Howard Crow: There's this classic phrase that people use where it's like the USB for APIs. It's actually a pretty good analogy, to be totally honest, because having a very standardized way of just plugging into an agent and building an agent on top of it has just made it so much easier to actually build a little app or whatever. I mean, we're seeing this, right? Like, apps are now a dime a dozen because you can make them in 10 minutes. And getting Planner sort of plugged into that and having that be a task fabric for the apps that people wanna build, I think is really important.

Tom Arbuthnot: Hi, and welcome back to the Teams Insider Podcast. This week we are talking Microsoft Planner. We're lucky to have Howard on the podcast. He took us through some of his history at Microsoft, how Planner came to be, the various integrations and where it is today, and also where it's going with its new agent, its new AI capabilities, and API surface as well. Really interesting conversation about Planner and where it's going. Thanks to Howard for jumping on the podcast. And many thanks to Luware, who are the sponsor of this podcast. Really appreciate all their support of the community. Hope you enjoy the show. Hey, everybody. Welcome back to the show. Really exciting to have Howard on the podcast. A different topic for us. We're gonna talk about Microsoft Planner, which I learned is kind of tied into the Microsoft Teams org, so we'll talk about that. Howard had a great session at M365 Conf, and there's a lot of announcements, including MCP, including new API access, but I don't wanna steal all the thunder. Howard, welcome to the show.

Howard Crow: Thank you, Tom. It's great to be here, very exciting. And yeah, as you say, there's a big Teams tie-in between Planner and Teams over the years.

Tom Arbuthnot: Yeah. So give us a little bit of history of yourself and your role and what you're doing today.

Howard Crow: Yeah. So I've been at Microsoft for a very long time working in the tech industry. I started in 1996, I think. I was like a wee intern coming out of school at the time, and so almost 30 years at the company at this point. My first job I did a lot of strategy work. I worked with a guy named Jeff Raikes, who was one of the early sort of SLT members, and we did a lot of business strategy, product investment strategy and things like that. And one of the things that we saw an opportunity in was this sort of collab space, and that was the genesis of a lot of the investments in SharePoint. And I joined the SharePoint team because of that, 'cause I'm like, "Oh, this seems really cool. I'm gonna go do this." And I was one of the first folks on the SharePoint team before it was called SharePoint. I actually wrote the code for Web Parts. And so I was, you know, dabbling around in the early days, writing a bunch of sort of JavaScript HTML code, and ended up joining the team with Jeff, Dee Burd, and Bo.

Tom Arbuthnot: Oh, interesting. Yeah, we had Jeff on the podcast a couple of months ago talking through the SharePoint 25th anniversary and the history and his journey with it. I mean, it's an amazing product journey.

Howard Crow: Yeah. We actually had a founder's dinner the other day for the 25th anniversary, and a bunch of us got together, went out to dinner, and it was just great to see everybody again. And so I worked on SharePoint for a really long time, like almost 12 years. I owned the front end. I did a lot of the back end stuff, like storage and platform things. Really amazing journey. And then after that, I did a whole bunch of different stuff. Some of those projects actually got canceled, which is, by the way, one of the great learning experiences in software. Sort of working on something and then having it canceled is a deep learning experience.

Tom Arbuthnot: Yeah. You poured your heart and soul into it, and it turns out it wasn't quite what you thought. I think the interesting thing at Microsoft is things that could be deemed successful in other companies might get canceled in Microsoft 'cause relative success is important.

Howard Crow: Yeah. No, it's absolutely true. Like, there are businesses at Microsoft that are, you know, a billion dollars, and it's like a tiny little thing inside Microsoft, but it would be a unicorn anywhere else. And so those things definitely happen. And then Jeff actually put me on the project team, and there's a real business aspect to the project team. It has its own revenue and all that kind of stuff. And he thought that I would do well in that environment, loved doing that. And one of the things that we discovered early on when we were talking to users was there was this real democratization happening in the project management space and work management space at the time. We saw that trend pretty early, and this was the genesis of Planner, was, "Hey, how can we make something that people can use to manage their work that is more accessible to everybody?" And things like Agile were starting to show up where people were coming in, updating their own tasks and things like that. And so that was the start of Planner.

Tom Arbuthnot: Just for those who don't know Project, there was Project Desktop, there was Project Server. It was like a bit of a beast. Like, you'd be a specialist. If you were using that, you were a full-time Project person, and you knew how to do all the dependencies and that kind of thing.

Howard Crow: Absolutely. And in some ways I used to call it the Photoshop of project management where, boy, it was like there were so many knobs and dials and features. It was an amazing product. Still there. People love it to this day. But Planner was really this evolution of, like, more people are gonna manage their work sort of more systemically. And so we started Planner early on. It was a great experience, super fun. We grew it up from literally nothing. So you might hear a theme there with me, which is a couple of moments of growing something from zero to 60 inside the company, doing it within the sort of the womb of Microsoft, which is both great and hard at the same time. Because little things can get lost in the giganticness of Microsoft. But it really took off. And to bring it back to the Teams thing a little bit, one of the main places that people started using Planner was inside Teams. Our sort of major usage is people pin the app, the Planner app, to the left-hand nav of Teams, or they make plans inside channels. This is where a lot of our usage comes from. And so this is why we're actually inside the Teams org now. We work with the Teams guys really closely. We do things like integrate with meetings, channels, and chats, and all sorts of different things inside Teams because the collaboration of people is really the Teams environment.

Tom Arbuthnot: And that makes a lot of sense, right? Teams is the place you come to collaborate, you're working together on a team that is often revolving around a project, so you would need a project tool around that, and having it be native in Teams makes a lot of sense for a lot of use cases.

Howard Crow: That's absolutely right. And you know, how many meetings do people have that are about the status of a project? Not just the channel where the conversations and the documents and all those kinds of things sort of revolve and enhance the work tracking, but just the meetings themselves where you're going over the work and talking about the status of things. And so there's huge additional opportunity to make those experiences more seamless for people, where you don't have to sort of cobble together the experience. Like, I'm doing this work over here and I'm doing it in a meeting and I'm having a conversation about the task, so how do those things integrate? Just making that whole thing more smooth for people. There's a long set of work we can do there to make that better for people, and we hear this from customers all the time.

Tom Arbuthnot: And in the spirit of the genesis of Planner of being accessible to more people, I feel like AI is really exciting, right? 'Cause there's a lot of teams that are gonna be working with it who don't have that full-time project manager specialist building the planner, chasing them up. I'm desperate to get into the AI agent conversation. Before we do, talk us through the different offerings. There's been Planner Basic, Planner Premium, there's been To Do. There's been a few different things there. What's the genesis, and where are we now with Planner as a service?

Howard Crow: Yeah, it's a really good question. I think over the years, Microsoft has kinda spun up quite a few task management, work management tools because it's a pretty horizontal thing. Everybody needs them, right? Everybody works on projects. Everybody has tasks they need to do. And as you can imagine, we've created a few things like To Do. And by the way, To Do is also on my team, so we're all in the same family basically. And so we've integrated all the families of sort of work and task management together in the same org, and what we're doing is effectively combining them all into one experience that will live under the brand of Planner. For example, we announced just recently that we've got a whole new My Task experience that's on Planner that is modeled after To Do. And so we'll be welcoming To Do users into Planner over the next couple of years just to get them into a more single pane of glass place where they can manage both their tasks and their plans in one place. The other thing is that we've got this Planner Premium offering that we'll be integrating into core Planner as well and unifying all the architectures behind everything. And the great thing about that is that it'll just make it more seamless and easy to understand the difference between a basic plan and a premium plan. And so for folks who use Planner, they know we have this kind of confusing division between these two things. And we're gonna fix that. I call it a theme in those two things. So Planner will be a unified environment for plans.

Tom Arbuthnot: That's good to understand. And Planner Premium, am I right that was kind of Project on the web? Is that where some of that DNA came from?

Howard Crow: Right. Exactly. It was originally called Project for the Web, and we were like, "Do we really need all these different architectures and brands?" And it feels kinda confusing. And I think people are welcoming this unification a lot. We're getting a lot of really good feedback that this will make it easier for folks to understand the way the system works. The interesting thing is that Project for the Web, and in the future Planner, is actually based on the old scheduling engine from Project. And so all of the logic and wonderful history of managing projects is actually baked into Planner.

Tom Arbuthnot: Yeah, the power of that to do things in the future.

Howard Crow: Yeah. The scheduling engine of Project, which manages your dependencies and gives you a critical path, like, here's actually the thing that's slowing you down the most, is kind of the gold standard. Literally people who go to get degrees in project management are trained on it. And so bringing that forward and making it a part of this unified system is really important to both us and our customers.

Tom Arbuthnot: Awesome. So let's talk AI. When did that first come... I mean, obviously, it came to Microsoft and every project team was looking at it. But when it came to you with a Planner hat on, what were you thinking? I know the new Planner agent has got a really good reception. Take us through that journey to the Planner agent and what else you're doing with AI.

Howard Crow: Yeah. You know, this is one of those "Hey, AI's gonna change everything" moments, right? I think we're all saying the same thing at this point. And especially in the last three or four months with all the evolution in the models and the harnesses that's gone on, it's really just amazing to watch the capabilities change. We actually got into AI really early, mainly because the idea that AI needs to plan out a path about what work it's gonna do, make a set of tasks and subtasks, and then follow a relatively thought-out path so it doesn't wander around and hallucinate is kind of endemic to what great AI is. And we're seeing this in all the new tools. This idea of, as an AI, I'm gonna make a plan, is really there. We actually saw this pretty early, and we shipped the first version of the Planner agent. We actually called it the Project Manager Agent. It got renamed 'cause we've done a lot of work and rebuilt it over the years. And that had underneath it something we called the Multi-Agent Runtime. For short, we called it MARS, the Multi-Agent Runtime Service. This was actually something that we got from Microsoft Research, and it was based on AutoGLM, which was an early harness that let multiple agents talk to each other and negotiate and pass off work. So Planner is literally the first production multi-agent runtime harness that ever got shipped in the world. I'm kinda proud about that. And we use this in a bunch of different ways. For example, if you're in a Teams meeting today and you ask... you can say "@facilitator, go make me a document that reiterates everything that happened in this meeting." That actually uses this multi-agent runtime service in the back. We actually make a task inside Planner. We hand that to the task runtime. We gather up the transcript. We hand that off to the multi-agents, and they go through a little loop. There's one that goes and writes the doc. There's an editor that edits the document. They run in a little loop, and they write the document for you, and then it will come back inside the meeting, right there, and you'll have a document that documents everything that happened in the meeting. That was the first sort of foray into agents and planning. And what's interesting is, boy, you're seeing this everywhere. If you're in Anthropic, you go up and say "Plan this thing," and you can then have multiple agents go, and it'll make a plan with a bunch of tasks.

Tom Arbuthnot: It's cool. Yeah, it's part of the reasoning logic, isn't it? There's also a finite context window, so actually it makes a lot of sense for AI to be like, "Here's the chunks. I'll go attack this chunk." And then obviously later on, we got multi-agent, so actually now I can spawn agents off to do all the subtasks. There's very good logic. It's funny that the way we organize work as humans turned out to be the same model as the way you would organize work for agents.

Howard Crow: Yeah. And this is something that I think is just showing up more and more. And when we did this two years ago, it was almost before its time in some ways, right? The models weren't totally ready. And as we've updated the models underneath Planner, we're just seeing it light up and get better and better. Because the models themselves, like this mixture of expert models and things like that, are literally designed around this pattern, and so it's just getting better and better every day. And so what we've done now is we've taken all of that and bundled it up into something called the Planner agent. We shipped this, I don't know, two months ago maybe, into our Frontier, to all of our Frontier tenants, so it's available to everybody who has Frontier in Microsoft 365. Stay tuned for our GA announcement. And that has been a huge success. You can actually now go into Copilot itself and you can say @mention Planner agent, go make me a plan. And we've done a lot of work to create a new UI around the plans that we make. We have these cards that we create that give you a very rich view of the goals and the tasks, and all of the different components of the plan inside Copilot. It's not just text, it's actually a very rich user experience. And the customer acceptance has been off the charts. Way more than we expected. People are loving being able to call up the Planner agent inside Copilot itself. And it uses Work IQ in the background. So when it makes the plan, it's actually looking at all of your content in OneDrive and SharePoint. And so you get a very personalized plan, not just a random plan with tasks that have nothing to do with your life. And so it's been really well-accepted. Stay tuned for the Vienna support.

Tom Arbuthnot: Yeah. That's an interesting unique thing to Microsoft, this Work IQ context layer where, if you're asking it to make a project and it has access to your data, it can leverage that. So we talked about the project in the meetings. It's got the meeting transcript. We wrote some emails about it. I've got some drafting in my OneDrive. It's exciting that we have that level of contextual awareness. It's almost like asking an in-house person to do it versus an outsourced person. An outsourced person can come up with a generic plan, but an in-house person has the context of, okay, here's what's been going on.

Howard Crow: Yeah, and this is the real difference between, I think, a lot of the work that we're doing inside Microsoft and some of the external tools that are a little bit more trained on the internet, right? That's great, but my company is very specific. Like, I'm a defense contractor, or I'm in the pharmaceutical business, or I'm in manufacturing something specialized. Those things just have a whole bunch of context that needs to be brought to the way that you build the plan that is different than the plan that's built on the internet. Now, for coding, there's so much content on the internet for coding, with Substack and all this kind of stuff, where the models have been trained on that. And so coding actually is this example of, hey, looking at the external world and the generic kind of content that's out there is fine, and it's proving to be incredibly successful because of that. Coding has been this one shining example of, yeah, if you train it on the internet, it'll be okay. But the way that a company works and its IT...

Tom Arbuthnot: Not many enterprise project plans get shared uncensored on the internet, it turns out.

Howard Crow: That's right, they can't be trained on. And so, funny story, a lot of the early training work was done on Enron data. I don't know if you know about this, but...

Tom Arbuthnot: Yeah, I've heard this actually, because it was all available.

Howard Crow: So if you want a project plan based on what Enron did, then you can get that on external data. But I don't know if that's a good idea.

Tom Arbuthnot: No, it doesn't sound like a great idea. And you've been doing a lot of work. I'm really excited about the MCP. You've been doing work with the API. We at Empowering.Cloud build a product where we've used the Planner integration. We've now done some stuff with Dataverse. I'm really excited about what you're doing with the APIs and the MCP, 'cause we wanted to be able to write custom fields to Planner, and I believe that's coming. Is that right?

Howard Crow: Yeah. Boy, we're all excited about the MCP, I have to say. It sort of unlocks a lot of things. I'll talk about custom fields in a minute, but I'll tell you about a set of work that we're doing right now, which is we released Cowork a while back. By a while back, I mean, like, a month ago.

Tom Arbuthnot: Yeah, yeah, ages ago.

Howard Crow: It's ages, right? It's like an ancient, crusty piece of software at this point in AI ages. The velocity's amazing. But yeah, so we released Cowork, and we're doing a bunch of work with those folks to get our MCP connected into that and make sure that we do the evals on it, and we're getting really good output. And we're already seeing a bunch of folks, like our MVPs and stuff, playing around with that combination, and it's super powerful. And so this idea of having that sort of integration available is really good. And as you mentioned, as we're releasing new capabilities into Planner, like the ability to have custom columns, custom fields, things like that, we will expose those in the MCP. That will just make the two even more flexible. The funny thing is, and this is where the MCP sort of showed up, the first thing that happened when we did the MCP is, if you're going in and using one of these coding tools, whether it be Codex or Claude or GitHub Copilot, the first thing every engineer does is make a Kanban board for what their agents are doing. And it's funny because that's the first thing we tested. We're like, "Can I make a Kanban board in all these tools, connected to the MCP?" Like, basically have Planner just be the Kanban board. And yeah, it works great. And this, I think, gets back to the role Planner has in the age of AI. We believe that the world is gonna get to a state where we are managing teams of people and agents, and your team is gonna be made up of, like, three humans and 12 agents. How do you track all that and manage it, and see what's going on? Today you can go into Planner, and this already exists. We did this using our agent runtime service. You can assign any task to an agent and go and execute that task, just like you would a human. And so being able to coordinate that work where there's a lot of agent tasks and there's people off doing stuff, how do I know where everything is in this process? It's almost gonna get harder rather than easier because so much of it is agent-run. So how do you know where everything is?

Tom Arbuthnot: Well, it's interesting. Project is a great service provider, right? And you said exactly the thing, which is the first thing developers do who are probably in GitHub CLI or Claude Code or something, they spin something up, and they might build some crazy markdown thing. They might use Jira, they might use ADO. They're developers. But once again, potentially Planner is the more accessible version of that, which is like I do something in Cowork, but now I can, as you said, have a Kanban board, have agents, and two weeks' time when I've kind of forgotten what Cowork did for me, I've got this SaaS-accessible interface that's robust and built to be like, oh yeah, if this agent has it or it stops here. That's really interesting.

Howard Crow: Yeah, that's right. And then the output can be attached to the task as a link to the document or the PowerPoint or whatever it was that you actually built in that process. And so you've got, as you say, a history or a memory of what you've been doing over time. And this really, you know... The one-on-one scenario is really interesting, and we love it, but the sort of what we call multiplayer experience is where it really shines a lot. Both of those are really great, and this is one of the interesting things about Planner. It has both what we call single-player and multiplayer. Single-player is me and my agent just kind of chatting together. Multiplayer is I've got a couple of teammates, a couple of agents, which is super leveraged, where we're all sort of working together. And this is where I think today we see a lot of single-player stuff, right? Like, you're in Claude CLI, and you make a board, and it's me and my two or three agents, and we're partying. It just unlocks itself in an organization when that gets exponentially bigger, and this is where we haven't quite gotten here yet in industry. But this is the next layer that's gonna get unlocked, which is my teams, these multiplayer scenarios with teams of agents and people, is gonna be the big change that happens in the next probably six, seven months.

Tom Arbuthnot: Yeah, it's interesting to look at development as a prediction of where knowledge work is gonna go. And it's like development was, "Oh, this is not serious." And now, okay, it's all right for autocomplete, and actually it's quite powerful. And then it's like, oh, I'm spinning off multiple agents doing multiple things, and Cowork, I feel like, as you said, is a... Like, suddenly that's a huge new massive amount of power we have in knowledge work. We go, "Oh, it can actually send emails for me. It can schedule tasks. It can build PowerPoints." Like, how many times have we had a go at building PowerPoints? The Cowork feels like the one that's like, oh okay, it's starting to really get there now. And so if you look at where dev is now, I don't know what timeline you put on it, but you can kind of see a polished version of that coming to knowledge work.

Howard Crow: Yeah, absolutely. This is the big thing. And developers, because they have high technical acumen, can sort of deal with working in a CLI and typing things in, and they understand how multiple agents work and all this kind of stuff. They can actually easily visualize in their head what's going on even though they're just typing text into a prompt. For knowledge workers, they need a much better experience for that kind of thing, and I think things like the Kanban board visualizations and being able to see my agents visually, like, "Oh, here's this agent. This is what it's doing. Is it done? Is it not? Does it need help?" Right? Like, that's another thing that happens today. I do a lot of work using CLIs and stuff, and you end up in this state where you have eight windows open and you're literally waiting for something to happen, right? And you spin up enough windows so that something is always interrupting so you can then help it and move it on to the next thing. That process is kind of inefficient. Like, you're sitting there waiting for it to need help. Things like having a visualization about when help is needed so I can step away, come back, do something else while it's reasoning, and get a notification that something has changed. All of this is what's needed to make it knowledge worker friendly. And it's just fun to start building that out and getting that ready for people. It's gonna be an exciting six months probably before we get that in a good place for knowledge workers.

Tom Arbuthnot: I'm jotting down a podcast catch-up in about six months' time then, Howard, so we'll see if the prediction is there. I just wanna talk lastly about the API, 'cause I feel like AI has given APIs a huge shot in the arm in the industry, and definitely inside of Microsoft as well because we need this interrupt. It sounds like you guys have been doing work on the API coverage as well.

Howard Crow: Yeah, I think it's a good observation. I think a couple of things have changed in that space. One, the MCP thing is, you know, there's this classic phrase that people use where it's like the USB for APIs. It's actually a pretty good analogy, to be totally honest, because having a very standardized way of just plugging into an agent and building an agent on top of it has just made it so much easier to actually build a little app or whatever. I mean, we're seeing this, right? Apps are now a dime a dozen because you can make them in 10 minutes. And getting Planner sort of plugged into that and having that be a task fabric for the apps that people wanna build, I think is really important. And so integration with things like Copilot Studio, where when you're in Copilot Studio you can just whip up an agent and have it store its tasks inside Planner, is just a core part of the Work IQ surface that people want. And obviously Build is coming up. Wow, it's pretty close. It's like a couple of weeks.

Tom Arbuthnot: Yeah, yeah, yeah. Weeks away now.

Howard Crow: And that is a huge event for us with developers, and you're just gonna see our MCP and Work IQ story shine there a lot. I think it's very unique in the industry because of the content that we have in our systems. The petabytes and petabytes of data that are accessible, plus our API surface is just amazing. All of the different components, the Lego blocks, that we offer for people building agents is just awesome. So I think Build this year will be a lot of fun because of that. A uniquely fun Build is my guess.

Tom Arbuthnot: Interesting. Yeah, I'm looking forward to covering that. So we'll be keeping a close eye on it. Howard, thanks so much for jumping on. Awesome conversation. Really appreciate your time, and yeah, it's an exciting time for Planner. We'll definitely have to check in again. I'm on board with you with the prediction and hypothesis that the planning surface becomes much more important as we go to use agents and unlock knowledge workers managing multiple agents, so I'm excited to see how that develops.

Howard Crow: Well, thank you, Tom, for having me. It's been a lot of fun to chat. It was great to see you at the M365 Conference, and I'm happy to check in in a bit when we're a little bit farther along with the knowledge worker thing and give you guys some more peeks into the future.

Tom Arbuthnot: Awesome. Thanks so much.