Saulius LazaravičiusHostinger’s VP of Products is a technology leader with a career spanning more than 20 years in software engineering, product development, entrepreneurship and e-commerce. In his early career, he was a programmer before co-founding B2B software company. He helped create a platform for transportation management from scratch. After joining NFQ he headed the cross-functional tech teams as well as client product developments. Lazaravičius joined Hostinger as Head of Product in 2022 and was promoted to VP of Product in 2024, expanding his responsibilities from web hosting and managed WordPress to a broad portfolio encompassing core infrastructure, VPS, domains, business email, product design, product growth, and emerging AI products. Recent work included the growth and launch of Hostinger’s email marketing service and the introduction managed solutions for Open-Source AI agents.
Hostinger It is a Lithuanian-founded tech company which has transformed from being a web host to a platform that uses AI for creating, launching and growing online business. The company, founded in 2004, now boasts 900 staff and serves over 5,000,000 users from 150 different countries. The company’s portfolio includes web hosting and cloud services, domains, email for business, ecommerce and AI creation tools. Hostinger’s strategy has increasingly placed AI in the forefront, with Hostinger Horizons allowing users to build and publish web apps and websites using conversational AI. It also offers integrated AI capabilities such as e-commerce and customer service, along with managed AI agents.
As a former software engineer you built teams and then led them. Today, your portfolio includes core infrastructures, VPSs, domains, managed solutions, as well as open-source AI agent agents. What has this progression done to your thinking about the needs of developers beyond just generating code from AI?
It’s not the writing of code that I find most fascinating. Finding the best solution for a user or technical problem is what interests me most. I form a hypothesis and build something to solve it. Then, I see how users and other people use it.
In the past developers spent a large amount of their time on repetitive tasks: looking for code and libraries, connecting pieces, debugging issues, testing and fixing them before they could evaluate the quality of the solution.
The AI shifts that balance. Many of those repetitive tasks can be done with just a few prompts. This leaves developers with more time to focus on the creative side: analyzing the problem, exploring alternative approaches, making decisions about the product, and learning through real-world user behaviour.
AI is most valuable for me in this area. The goal is to generate code more quickly, but it’s not the only thing. The goal is to remove more work that’s not interesting but mandatory, so that developers have more time for the important problems.
AI tools for coding have drastically reduced the time needed to get from an initial idea to a working program. You think that deployment and infrastructure are the biggest bottlenecks for AI-assisted Software Development?
The next big source of friction is likely to be the upcoming social media.
AI is dramatically accelerating the process of turning an idea into code. Once the code is written, it still needs to be placed somewhere, configured correctly, connected services, handled domains, security and kept running. When you still have to navigate between multiple dashboards or documentation pages and terminals in order to complete these steps, your workflow slows again.
It is clear that developers are looking for the same level of speed in moving from an idea to a code. Hostinger Connector is used daily by around 1500 customers, who are performing between 30 and 40 actions per day. This is an indication that AI-assisted deployment and operations will be integrated into development workflows.
Hostinger states its Connector is already processing 30,000 to 40,000 actions per day You can find them on websites and other services. Have you been surprised by any of the usage patterns that developers have asked AI agents to perform in production?
Connector is available directly in the tool that developers are already using, including VS Code Antigravity Cursor. This shift is important because it integrates infrastructure into development flows instead of forcing developers jump between dashboards.
Deployment is one of the most obvious use cases. MCP is responsible for the deployment of more than 14600 unique sites. Weekly deployments have increased by 300% during the past month.
Connector allows agents to do more than just deploy. It also manages domains and DNS. This means that an agent could move from helping to build a website, and then put it live online. They can also manage the services surrounding it.
The difference between giving an AI agent the power to write code versus letting it change DNS records or SSH keys and firewall rules is significant. Once agents have this much operational power, what new safeguards are necessary?
Agents are no longer merely suggesting actions. This is a real change to a system in use, and security and control becomes more crucial.
This means that agents must have defined permissions, secured access and transparency into their actions. Developers should, in an ideal scenario, not have to deal with all this complexity.
Hostinger Connector, our agent platform and other products are all based on this approach. Platform handles infrastructure, security, and access layers for the agent so that developers don’t have to worry about configuring or maintaining anything underneath.
“Vibe coding” makes software development available to those with limited infrastructure and DevOps knowledge. What happens if these users start deploying apps through agents before they fully understand the systems beneath them?
It seems to me that this places more responsibility with the platform than the end-user. We cannot expect everyone to be an infrastructure expert if AI allows more people to create software.
Platforms should take on more complexity, including monitoring, backups, security, resource limitations, and defaults that are sensible. The users should be able to understand the high-level issues, but not have deep DevOps expertise in order to safely get their application up and running.
AI agents have shown a pattern similar to ours. Managed OpenClaw was a product that many users found challenging. The challenge wasn’t what it could do but how to make sure the software ran reliably. This included hosting, configuring, obtaining API keys, securing, and maintaining the system. The technology became much more available when the setup was removed.
I expect infrastructure to increasingly operate in the same fashion. Platforms should allow users to concentrate more on their projects, and let the platform handle the complexity of operations in the backend.
What context is required for an AI agent operating infrastructure to do so reliably? This includes dependencies on server resources, policies and credentials as well as security.
Agents need enough context in order to fully understand the tasks and the surrounding environment.
Simple actions like restarting the server or changing DNS records can have a significant impact on other components of the system. The agent must have structured data about resources that they operate: What is available, which resources depend on it, how many resources there are, the permissions of the agent, and previous events.
Having more context does not mean that you have unlimited access. Only the data and permissions that are necessary to perform the task in a safe manner should be given to the agent. The wrong choice would be to grant an agent access that is excessive, just so they can have more context.
When should humans remain required to approve certain actions? What are the categories of actions in infrastructure that AI agents could perform independently and which ones should require human permission?
AI agents are capable of operating largely independently, provided the system has been configured correctly from the beginning.
Rather than relying solely on one agent, multiple agents could have different responsibilities. The first agent could be in charge of coding, the second can do the testing, the third monitors the system with alerts and the fourth one can escalate the issue when it goes wrong.
This setup does not require the approval of every single action. They are responsible for overseeing the entire agent system, including defining the rules and setting the permissions.
The key question for me isn’t so much which actions require human approval, but whether or not the entire system was designed to include the correct checks, responsibilities and escalation pathways.
Do you anticipate that the boundaries between DevOps platforms and cloud management interfaces will disappear as development environments become more and more interfaces for deploying apps, managing domains and servers, or controlling external services?
The boundaries will blur, but I do not expect every interface to disappear.
Today, to bring an application from development into production, developers must often navigate between a Web IDE, a Domain Panel, a Hosting Dashboard, a Terminal, and multiple external services. The AI agent can integrate many of the steps, bringing them to the place where developers are already working.
This does not mean that dashboards and cloud-based management tools will be gone. These tools will be still useful to perform deeper configurations and overseeing. Developers may not need to worry about the next interface to use for everyday tasks.
It’s less important which tool is traditionally associated with a particular task and more about what you are trying to accomplish.
The AI agent can make improvements to infrastructure faster than humans but can also make errors at the speed of machine. What will become more important as the agentic technology matures in terms of capabilities like audit trails and rollback mechanisms?
As agents begin to take on greater operational responsibilities, these capabilities will be essential.
AI allows infrastructure to be changed much quicker than by a human, which can prove useful if everything is going well. However, the same speed also makes a mistake more likely to spread. Users need to be aware of what has changed, who the agent can access and what they should do if things go wrong.
This type of automated system is only trustworthy if it has audit trails, permission limits, monitoring and rollback mechanisms.
When agents are more intelligent, the quality of their safeguards matters just as much.
In the near future, will AI-development platforms that have the best code models win? Or, will agents with secure access gain a competitive advantage by having access to organizational contexts, production systems, and infrastructures tools.
It is not enough to rely on the quality of a coding model.
After models have proven to be good at generating codes, what comes next is the real question. Can the agent actually deploy the application? Does it know how to use the services correctly, can it understand the environment? Can the agent make sure everything is working properly and safely?
Infrastructure access, tools and permissions as well as context are all important.
My opinion is that the best AI development platform will have a combination of good models, and secure access around the system code. It’s not about writing software faster. The value comes from helping people go straight from an initial idea to a finished product.
You have done a great job. Readers who want to know more about various hosting options should visit Hostinger.
