Your Agent Needs a Roadmap From You

Ben Sufiani: Hey everyone, this is Ben. Today I'm sharing my latest captain's insights. This time, why I think that your agents that you are building new products with need a roadmap from you. They can now build anything, and the question becomes what is worth building? And this is what they look to be still very bad at. And we need to provide them guidance as humans. This also frees up again a lot of bottlenecking. that I experienced after building like four projects in parallel and I want to share that with you today. So here is what happened to me. I still recognize that I am the bottleneck by the following behavior. Now that I have a full schedule, I'm busy, I have multiple projects, I tend to work in bursts on specific projects, and in the meantime, they sometimes don't make any progress for a whole week. So here on the graph, if you're reading this on pirateskills.com or watching this on YouTube, you can see that on every given any given month. I make occasional progress, maybe sometimes just Tuesdays, then next week it's Wednesday, Thursdays, or Fridays. And it's all very sporadically. What you want in contrast is to make progress by default on an ideally daily basis or even more frequent. And then it's totally fine if you drop out on the weekend But you see the agents were never the slow part. I was. it is my fault that I didn't guide it properly with the help of what we usually call a roadmap. So that is our new job. So the job that stays very human and with us are two things. Number one is we decide what is supposed to be built next and maybe guide the agents with some additional context for what exactly we want to be done next. And then we let it do stuff, and then we review what the agent has built before the customer comes in. In practice, this looks something like this. Here's a good example for customer magic. On Sunday, I had the idea that when I want to add pirate skills to customer magic, there should be some transparent process of how the data is being used, like in terms of compliance. So I had a chat with the roadmap session in Customer Magic about that. And We defined that we wanted to have one place that is human and agent accessible on how the data is handled. Now this becomes a roadmap item. And it can, given the existing context, write its own brief. But in this case, I had it chat with the other sessions in PirateSkills to define the clear needs for this new compliance page. Might also be a bit abstract, so let me show you. This is my cloud. This is the customer magic roadmap session. You can see on the right hand side there are the upcoming projects and what has been completed. And there is this session called the trust tab. This is what turns out to be the thing that I wanted to build. And to show you how that looks like in real life. Here it loads. We now have a page in the settings in Customer Magic that clearly shows what is being saved with the workspace and how AI is being used, what data is processed, the different subproviders that I'm using and subprocessors, all with the associated agreements that I have with them. So an agent has it a really easy time to process. all of this. I thought this is what I would like a lot of American companies to do to make my life easier, but they don't. This is the GRACE session. this is my governance, risk, and compliance partner that handles everything in that direction. And naturally, if another app wants to integrate the PirateSkills data, there needs to be some process here. And Grace provided the context for that session so it could write its own better briefing. I can review that and then it can start the process and starts building that stuff out. And then I can review it. But the whole building process is off my plate. It's usually like running overnight. And I get to review it. Is that what I wanted? Do I think this is good enough? Do I need to do final changes or require that? And then get it into the hand of the customer. So I remain the job of being the initiator of telling what's next and being the reviewer at the end, freeing a lot of my time up and getting out of that mode where I was just focusing like on Tuesdays, all on pirate skills. And now I can just keep those sessions going and just provide what is next and review what is open. You might ask yourself why deciding stays human. That is a good question. And I think we are currently in a time where the agents are simply not that good at deciding on a macro level what is good. It is very good on the implementation level, it's very good as a strategic partner, and I use it in a way where I ask it to give me ideas. For the roadmap, but then I decide what. For example, when I asked the agents on the next on the roadmap for pirate skills, it told me, hey, integrate with customer magic, finish up the mobile app, compliance, slides, messaging. And I give it some feedback and we ended up, yes, customer magic is the correct number one priority. but we need to think about the hackathon that happens in November. And it just didn't have the right context on that, which is my fault. Then we are also working on the slides that is tied to an upcoming event, and I need to prepare that. The mobile app is essentially just on hold with Apple for now. Google's already live, Apple is still waiting, and I'm trying to figure things out, what is going on wrong there. And then compliance and its tool. and we cut out the messaging stuff. So pretty close, but I'm still the person. making that call. And you could argue correctly that the agent just doesn't have the correct context to make that decision. And I see that as my responsibility as well. When I notice that an agent is clearly not aware of a certain project, then I ensure that it has access and understanding about that context that goes into the project rules and skills and again expands the capability of augmenting me as a human. I'm not worried about it replacing me. So, how do I want such a roadmap to look like and where is it supposed to live? let's take an example of Zorg. Remember, Zorg is my little project where I create an AI content team that really absorbs your taste before providing repeatable content formats for you and publishes them to the channels that you desire. And this is a Big project in a sense that it has many items. I need to keep an eye on. And I completed several steps where there are some key parts that are not done yet. It realizes concretely: okay, I need to create the authentication feature. So I have accounts with sign-in and sign-up and workspaces. But the other ones are like semi-important, and I need to have a discussion on that. And it's Totally fine to have this in a simple document. I think there are three good places where to manage that. And I want to show you all three. So let's go to Cloud Code and look for the ZOG roadmap. Here is my session. Again, I have a strong tendency to have sessions with certain responsibilities. And here is what is called an artifact. An artifact is the way Anthropic has decided. That it can create like documents that are straight online. So they don't live in the code base, but they're very easy to use, also on the go. And I have this one important document with lots of sub pages where I describe like what is Zog, what is the audience, what is our current tech stack, how's our pricing planned, the naming, research, all of that stuff was in here. And now we get a new page called the roadmap. Very simple. Four phases, ideas or a backlog of everything that I want to get done, but that is not in my current priority. Planned is already sorted by priority. So these are the ideas that are good enough to be next in the pipeline. In progress is everything that I'm working on right now, can be one project, can be five, and in completed, it's everything that I've already done. So I have that context aware as well. So this is an artifact. The good thing is it's the easiest to set up if you're working in Cloud Code, just sell it, hey, create a document for this and that, and it will create these sidebar artifacts. But let's say you're switching between codecs and Cloud Code. These artifacts are not easily accessible and editable for codecs. So there you just want to tell it, create it as like an HTML or markdown file in the code base. That is the second way. Then the third way is to put it into an external tool. So this external tool for me is linear. You can use whatever you like, Jira, Linear, Trello. For me, Linear works wonderfully. Here is my Zog dashboard. It already looks more better designed, thought out, and everything. And another huge advantage is here that everything can have like subtasks and all the code changes are associated with these tasks. It's very transparent for the agent what has been done compared to just having Artifact. But you need to want to work with another external tool. It's definitely the more complex solution, but the one I currently prefer and that I'm willing also to pay 10 bucks a month for. what you can see here, what is very important is I know this is working right now and this is ready for me to review. this is this looks like this in my code base. Here is the Creative Gen generation. I ask in the roadmap session, hey, can you please create a new session for the design project? that is then renamed into creative gen. And here it has already the local dev server ready for me to review the data and has written me a guide on how to test this. Like open this page, do this and that steps, give me your feedback throughout, I'll give it its feedback, it usually wraps up, finishes, I approve, and then boom, I push it live to the customer. And then this part is completed. Now the question becomes what's next? And this is the biggest difference in having it in a roadmap. I can now easily say, hmm, what do I agree with most? I want to wait with the accounts until Wednesday when I do it during the Pirate Forge workshop. So I don't want to pull ahead that. So I can now tell it, hey, let's start working on projects. Please don't create the accounts and the workspaces yet because I want to do that in the workshop on Wednesday. And then I kind of have developed a habit of nudging this process on like a twice a day basis. So what I mean by that is that usually before any meetings in the morning, I look through here, I essentially just have the discussion here in Cloud. I go to the roadmap and I say something like this. Okay, let's assume design is finished. What are the next top three things you would do that does not affect authentication yet? So I can have a normal discussion about this. But I'm again and the person that decides, and every feedback that I give it now in these little road mapping sessions, I will be able to put straight into the plan. It sinks here between the artifact and linear. And I, for example, would say now, okay, let's work on the review inbox. I don't want the funnel numbers, I want to do that in another workshop. Again, but these are my personal reasons that the agent just cannot know everything about. And then this would create a new session, interview me shortly on what exactly the review inbox is, and then start designing it, building it, verifying its work, and then telling me that it's ready for review again. And if I do this in the morning, I go like this is the customer magic thing, this is my pirate skills roadmap. And I kind of heard these little agents in front of me making consistent progress in the morning and in the evening. I love letting them run overnight. And then review things again in the morning. That way I get my normal work done while I make really significant progress on my projects. So now it's time for you to give your agent a roadmap. If you want to try this and you are on the article, otherwise the link is in the bio. You can scroll down to this section called Give Your Agent a Roadmap, and there is a prompt to be copied. If you put this in a new session in your project, it will read what the project is about. It will look at the recent Work and it will ask you a couple of questions to clarify and then propose projects in the four stages, ideas planned in progress and completed. And you do not have to use any external tool with this. This will set it up as an HTML page. Once you like it, you can put it into a more professional project management tool. Maybe you don't like the first version that came out of that. I didn't like mine either. then I just took a screenshot from Linia and then it proposed something much more in that direction. So now you're one step further along this genetic journey that we are all on. We know the agent can now build almost anything, but it's not good at telling what is worth building for us humans and for other agents. And that is now our job, maybe once or twice a day to hurt your AI agents and update your roadmap. I hope you like this and I see you in the next one. Before you leave, let me tell you how we can go deeper into the Pirate Skills journey. We have a couple of events I want to tell you about. The AI compass is the right one for you if you think, Ben, but where do I get started with all of this AI stuff? This was pretty advanced. Then the AI Compass is for you, happens always on the third Wednesday of the month, sometimes the fourth. And this is where we give you the best introduction how you can essentially go beyond ChatGPT. This is what we did today, it was something we would discuss in the Pirate Lab. This is more advanced. Both events are online. And if you want continuous new input, Pirate Lab is excellent for you. Both free, both online. Our offline events are happening in Cologne, Germany. We have our monthly meetup called Vibecoding Cologne, always on the first Wednesday of the month. You're very welcome to join. And every quarter we do a big long Weekend, a whole hackathon weekend that really allows you to emerge yourself into this world. that is usually priced between 44 and 90 bucks. Then if you want to take your app building profession more serious and you want a bit more handholding in learning that, I have a program called the Pirate Forge that takes you six weeks of vibe coding, six weeks of vibe marketing. And helps you create a revenue-ready SaaS product for the first time without any coding background, and then applies those skills also to the marketing process. If you're a developer and you think like, I don't need that, just the growth thing, then of course you're welcome. Just talk to me, book a call, and we'll figure something for you out. That's it from me today. Thank you for joining, and I see you in the next one.