Yann Kronberg is presenting a suite of AI agents that can generate a full project pipeline—from SOW to skeleton code in three days, boosting developer productivity by 10x to 100x.
Arham: Welcome back everyone. Today we’re going to be talking to Yann Kronberg about some of the pretty interesting applications being developed internally within Zazmic. Yann, I want you to start off with the origin story and just walk the audience through an overview of what these applications entail. Why did you even kind of begin developing them in the first place?
Yann: Thanks Arham. All these things that we’re building are born from a necessity and are tasks we were doing repeatedly every day. For instance, we’re always trying to make proposals for big software or AI development projects. Usually, what we do is we show a prototype, and based on that prototype, we derive an SOW. In that SOW, there are obviously estimated timelines and so forth. When we started this business, we would just use Figma and produce a clickable prototype.
We meet hundreds of clients a year and do something like 300 SOWs every year, which is huge—almost one a day. We decided to streamline the entire process. We would record a meeting with the client and derive the scope based on that. Based on that scope, our application would automatically produce an SOW with a timeline and a primitive prototype. Once we achieved that, that was the first application.
We then pushed the envelope to automatically build Epics and features in Jira using the scope and the Figma file. Our designers would iterate on the Figma prototype based on client feedback, and we would then automatically derive Epics and Features in Jira so our engineers could get organized and develop faster. A third thing we did was take the auto-generated code that Figma provides for React, plus backend skeleton code generated by several LLMs, and we would have essentially skeleton code for the SOW.
Yann: Basically, we have a meeting with the client, we extract the scope, based on that hour meeting, and we make an SOW out of that. Then based on that SOW and the same meeting, we produce a prototype in Figma. Then based on that, we produced a project and its corresponding ticket in Jira.
Arham: The entire project management journey that they’ll be going through.
Yann: Yeah, that’s right. The first one creates the SOW. The second one creates the UX/UI based on the SOW. The third one creates the JIRA epics and features in order to build that project. That’s the three things we’ve created so far.
What you see is a process that would take us six to eight weeks. In three days, we have an SOW, a pretty good clickable prototype, features and epics in JIRA ready for our development team, and some skeleton code ready to be reviewed and revised by our development team. Essentially, that is speeding up the entire manufacturing process of making software for our clients.
We applied that not only for applications but also for our AI dev projects, traditional software projects, and cloud migration. Building these three things resulted in a massive amount of productivity.
Arham: Awesome. There’s often a huge gap between making a tool that is very functional internally and making it a commercial product. Do you plan on taking this to the market and making such a product available to other large organizations that are building? Could you talk a bit about going from that internal app to actually making a commercial product?
Yann: There’s a gap for sure. We’ve started publishing this agent on our marketplace and are preparing to publish it on the Google marketplace. The gap lies in how people work. The agents are really useful for people who share our design-focused workflow, where we start with a design and derive everything else from it. But if you work differently, the agents won’t be as effective for your organization. So basically, it’s more about training.
We’re trying to bridge that gap. We’ve successfully done so for the manufacturing, UX/UI, and SOW aspects. However, the engineering part still needs some work to fully fit the workflow of the companies using the agent. We’re getting close, and this progress is driven by how good these tools are getting. They’re getting better and better at extracting useful “skeleton code” to help engineers go faster, so there’s still a bit of work there.
Arham: AI tools are often discussed in terms of task automation and increased productivity. I’m curious to know, as you roll out your commercial product, how do you see it impacting the jobs and workflows of people in other organizations? In what ways do you think it will change things for the better?
Yann: For traditional engineering teams, productivity can go up by 10 to 100 times. So, customers expect us to deliver a lot more and a lot faster.
The way developers work has changed, even if their involvement hasn’t. Instead of starting from scratch, they now get to begin with skeleton code—sometimes a pretty advanced set of code—for their project. The developer’s role is more like a reviewer, taking that initial code and building it up.
The developer is still completely necessary at every step, maybe even more so than before. Their role has changed a bit, though, to require more analytical and reviewing skills. This also means more experienced developers will likely get more out of it. A junior engineer, for example, won’t get the same level of productivity. Ultimately, this change in workflow makes the job more interesting for the developer, too. So, everybody benefits: more productivity, more engaging work, and a faster process for the customer.
Arham: Seems like a win-win-win situation that you’re trying to build out for, which will help both the customer and the organizations using it. My final question is about AI itself.
Looking back at the entire process—from a simple idea within your company to a solution used by your customers—what do you believe was the most important non-technical element that ultimately determined its success? And what do you think that element will be moving forward?
Yann: Same as before, really. Developers have to be creative and build solutions that solve real problems.
With AI, what’s interesting is that solving issues involves a lot of experiments. There are a lot of things where you just don’t know if it’s going to work out right away. So you can do a lot more, and solve a lot more, with AI. You just have to be brave and try a lot of different things.
But I think it’s the same as before. Software was always about solving a lot of issues by being creative and doing new things. Incorporating AI into that process is kind of the same thing; it’s just an additional dimension to the work.
Arham: An AI tool is only as good as the person using it. Do you have a timeline for when the product will be on the marketplace?
Yann: We’re releasing two agents, actually—one for social media and one for manufacturing processes. We expect them to be on the marketplace by the end of September.
Arham: That’s great to hear!
Let's get in touch!
We'll send you more details