Saulius LazaraviΔius, VP of Product at Hostinger, is an skilled expertise and product chief whose profession spans greater than 20 years throughout software program engineering, entrepreneurship, e-commerce, and product improvement. He started as a software program engineer earlier than co-founding a B2B SaaS firm, the place he helped construct a transportation administration platform from the bottom up. He later shaped and led the software program improvement workforce behind BARBORA, one of many Balticsβ largest grocery e-commerce platforms, earlier than becoming a member of NFQ, the place he managed cross-functional expertise groups and shopper product improvement. LazaraviΔius joined Hostinger as Head of Product in 2022 and was promoted to VP of Product in 2024, increasing his obligations from internet hosting and managed WordPress to a broad portfolio encompassing core infrastructure, VPS, domains, enterprise e-mail, product design, product development, and rising AI merchandise. His current work consists of the launch and development of Hostingerβs e-mail advertising and marketing providing and the introduction of managed options for open-source AI brokers.
Hostinger is a Lithuania-founded expertise firm that has advanced from a standard internet hosting supplier right into a broader AI-driven platform for constructing, launching, and rising on-line companies. Based in 2004, the corporate now serves greater than 5 million customers throughout over 150 international locations and has a workforce of roughly 900 staff. Its product portfolio spans internet and cloud internet hosting, managed WordPress, digital non-public servers (VPS), domains, enterprise e-mail, e-commerce, and AI-powered creation instruments. Hostinger has more and more positioned AI on the middle of its technique, together with Hostinger Horizons, which permits customers to create and publish web sites and internet functions by conversational AI, alongside built-in AI capabilities for e-commerce, buyer help, and managed AI brokers.
You started your profession as a software program engineer, later constructed and led improvement groups, and now oversee a broad Hostinger product portfolio spanning core infrastructure, VPS, domains, product development, and managed options for open-source AI brokers. How has that development formed the best way you consider what builders really want from AI past merely producing code?
For me, essentially the most fascinating a part of software program improvement has by no means been writing code itself. It’s discovering the best resolution to a technical or person downside, forming a speculation, constructing one thing round it, after which seeing how individuals truly use and profit from it.
Prior to now, plenty of the developerβs time went into the repetitive work round that course of: trying to find libraries and code snippets, connecting completely different items, debugging, testing, and fixing small points earlier than you can even correctly consider whether or not the answer was good.
AI modifications that stability. A lot of that repetitive work can now be dealt with with a number of prompts, leaving builders extra time for the artistic half: understanding the issue, exploring completely different approaches, making product selections, and studying from actual person conduct.
That’s the place I see the most important worth of AI for builders. It’s not nearly producing code quicker. It’s about eradicating extra of the obligatory however much less fascinating work so builders can spend extra time fixing the issues that really matter.
AI coding instruments have dramatically decreased the time it takes to go from an thought to working code. Do you consider deployment and infrastructure administration are actually changing into the larger bottlenecks in AI-assisted software program improvement?
Sure, I feel they’re changing into the subsequent main supply of friction.
AI has compressed the time between an thought and usable code dramatically. However as soon as the code exists, you continue to have to put it someplace, configure it appropriately, join companies, deal with domains and safety, and preserve the applying working. If these steps nonetheless require switching between a number of dashboards, documentation pages, terminals, and instruments, the workflow out of the blue slows down once more.
We will see that builders need the identical velocity from code to manufacturing that they’re now getting from thought to code. Greater than 19,000 prospects have already tried Hostinger Connector, round 1,500 are utilizing it each day, and they’re finishing roughly 30,000 to 40,000 actions per day. That could be a sturdy sign that deployment and operations are shifting instantly into the AI-assisted improvement workflow.
Hostinger states its Connector is already processing 30,000 to 40,000 actions per day throughout web sites and companies. What are builders truly asking AI brokers to do in manufacturing, and have any of these utilization patterns stunned you?
Builders use Connector instantly from the instruments they already work in, similar to VS Code, Antigravity, and Cursor. That shift issues as a result of it brings infrastructure actions into the event stream as an alternative of forcing builders to leap between separate dashboards.
One of many clearest use instances weβre seeing is deployment. Greater than 14,600 distinctive web sites have already been deployed by MCP, and weekly deployments grew by practically 300% over the previous month.
Past deployment, brokers use Connector to handle domains and DNS, web site information, server assets, firewall guidelines, SSH keys, ecommerce duties, e-mail campaigns, and different Hostinger companies. In follow, meaning an agent can transfer from serving to construct a venture to truly placing it on-line and managing the companies round it.
There’s a vital distinction between permitting an AI agent to jot down code and permitting it to vary DNS information, firewall guidelines, SSH keys, or server configurations. What new safeguards develop into crucial as soon as brokers are given this stage of operational management?
The important thing distinction is that the agent is not simply suggesting what to do. It’s making actual modifications to a reside system, so safety and management develop into rather more essential.
Which means brokers want clear permission boundaries, safe entry, and visibility into the actions they carry out. However ideally, builders mustn’t must handle all of that complexity themselves.
That’s the method we’re taking with Hostinger Connector and our agent platform. The platform handles the infrastructure, entry, and safety layers across the agent, so builders can deal with what they need their brokers to do moderately than on configuring and sustaining all the things beneath.
βVibe codingβ has made software program improvement accessible to individuals who could have restricted information of infrastructure or DevOps. What occurs when these customers start deploying functions by brokers with out totally understanding the programs beneath them?
I feel this places extra duty on the platform, not on the person. If AI makes it potential for extra individuals to construct software program, we can not count on all of them to develop into infrastructure consultants as nicely.
The platform must handle extra of the complexity beneath, from safety and backups to monitoring, useful resource limits, and wise defaults. Customers ought to perceive what is occurring at a excessive stage, however they need to not want deep DevOps information simply to get an software on-line safely.
We’ve got already seen an analogous sample with AI brokers. With merchandise similar to Managed OpenClaw, the problem for a lot of customers was not what the agent might do, however all the things required to run it reliably: internet hosting, configuration, API keys, safety, and upkeep. Eradicating that setup burden made the expertise rather more accessible.
I feel infrastructure will more and more work the identical method. Customers ought to be capable of deal with what they need to construct, whereas the platform handles extra of the operational complexity safely within the background.
For an AI agent to function infrastructure reliably, how a lot context does it want in regards to the surrounding atmosphere, similar to dependencies, server assets, safety insurance policies, credentials, and former configuration modifications?
An agent wants sufficient context to grasp not solely the duty, but additionally the atmosphere round it.
A easy motion like altering a DNS file, restarting a server, or deploying a brand new model can have an effect on different components of the system. So the agent ought to have entry to structured details about the assets they’re working: what’s deployed, what is dependent upon it, what assets can be found, what permissions the agent has, and what occurred earlier than.
On the similar time, extra context mustn’t imply limitless entry. The agent ought to solely obtain the knowledge and permissions it truly wants to finish the duty safely. Giving an agent extreme entry just because extra context would possibly make it smarter could be the improper tradeoff.
The place ought to human approval stay obligatory? Are there classes of infrastructure actions that you just consider AI brokers ought to be capable of carry out autonomously and others that ought to all the time require specific human authorization?
I consider AI brokers can function largely on their very own, so long as the system is configured correctly from the beginning.
As a substitute of counting on one agent to do all the things, completely different brokers can have completely different obligations. One can deal with coding and deployment, one other can take a look at the outcome, a 3rd can monitor the system and lift alerts, and one other can escalate points when one thing goes improper.
In that setup, the human doesn’t have to approve each particular person motion. Their function is to supervise the entire agent system: outline the foundations, set the best permissions, monitor how the brokers work collectively, and preserve enhancing the setup over time.
So for me, the important thing query is much less about which particular person actions all the time want human approval and extra about whether or not the general system has been designed with the best checks, obligations, and escalation paths.
As improvement environments more and more develop into interfaces for deploying functions, managing domains, configuring servers, and controlling exterior companies, do you count on the normal boundaries between the IDE, DevOps platforms, and cloud administration consoles to vanish?
I count on the boundaries to blur considerably, though I don’t suppose each specialised interface will disappear.
Immediately, builders typically transfer between an IDE, a internet hosting dashboard, a site panel, a terminal, and a number of other exterior companies simply to get one software into manufacturing. AI brokers can join a lot of these steps and convey them into the atmosphere the place the developer is already working.
That doesn’t imply dashboards or cloud administration instruments will disappear. They may nonetheless be helpful for deeper configuration and oversight. However for a lot of on a regular basis duties, builders could not want to consider which interface they must open subsequent.
The expertise turns into extra in regards to the activity you need to full and fewer about which software historically owns that activity.
AI brokers can doubtlessly make infrastructure modifications far quicker than people, however they will additionally make errors at machine velocity. How essential will capabilities similar to audit trails, rollback mechanisms, permission boundaries, and steady monitoring develop into as agentic improvement matures?
These capabilities will develop into important as brokers tackle extra operational work.
AI could make infrastructure modifications a lot quicker than an individual, which is helpful when all the things goes proper. However the identical velocity also can make a mistake unfold a lot quicker. That’s the reason customers have to know what modified, what the agent was allowed to entry, and learn how to recuperate if one thing goes improper.
Audit trails, permission boundaries, monitoring, and rollback mechanisms are what make that sort of automation reliable.
As brokers develop into extra succesful, I feel the standard of those safeguards will matter simply as a lot because the intelligence of the agent itself.
Trying forward, do you suppose the successful AI improvement platforms can be these with the most effective coding fashions, or will aggressive benefit more and more come from giving brokers safe entry to infrastructure, instruments, organizational context, and manufacturing programs?
The standard of the coding mannequin will nonetheless matter, however I don’t suppose it will likely be sufficient by itself.
As soon as fashions are good at producing code, the larger query turns into what occurs subsequent. Can the agent deploy the applying? Can it join the best companies, perceive the atmosphere, make modifications safely, and confirm that all the things is working?
That’s the place infrastructure entry, instruments, permissions, and context develop into rather more essential.
I feel the strongest AI improvement platforms will mix good fashions with safe entry to the programs across the code. The true worth is not only serving to somebody write software program quicker, however serving to them go from an thought to a working product with fewer steps in between.
Thanks for the good interview, readers who want to be taught extra in regards to the varied internet hosting options on supply ought to go to Hostinger.
