You might think developers would be worried AI is coming for their jobs, but it's the complete opposite. As the Chief Technical Officer (CTO) of website experience platform Webflow, Allan Leinwand understands the importance of AI in the development process. He argues:
Developers love tools, and they love new things that help them be more productive, and AI is definitely helping our developers be more productive.
Webflow started working with generative AI as early as Generative Pre-trained Transformer (GPT) 3/3.5. It wanted to understand how AI could be leveraged in both the product and the development lifecycle. Over the past year, things have accelerated even more with new tools, new offerings, and new ways to use AI in the development workflow. There are more ways to develop, test, and review code using AI, and Leinwand wanted to empower Webflow's developers to do what they wanted to do, better, stronger, faster.
Initially, the engineering team began with code assist and code generation tools. They started playing with Cursor when it first came out and then tested other code generation tools. The firm conducted an internal "bake-off," where Leinwand told his developers to go try out the tools available. But before they did, they ran the tools through all the security processes, made sure they were firewalled, and there would be no loss of customer data:
We really just wanted people to sort of go off and try the state-of-the-art.
The company also played with ChatGPT and has a team license ,which means OpenAI can't train on their data. Leinwand says that many people were creating internal GPTs and agents, using them across product, marketing, and sales. He uses it himself for quarterly reviews, creating a GPT with all the documentation he would use to perform an evaluation.
On the developer side, the firm leveraged the Sonnet models from Anthropic for code generation and utilized Cursor as an Integrated Development Environment (IDE). It also uses Augment Code, which is a chatbot/remote agent that looks at your entire code base and allows you to talk to the code and make changes.
With Cursor, you have to upload your files, but with Augment, it can read your entire GitHub repository and build its own knowledge base. Developers can ask questions like "Where in the code can I do this?" or "How do we do that?" or "Where should I add this function?"
Leinwand said it's been very cool to see how developers are using the tools to support their work:
I think about when I started in software engineering. If I got stuck, I'd have to wander down to the cubicle and talk to the person and say, I don't quite get this. And we looked at the code together and tried to figure it out. Now, I can just ask my robot friends, and away we go. And, maybe it's not 100% right, but it gets me pushed in the right direction. And I think that's really, really useful. I think that's kind of overlooked: how AI can help engineers understand the existing code base. There are very few jobs that are purely starting with a blank file. They're generally inheriting some code and trying to detect it and trying to understand it.
Developers aren't worried that AI will take their jobs, insists Leinwand, although there were some concerns early on:
I think what we did have is, in the early days, maybe the earlier GPTs, people didn't necessarily trust the code gen output as much as they should have. They thought, 'Yeah, this is good, but not great, so I'm not really sure I want to spend time reviewing this code and turning it into something great'. But as the models have evolved, we've seen their initial skepticism and cynicism come around.
The narrative of losing their job hasn't landed with developers, he suggests, despite the mainstream media suggesting otherwise. As people realized what the tools could do, such as generating code, reviewing code, and writing Product Requirements Documents (PRDs) faster, they embraced them. For his part, Leinwand has seen an almost 90% adoption rate among developers who use some form of AI code generation and AI tools daily.
At the same time, he doesn't mandate or require the use of AI tools.
I don't think developers really like mandates, but I think when shown how to be more productive and how to move faster and be able to get code written, tested, and into production, and be more productive, I think people lean into that, and that's kind of been fun to see.
What he does do is track usage on a team basis (not an individual basis). He wants to understand how teams are using the tools and what he can do to support them, such as setting up the development environment differently or configuring the tools for specific use cases:
There's a metric we do track called cycle time. And cycle time is sort of the ability to track from the first time someone touches a bug or feature, all the way through to when it hits production. We've seen that come down by about 21% overall. So we're seeing a 20% increase in cycle time, and that's pretty impressive in order to see what we're actually doing.
Webflow is not leaving everything to AI. Their processes are a combination of human and AI elements. Leinwand said they have pull request reviews where each change is labeled as low, medium, or high risk, and these are generated by AI. Everything pushed into production is tagged appropriately. They let things like typos or syntax errors go. But having AI evaluate for changes that are medium to high has been very useful. Humans perform code reviews of these riskier changes (or they engage in pair programming, which combines human and AI efforts, where the AI performs the navigator role, suggesting tests and reviews) and then push the code to production.
I was curious to understand how Webflow was building AI capabilities into the platform. Leinwand said the company started building AI capabilities into the platform two years ago. It began with a feature that integrated with their educational content and videos. A user could ask how to do something, and the AI would drop them directly into the relevant document or video.
Then, they built the backend to connect to some of the big foundational models, implemented rate limitations, and ensured everything remained secure.
Webflow wants customers and users to be able to use the platform's features more easily. AI features such as text improvement, image generation, sample Content Management System (CMS) item generation (e.g., blog posts or product descriptions), or the ability to pick your favorite design and create it are built into the product.
There's also the AI Site Builder, which enables users to create a new website from a single prompt. The Site Builder brings in all the design system elements you've already used from fonts, colors, padding, layouts, and image types, and creates the new site.
For companies that can't afford to hire large development teams to build their websites, the AI Site Builder is a welcome tool, allowing them to get their sites up and running faster, working better and more efficiently, and improving all the time. Optimization is super critical.
Leinwand points out that AI isn't a product. It's a capability that extends from building a site to generating new sites and optimizing existing ones. Webflow will continue to put that AI thread throughout its product.
As a CTO, there are a lot of things Leinwand has to think about. But what excites him most about AI going forward? He says:
I think the most exciting thing that I see people doing right now with websites is this personalization aspect. Being able to make the website respond to so many different things. Maybe it's the geography you're coming from, maybe it's the language you're speaking in, maybe it's the products you're searching for, maybe it's how you filtered and searched on the website. Being able to really pursue that experience and personalize it to the visitor is something I think is super powerful for our customers. Because they really want that engagement. At the end of the day, the website is to engage for something. It's to engage for marketing materials. It's to engage for a product. It's to engage sales or support. There's some business need, and if we can help our customers and their visitors get to that engagement faster, that's just super cool.
As for the website becoming obsolete because of AI? Leinwand doesn't see that. He does see two types of visitors to the website: bots and humans. You have to build for both audiences. He said brands want to know how their product is being represented in LLMs (Large Language Models) and chatbots:
I'm going to spend a lot of time thinking about that. And I think being able to develop features and products on our side that allow our customers to influence how that AI engine optimization is occurring, as opposed to how classic SEO is occurring for humans, is going to be a really exciting world. I'm looking forward to seeing that evolve, because, from my perspective, honestly, it just means more people coming to this thing called the website and the consumption model changes. And as the consumption model changes and grows, there's just a lot of really cool engineering problems there.
I enjoyed my conversation with Lienwand because it was clear that he, and Webflow, were looking at AI with a clear idea of how they wanted to leverage it. The first step is understanding its value and how it can help developers (and other employees) work more efficiently and effectively.
It isn't about taking away jobs; it's about helping people work smarter. This approach seems more straightforward in the development world than in other areas, such as marketing and customer success. But it's also not without its challenges, and Leinwand seems prepared to address any head-on, while also encouraging his team to continually test and try new tools.
