐩𝐫𝐨𝐚𝐜𝐭𝐢𝐯𝐞𝐥𝐲. I see this new agent experience getting closer and closer to all the tools we use everyday. And your agent texting you is one of the experiences that makes personalized agents more accessible than ever before. So I built OpenInstinct CrewAI edition: a personal agent I can text to research a webpage, save a note, set a reminder, or schedule a GitHub briefing all surfaced by texting our agent through its own phone number. One test made it feel real: “In five minutes, brief me on PRs I should review.” I restarted the worker before the deadline. The saved task survived, fetched my GitHub review requests, and delivered a briefing with two PRs to my phone. An ordinary conversation, backed by persistent state and tools that can follow through. The stack behind it: → CrewAI orchestrates with conversational flows. → Composio handles Gmail and GitHub and many others integrations through hosted MCP. → Supabase stores conversation history, saved notes, jobs, and schedules, memories. → Sendblue powers the iMessage channel. → Browserbase handles public-web research and browser use and control for related tasks. Github: https://lnkd.in/gTAMGrWb

I gave my agent a phone number and now it 𝐭𝐞𝐱𝐭𝐬 𝐦𝐞 𝐩𝐫𝐨𝐚𝐜𝐭𝐢𝐯𝐞𝐥𝐲. I see this new agent experience getting closer and closer to all the tools we use everyday. And your agent texting you is one of the experiences that makes personalized agents more accessible than ever before. So I built OpenInstinct CrewAI edition: a personal agent I can text to research a webpage, save a note, set a reminder, or schedule a GitHub briefing all surfaced by texting our agent through its own phone number. One test made it feel real: “In five minutes, brief me on PRs I should review.” I restarted the worker before the deadline. The saved task survived, fetched my GitHub review requests, and delivered a briefing with two PRs to my phone. An ordinary conversation, backed by persistent state and tools that can follow through. The stack behind it: → CrewAI orchestrates with conversational flows. → Composio handles Gmail and GitHub and many others integrations through hosted MCP. → Supabase stores conversation history, saved notes, jobs, and schedules, memories. → Sendblue powers the iMessage channel. → Browserbase handles public-web research and browser use and control for related tasks. Github: https://lnkd.in/gTAMGrWb

We're partnering with DXC Technology to bring agentic systems into production across Latin America. DXC is a global IT services company operating in 70+ countries, with 60 years in Brazil and 7,000 people across the region. They bring consulting and implementation. CrewAI brings the platform to build, orchestrate, and govern agents. Big props to our own Pedro Moura for helping to make this happen, and to Ricardo Ferreira, and the DXC team. Excited for what's next. #AgenticAI #CrewAI #DXC