[On-demand] Product Management Webinar: Agent Workflows

[00:00:00] Julie Hammers: Hello. I see some folks popping in here. Welcome, welcome

Happy Tuesday on a formerly rainy, but not so rainy evening here in London. I hope everyone is well. We’ll get started here in just a few. We’ll give some time for folks to join us. In the meantime maybe let’s talk about whether folks are sad to see summer go or if they’re excited for autumn or fall, as we say in the States.

Today. That’s okay Just wanna make sure I’ve got my

Comments on here. They always change the control position every time I [00:01:00] use Zoom, so it’s fun like that

Okay. We’ll give it just one more minute. Maybe while we’re waiting, another question is what are people most excited to learn about or automate with MCP?

Best practices. Yeah. Yeah. That’s kind of what I’m hoping to get away from B- Steve, is the ad hoc with everything. Okay. All right. We’ve got a good little crowd here, so I’m gonna start, go ahead and kick off here, ’cause we’ve got, I think, a fair amount to get through and talk about. So welcome everyone.

I am Julie from ProdPad, and today we are going to talk about MCP and basically using MCP as a connection layer with our whole product stack. And talking about why [00:02:00] thinking about it more holistically and at a organizational level is potentially the better way to go than thinking about it as an individual PM or as just a civil person kind of trying to you know, use AI agents and automate our work.

So let’s get started. So today we’re gonna be focused on three different areas. The first is maybe a little bit of a review if you’re not familiar with product operating models. But what I wanna walk us through is all the parts of the product discovery and delivery flow that most companies or organizations have, and how you might apply that to your own company or if you already do just to kind of see how that maps.

And then we’re going to basically do a mapping exercise and talk about, okay, so where does that work and where does that data live? And then where we’ll talk about, okay, like once we’ve done that, then we’re gonna plug and play into MCP. So that’s the [00:03:00] general flow for today’s chat.

So if you are not familiar with ProdPad we are the original product management platform that connects your product ideas into outcomes brings everything together with roadmaps, ideas, and feedback management. Connects everything from your vision and strategy to prioritizing ideas to experiments.

We have loads of integrations and we are free to try, of course. And we have a sandbox where you can get in and take a look at the product yourself if you are so inclined. So that is our spiel done. But yeah, let’s talk about the product decision stack. So I can’t claim credit for this particular kind of graphic or this orientation.

There’s a fellow named Martin Eriksson who is well-known in product circles for founding Product Tank. But he has a new book out called The Decision Stack. And I like [00:04:00] this framing for this particular exercise, and I’ll tell you why here in a second. But what this is showing us is the top part of our product kind of operating model.

What you have at the top is your vision where we’re going our strategy, how we’ll get there. And then generally you either have goals or you have objectives, and that is kind of what matters right now and what… how you’re going to measure success as a key result.

And underneath of that is our opportunity layer, right? And opportunity is called different things in different places. We refer to them as initiatives and within ProdPad, but it maps the same. So the opportunity is not a feature. It is a space or area. What is the strategic area that you want to focus our problem-solving or solutioning in?

And that last piece is interesting, those principles. So that’s kind of our guardrails, right? [00:05:00] That’s why we do what we do. How do we decide where we will and won’t go and what guides you as a company. And so that principle, and you can learn more about that in Martin’s book, but the principles kind of ladder up to strategy, right?

And so it’s important to kind of have these pieces in place before you start mapping things out across everything. So yeah, that’s kind of the first… And I’ll show you where how all these pieces are gonna kind of come together with MCP here in a second, so just bear with me as we’re getting through this if you’ve heard it before.

But the next piece is what I’m referring to is a product operating stack, right? So if you go back to this decision stack, right? So this is a simplified version. Most companies will have multiple objectives, right? And they’re gonna have loads of different opportunities or initiatives that they’re approaching.

But if you take one of these opportunities out, what does that look like going down? So what AI has [00:06:00] done is allowed us to very quickly iterate and compress where we had a double diamond where we had our discovery and we had whatever, delivery. And even before that, we had more of a waterfall thing.

So this incorporates ideas from people like Teresa Torres in terms of opportunity solution trees where you are quickly testing ideas out, and that’s what working with AI allows us to do. And even just without AI, I’ve seen this compression was already happening, right?

So the key here is that you have your opportunity space. You have identified kind of solution or feature that you wanna look at, and then you have experiments that you wanna try that are related to that solution. And then you have a very quick iteration through definition, ideation, that includes design and prototyping and deliver, right?

So that obviously includes you know, setting up your dev tools managing the, the GTM process there, and then measurement. And then in this bottom [00:07:00] layer this is kind of like the principles layer on the other graph. But insights is the feedback that you get through the whole process.

So this actually informs this cycle, right? And then it also informs the other graft as well. So you take these insights back to this graph as well, or stack, and then it also feeds into your outcomes. So the outcomes at the bottom are based off of your measurements and your insights, and that results in your outcomes.

So that’s all to say is that these two stacks kind of go together, just in terms of… you can think about this as your top of funnel, if you will, for this is like your stakeholder view, right? And then this is getting down to the team level. So why have I… I’ve got a raised hand here I wanna take.

Thank you for raising your hand.

If you can type your question in, whoever raised their hand, that would be great. I’m happy to answer it real quick before I go on to the next piece.

Unless that was [00:08:00] a… Clicked the wrong button. Awesome. Okay, great. Just wanna make sure, ’cause I know that was a lot to digest. So why did I show all of that? So I’m gonna show you. So we’re gonna take all of that info, and we’re gonna plug it into how we work, right? Okay, coming back to our graph.

Next thing we’re gonna do is mapping our tools. So bear with me for just a second Feel like I’m the queen of swapping windows on these webinars. We have got

Too many windows open. Okay. So hopefully can, everyone can see my lovely Figma board here. So I’ll be sharing this out as a template that folks can use just as your own mapping exercise. So now we’ve got these mapped up next to each other, right? So we’ve got our opportunities and then our full operating stack here.

The next piece is Now take a look at these graphs or maybe just think [00:09:00] about the way that this kind of looks in your own organization. Then think about for every piece of the stack, what tools are you using? Where does that data live right now, right? And you know, what’s involved with that, right?

And the good thing about MCP is that it’s really taken off in the last couple months. Every single one of these apps on this kind of screen here, they all have MCPs available. So we have an opportunity available to us where previously you know, we’d used manual like Zapier flows sometimes.

But there was nothing really that could help maybe connect or send notifications, but it couldn’t really do any definitive thinking or reporting around what was happening in this whole stack. So think about what you wanna do. For us we, I will say on the top of the stack, so this may vary if you use a different product tool this part of the stack might [00:10:00] be in a different tool.

So but we have all the kind of tools for product leaders to have all of, you know, this part of the stack all in ProdPad. But on the other side so we live in the idea level, like initiative and idea level up through ProdPad, and then our team uses Figma and Claude Code to design and prototype Linear and Trello for development, Mixpanel, Segment.

You know, you wanna look at your sales tools, and especially for this insights piece, that’s really important. Think about all the sources that you might wanna connect up into your product tool. So for us we have all of our sales recordings in Fathom. And we use Help Scout as our support tool.

Our CRM’s HubSpot. So whatever those are for you, you can just map these out of And this should be org-wide too, or at least team-wide to get the most benefit out of this. Because what we wanna avoid is kind of individual markdown scripts on your personal instance [00:11:00] of Claude or ChatGPT or whatever, which works for one PM on one team.

But as soon as you start talking about okay, well, we actually have a dedicated process or way of working on our team, then that kind of falls down, right? So what we wanna do here is actually create set it and forget it processes and routines that go across your whole product stack.

Okay. So the next piece… Here, I’m gonna stop here for a second. Make sure we don’t have any questions. Okay. Cool. All right, I’m gonna go back.

All right, next piece is automation. So this is again, a great exercise to do with your team, We can guess what kind of– And I’m gonna offer a bunch of the prompts and scripts that I’ve created as part of this. But your setup will vary, right?

So just talking about you know, [00:12:00] starting for each part of the stack and for that tool you can think about what you might wanna automate. So you might wanna say at a given cadence, like quarterly, I want my AI to look at our opportunity backlog or our candidates and think about like which ones should we prioritize for our upcoming roadmap based on, insights or other our strategy stack.

Same thing for at the lower level. So this kinda goes down further and further grain down to the stack. But you can have it triage new ideas at a certain cadence. Within ProdPad, we do a lot of automation just to like help with that flow already. But this, I’ve found that most of our customers usually have like their own special sauce of triage or things that they want to tag or you know, have a certain workflow, so you can set that up.

You can obviously have it once an idea is in the definition part of your workflow, you can say “Okay, [00:13:00] once this idea is here, then go ahead and make sure that we’ve defined the scopes and we’ve generated a draft spec for someone to review.”

Definitely recommend having a human review as part of the spec phase because no one wants a spec that they don’t, they can’t read. But you can also have it generate user stories or define scopes. Same thing with prototyping, right? So this is actually incredible. You can connect you know, at this point you can have your product tool synced up with your design tool for once, right?

How long did that take us to do? But you can basically have it say “Okay, for ideas that we need to work with our design partners on go ahead and based on the idea description, go ahead and create a prototype based on that.” Or you can actually have it with the Figma MCP, you can also have it create diagrams or other things to work from.

Miro does that as well, right? And then over in development town we had integrations, but I think everybody [00:14:00] can agree that it’s not just getting a status back from the dev tool back to your other tool. You sometimes want a full “give me the full answer”, right?

Look at what’s in our product tool, look what’s in our development tool, look across everything that it’s connected to and give me a more of a richer status update, right? And then yeah, going into when you’ve released something you can ask it to measure the outcomes and then update your outcomes back in ProdPad or whatever your product tool is.

Same thing for helping maybe to test. When you’re doing testing and you wanna just make sure that your event tracking is working, you can have it do that as well. So there’s loads of use cases. But what matters is thinking about what is mission critical for you. What are the processes that your team is burning admin time on that processes like this can help you.

And last one on this column is about outcomes. Yeah, looking at your CRM or your finance apps to say what did we end [00:15:00] up with, right? And then I’m all about mixing the qualitative with the quantitative. You know, maybe you’re running a survey and you wanna actually get your NPS score and have that kind of inform what you think the final outcome was for a particular opportunity.

Okay, so I went backwards a little bit, but on the other side, here are some just ideas, right? On the vision side, A great one which I can show you that I have an automation set up for is checking roadmap alignment with your vision, right? Or strategy, right?

So just doing a check-in, does this align? Sometimes, oftentimes, you might even be at the product level, and you might have a product vision that you need to set, not necessarily a company vision. And you know, you can actually have it– That would be more of a one-off, but you could have it kind of create help to draft a vision for you or a new vision based on an updated strategy.

Say, if you wanted to totally pivot the business, right? So beyond that, you also have things like looking at summarizing your strategic [00:16:00] themes maybe linking initiatives to objective, objectives or drafting OKRs. All pretty straightforward. I think you can probably see the thing is just connections, right?

That’s what we’re hoping to achieve here, is it’s not just like level one with MCP is just doing things within one app, but the next level is basically seeing across your whole system and then having a mapping of how you want all the pieces to come together. Okay, I’m gonna stop there, come back to y’all.

We doing good? Okay. The next thing I’m going to show you is actually MCP set up, right? And ins and outs of it. So before I open all the Claudes, I’m gonna demo in both in Claude Code and in CoWork just because they have the most straight-ahead MCP setups. So one thing worth knowing is OpenAI has been shifting their approach to MCP.[00:17:00] 

They’ve been leaning more heavily on plugins, so it’s a little bit easier to set up plugins and if you’re working in ChatGPT than it is to set up MCPs. Doesn’t mean it’s not doable. You just have to click more buttons. And that is because MCP is an open standard but it is Anthropic’s open standard.

And so competition. Anyway so just some things to note as in terms of setup. So let me go ahead and come back to

My lovely black window here. Okay, so we are looking in I did a previous webinar showing just some things straight. If you wanna see just more straight one-on-one, like just one-off kind of MCP questions I’ll put the link on our recording. But you can check out my last webinar that I did.

So what I wanna show you here is just looking at MCP in general and understanding what you can do. I find it’s most easy to actually just come straight [00:18:00] into my list of MCP servers. And I can see here, I’ll look at Help Scout. They have a lot of tools, right?

And so you have the ability to view the tools or authenticate. One note I will say which I don’t think is a totally solved problem, the new spec for MCP is out, but I still find that sometimes you do have to reconnect your servers. You’ll usually get a pop-up from your AI co- agent with some things gone unconnected.

But sometimes the AI will not prompt you and will keep trying to do the thing even if it’s not connected. So before I start a session- and I know it’s gonna be running MCP, I sometimes just pop in and make sure that my tools are still connected. Okay. So coming through you can see the tools.

So there’s three parts to MCP. I’m not gonna get into that right now, but there are three different parts of it that let you do different things within MCP. But I find in order to get my brain going about what’s possible and what’s exposed [00:19:00] whether if it’s just literally the same as an API or if it’s a little bit different.

You can kind of page through here and see what’s available to you. For example you could say maybe I wanted to see all the tags that we have. And yeah, so get a specific tag by ID. So you could do that. So this just kind of helps you to get a mental model of what MCPs are available to you, and that’s also where you’ll manage, and you’ll reconnect and do that.

Now, if you’re working in Cowork or on the desktop app it’s obviously a little bit more plug and play. You just go to your settings, and I’ll show you where that’s at in a second. See, for ProdPad, we’ve got 11 tools right now, but there’s more coming very soon.

I’m going to stop sharing for a second. I’m going to-

Okay. Now everyone should be seeing we are looking in Claude CoWork. So what I wanna show you real quick is there are two different… it really depends if you or whoever’s setting this up for your organization is used to working in Claude Code. If it [00:20:00] is work involving your code base I would recommend using Claude Code.

If it isn’t, if it’s using just cloud-based like MCPs, then you don’t necessarily need to do that. For these kind of demo things I’m gonna show you, you don’t necessarily need to do that. I’m going to go to my settings and I’m going to go to connectors. They keep renaming things.

But yeah, instead of going to /MCP, this is where you can find all the things that you’re connected to. And you can set it where you can basically have Claude ask for permission or always ask or do it always unless it seems really fishy.

So that’s- Helpful to set, especially as we’re going into this new era of things. It’s good to know what your data policy is if you are connecting these tools in MCP whether that’s allowable, for example, for PII or any type of customer [00:21:00] data to go from your CRM to your product tool for example, in this, in any of these flows.

So once … And just as a review here, I’ve just set up some simple examples here. So in Claude Cowork, you don’t have to do routines. You can just do straight ahead questions using MCP. I find it, at least to start off, I always find it’s helpful to at least say use it, like this particular MCP.

For me, it’s more helpful because we have a staging server and a production, so I wanna make sure it uses the right one. So we can easily ask questions like what’s in our feedback inbox right now, group it into themes. Do we have matching ideas to link these to? And you can have that go back and it’ll do that.

And then, you can say if there are matches, go ahead and link them. If not, then create new ideas for that. And you can kind of go through a whole flow. And then if you’re a stakeholder, you could have the regular question of what’s everybody working on, right?

And can you make me a weekly report, right? And so [00:22:00] I’m not gonna open this particular one right now, but anyway, you can have it look across your roadmap and generate a report for you. Okay, going back to scheduled. Okay, so you’ll notice all of these are paused, and that’s just because I wanna show you the prompts versus actually running these.

So if you’ll remember back on our big Figma with our other stickies, right? So what I would suggest doing is for each of those things, set up routine, figure out how frequently you want something to run whether you want to manually approve it or if it can just run on its own.

For this one this is doing a theme summary of maybe you wanna just be able to quickly share what the themes are on your roadmap and group the kind of initiatives into strategic themes just for easy sharing, right? That’s a simple, straight ahead one.

So these I wrote using Claude Code first and then brought them [00:23:00] into Cowork. You don’t necessarily have to do that, but what you’ll notice is that the instructions say specifically what tools to use. And so if you have this and once you do have something running, you can check and see the status of it and whether if it breaks or if it’s going, whatever, et cetera.

And if all goes well, then all of your systems will be automatically all talking to each other, and you’ll get a lot of extra enrichment. So my, thoughts on AI making things faster or making teams faster. I don’t necessarily think that AI… if you already have a team that already has a good velocity, it’s not necessarily going to make things faster.

But what it will do is that you’ll be able to pack more into the same amount of time. And so it’s… that’s what I kinda think about is just it would it… Realistically for a [00:24:00] smaller product team, they’re not gonna spend hours and hours going through tagging up customer transcripts, right?

But now we just have more data to act on. And so it’s not necessarily making us faster as in terms of that strategic thinking but it is providing richer kind of data to make better decisions on. So this is the same as we were looking at before, but we’re looking at it in Claude Code.

So in Claude Code, it’s routines. You can start these on the command line, or you can actually do it through the Claude Code interface as well. I know I just showed you this one, but it looks a little bit different. But it shows you what tools you can set, what model you wanna have and you can have notifications set as well.

So again, this is saying ” okay using the tools, list the backlog ideas, list ideas check if it’s got a spec yet summarize which backlog ideas are missing scope definitions or user stories or no functional spec.” And so the team can prioritize. So that’s one routine that’s set up right there.

One cool one that is [00:25:00] actually a real one that I do run is using our support tool, Help Scout to, it’s a daily job. And our support team is great generally about passing stuff from Help Scout into ProdPad for us to review. But what I’ve found using Zapier over the years and then also other tools like Intercom because those conversations are threaded it’s sometimes it’s hard to use those automations, and it’s hard to find the feedback in them.

So what this automation does is it basically goes through the customer queue, and it reads, and it basically says “Okay, don’t report, don’t bring… add a bug report. Don’t ignore admin or billing.” Things that are already addressed probably already have a Trello ticket on them or something like that if it’s a bug.

But if it looks like valid feedback send it over to our product tool, into ProdPad. And so that checks every day, and that’s less work for the support team, and the feedback that you get in your product tool is already triaged and set up [00:26:00] ready for you to go. So that one is super-duper helpful.

Yeah, you can do all kinds of things. So it’s kind of sky’s the limit, but the main idea that I’m trying to share here is just think about your whole system, think about where everything lives, and think about how you want everything to connect. Map that out and then create the routines

and that’s why maybe having it as a Figma board or a FigJam or a Miro board is helpful just in terms of that everybody knows that’s how everything’s connected and what routines are running, and so you have just a nice visual of how everything goes together. One thing that we also wanted to talk about was what’s the difference between, like why would you do an MCP versus an integration?

Does MCP like automatically solve integrations for us? And the answer is not quite yet. I think it’ll probably happen, honestly. But right now we don’t have a [00:27:00] third-party neutral integration standard. So you would need to create one within your own code base to allow AI to automatically create integrations for you.

And it’s definitely, it’s not a plug-and-play thing. But what MCP can do is for something that is, more one way. Like it’s just check in check in Mixpanel to see if this event has improved by some measurement, over the last three months, and then send that data to ProdPad.

Like those types of workflows which are more like traditional Zapier workflows, like “if this happens, do this” those fit very nicely. But for your more robust dev tool integrations like we have at ProdPad or other people have as well where you have bidirectional syncing or more things are syncing in between different applications, that’s a lot more complex.

And it’s also gets tricky I think the major sticking point is just authentication and [00:28:00] who or what user is doing that motion. So I would say, you’re not gonna replace all your integrations with MCP in the next week, but maybe in the next three months, right?

Might be available to us the way things are going. But for right now, I would stick to straightforward like workflows. The other thing is that every tool we have has AI outside of the app and then AI inside of the app, and ProdPad’s no different. We have generative buttons and tools, and we have a copilot that you can talk to also automate your work.

And so how Do you need a standard for your product teams of being like, “We only use our AI agents,” or, “We use this particular… You know, we use Copilot, Microsoft Copilot in everything that we’re doing.” I think there’s kind of a how hard is it and how often are you doing it question.

If it’s something that’s maybe generating user [00:29:00] stories, like just clicking a button that says generate user stories is a lot easier than having to faff with you know, going through your agent, getting them to understand, and setting up a process for it always. And if you are already in that tool, then go ahead and use it. I just feel like all of these are tools for us, and we shouldn’t be like necessarily dogmatic about oh, we should only use this one, or we should only use that one. Use the right size AI tool for the job, right?

And yeah, I think that will set our teams up for success just in terms of picking the right option. All right, so I think I’m gonna pause at this point and check back in here. So do we have any questions, Yes. Yeah. Dominic had a good comment about MCP. That’s what we’re actually seeing with our own customers more and more, is that there are more questions about security surveys and things like that for AI. We’ve had a open season period for a while, [00:30:00] but I think it’s, for especially for larger sized companies, it’s gonna become a lot more regulated.

So I think as leaders, we just have to be clear about the value that we get from connecting these systems. Because the downside is that we’ll be back where we’ve been before, which is you know, engineering’s in a silo, product’s in a silo, design’s in a silo, and stakeholders are just trying to figure out what we’re all working on.

And so the, the dream is that by defining this it’s not enough to just say that, “We’ve got this operating model.” By plugging in the systems and actually connecting them together, we can tell a much more cohesive story to our stakeholders.

Yeah And it’s, and it can be expensive though too. So that’s actually one thing that we’re also thinking about as well for our in-app copilot, we basically choose the right model for your prompt. So it’s always [00:31:00] the most efficient what’s gonna be the quickest, cheapest model for us to do.

And by, You don’t necessarily, don’t, obviously don’t need to be using you know, Fable for prioritizing your backlog Well, thank you, Dominic. Appreciate it

Thanks, Emma. Any other questions from folks? Any other burning, like, how-tos for MCP while you’ve got me?

Oh sorry Matthew if you’re still there. Matthew had a question earlier about what application I was using to run Claude Code, and it’s called Warp, so W-A-R-P. And i use it as a harness for Claude Code. So that just allows me to run different agents and also have different tabs open instead of having a million terminal win- windows and all that stuff going on at the same time.

Just gives a little bit of extra functionality and window management for me. So [00:32:00] highly recommend it. Doesn’t cost me anything right now if you’re looking for it. Yeah we use VS Code too as well. And so yeah, give it a try, see what you think. It’s interesting, I tend to go, if I’m coding, I tend to go over to VS Code, but if I’m talking or just doing more kind of writing work, I tend to work out of Warp. Well, thank you, Matthew

Yeah, probabilistic tasks. I’ve had a reaction to a lot of the LinkedIn kind of influencers over the last few months being like, “Here’s my… Just use my 30-step process to automate your whole product flow,” and I’m like, “Don’t do that.” I think it’s just a, a lot of overwhelm.

So we have a lot of ways of working and systems already that do work. I think it’s just thinking about how you connect those things and how you would like everything to work [00:33:00] together going forward. You don’t have to throw everything away just because we have Claude, all right, I think I will leave it there.

Thank you so much. I’ll give you guys a few minutes back on the clock, but we’ll have this recording up on YouTube for viewing later for folks that are joining us another time. I’ll also be sharing out those templates if you’d like to do your own mapping exercise for mapping your stack to MCP.

All right, thank you so much everybody. Have a good night. Bye-bye.