“Don’t be evil” – as a baseline. For humans and for AI.
Illia Kovtun is a Software Developer at Substantio and is currently working on AI-powered solutions inside the platform. In this interview, he talks about what makes product development in the compliance space unique, how AI can eliminate tedious routine work and why he remains critical of today’s AI hype.
Illia, you’re a Software Developer at Substantio. What’s special about software development in the material compliance space and what do you enjoy about it?
It’s not that unique, really. Sure, it takes some time to build at least a basic understanding of regulations, standards, and tools. But in the end, it’s just code.
What I’m proud of – and what makes the work special for me – is where feature requests come from, and how flexible the product is in implementing them. In compliance software you often see this: a company builds a product around its own rules and formats, and then tries to force every customer into that workflow. Substantio is different. We have a core structure but every feature you see was once requested by a customer.
Sometimes a feature gets requested and then isn’t used – that’s a shame. On the other hand, that’s a pretty natural evolution process.
What’s the fundamental challenge in building software?
I see two main challenges: consistency and security.
No matter how perfectly you design the architecture or how strict your guidelines are: software development means different people collaborate, and you constantly have to compromise between performance, flexibility, and development speed. That turns into an ongoing fight for consistent product behavior.
And when it comes to security, my main concerns are: over-dependence on single-zone cloud structures, AI-generated code (if it isn’t properly reviewed), and the massive compute power that’s now available for exploit hunting with AI tools.
What I mean by that: AI can now find and exploit security vulnerabilities in software at unprecedented scale. It knows every documented weakness and can test them in seconds. That makes outdated or poorly protected systems extremely vulnerable.
You’re currently working on AI integration. What problem are you trying to solve?
I’m focusing on eliminating tedious routine processes with AI and I’m just one of many trying to turn this AI hype into something that’s actually useful.
The reality is: every manufacturer and every supplier has their own document standards and sometimes there are no standards at all. Compliance details then arrive in all kinds of formats; sometimes it feels almost handwritten. The result is chaos in the communication chain and a lot of manual work.
How does your new AI feature work in practice?
It will be a feature we can expand step by step – I’m talking about the current implementation state.
Here’s how it works: a supplier receives a compliance request as usual, with a list of parts. But instead of manually clicking checkboxes and selecting substances, they get an AI-generated suggestion they can confirm or adjust.
They can also upload any document, in any format, with any structure. As long as it contains compliance data, our software maps that data to the parts list and prepares suggestions. What’s left is: quickly check if it fits and click Confirm.
Important: we won’t allow AI to control the entire process. Final assessment, any changes, and submission remain human responsibility. And AI is only used if the user explicitly wants it.
What’s the value for customers and who benefits most?
Fewer clicks, less time spent and therefore a higher chance that the data actually makes it into your system.
Closing data gaps is good. Keeping supplier response rates high is good, sometimes even priceless. It’s hard to argue with that.
Everyone talks about AI. How much can it realistically help?
I think AI is generally overhyped, but the potential is still huge. For me, the comparison to nuclear fission fits: we use it for energy, medicine, science, and industry, but also for very dangerous things.
On the positive side, AI drastically reduces effort: physical simulations in a fraction of the time, fast answers, less manual data mining and “translating” highly complex documents into usable insights.
Roughly when will your AI solution be available?
Technically, I have a working prototype right now, but there’s a lot of testing in it until we can roll out in practice without security concerns.
A secure base version is expected to be available around April, but for the full-power version, it’s too early to say that in a serious way.
Where do you think AI is headed – specifically at Substantio?
We’ll turn AI into real value. We’ll use AI in a way that helps not in a way that quietly makes decisions behind your back.
Expect non-invasive helpers: document parsing, a chat that helps new users navigate, compliance statistics, easy adjustments of requests with minimal effort – a lot is possible.
Substantio is about data, and AI is good at that. As long as the customer explicitly allows AI to access certain fields, I don’t see hard usage limits.
What’s your biggest personal wish as a software developer?
“Don’t be evil” applied to every brain, biological or digital.
