AiThority Interview with Matej Bukovinski, Chief Technology Officer at Nutrient

Matej Bukovinski, Chief Technology Officer at Nutrient chats about the evolving skills modern software developers need in an AI driven coding ecosystem:

_______

Can you tell us about Nutrient’s latest AI assistant and how it’s changing the game for end users?

For years, documents, especially ones in portable formats such a PDF, have been the most static part of any digital workflow. You can automate almost everything else, but the moment a PDF contract enters the picture, someone has to stop and do manual work. That’s the problem we set out to solve.

We recently shipped a major update to Nutrient AI Assistant that adds an autonomous document editing agent alongside the existing chat agent. The chat agent lets users ask questions, get summaries, and translate content — essentially turning any document into a conversation. The editing agent goes further: you describe what you need in natural language — “extract all liability caps from this contract,” “redact patient identifiers before sharing with the research team” — and it plans, executes, and adapts across multiple steps using our purpose-built document tools for rendering, extraction, form operations, annotation, and redaction.

What makes this different from bolting a chatbot onto a PDF viewer is that the agent operates under configurable governance. Developers can define which actions run autonomously, which require user confirmation, and which are prohibited entirely. You can inject domain-specific skills — your company’s compliance rules, pricing data, validation logic — so the agent doesn’t just work on documents, it works the way your organization works. It runs across web and mobile, and connects to whatever LLM provider you choose — in the cloud or on premise.

Also Read: AIThority Interview With Rohit Agarwal, Founder & CEO of Portkey

As developers open up to AI assistants supporting their work, how will the future scope of work for a typical B2B SaaS development team start to look like?

The honest answer is that the team of the near future looks less like a pool of coders and more like an engineering leadership group. When agentic AI handles more of the implementation — writing code, running tests, fixing failures — the valuable human work shifts to defining outcomes, designing systems, and reviewing results. At Nutrient, we’ve mandated the use of coding agents across the team: we only write code manually as a fallback. The engineers who’ve leaned in are operating at a fundamentally different altitude, thinking in systems and outcomes rather than lines and files.

For B2B SaaS teams specifically, I think we’ll see smaller teams producing more, with engineering judgment becoming the scarce resource rather than engineering hours. Planning, architecture, and the ability to orchestrate AI effectively will define team composition more than raw headcount. The best performers in the age of AI are not necessarily going to be the same as before it.

What skills matter most in a time of AI-driven coding?

Three things stand out. First, communication — and I mean that literally as a technical skill. This is not about prompt engineering. As AI advances, being clever with your wording matters less and less. What does still matter is the skill of being able to describe a problem with enough precision to get a useful result, without overspecifying in ways that constrain the solution space. This in many ways separates great engineers from the rest right now.

Second, systems thinking. When AI handles more implementation, the scarce resource isn’t code volume — it’s architecture, contextual knowledge, and the judgment that only comes from having built and maintained real systems at scale. Those skills matter more when the tools around you are more powerful, not less.

Third, critical evaluation. Developers who use AI as a black box, accepting output uncritically, create a different kind of problem than the one AI was supposed to solve. The ability to recognize when the tool is wrong, and to push back, is crucial.

Can you talk about some of the challenges you foresee with developers using AI without the right protocols? What would you share as top tips and best practices?

The biggest risk is invisible: teams that speed up code production without equally speeding up code review, testing, and release processes just move the bottleneck. You end up with a firehose of pull requests that nobody is reviewing or testing properly, and quality drops.

The typical advice someone might give here is to not let AI-generated code skip human review. In fact rather double down on it. However that brings us right back to the bottlenecks described above and to not fully leveraging the speedups AI promises.

Instead, make your review more effective by augmenting human reviews with AI tooling. From automated LLM based reviews and security checks on CI to using agents steered by humans to dig into specific areas the reviewer knows can be problematic. This is what we practice at Nutrient. Every stage of the development lifecycle needs to get faster together. The same applies to QA testing, where agents can do exploratory testing next to humans or releases where agents can take over polishing release notes.

A few thoughts on the future of AI?

We’re not incrementally improving how software gets built — we’re dismantling its fundamental assumptions. The notion that building software requires large, specialized teams working in lockstep over months-long cycles is becoming obsolete. A single engineer with the right agentic setup can now achieve what used to require a squad or engineers. This fundamentally changes how engineering organizations are built and what skills are being prioritized in hiring. Everyone who’s not investing is bettering themselves on this front is risking becoming obsolete.

The software industry is a frontrunner here, but it’s clear to me that it’s only a matter of time before the same hits other intellectual work. As agentic software progresses, the specific industries that software serves get disrupted one by one. In that sense it’s very obvious why AI labs are investing in software development tooling first, to gradually extend their reach. People who want a space in this new world need to treat this inevitability as an opportunity, not a threat and learn ways how they can leverage the new tooling to their advantage, as it’s becoming available.

Some top AI innovations and AI innovators you’d like to shout out?

Anthropic’s Claude Code and OpenAI’s Codex have genuinely changed what “agentic coding” means in practice — these aren’t assistants anymore, they’re collaborators that execute across entire codebases. Cursor deserves credit for proving that deep IDE integration and model flexibility is what professional developers actually want. On the open-source side, projects like PI (the minimal coding agent) and OpenClaw (for agents that go well beyond code) are pushing boundaries in important ways.

Beyond coding tools, I’d highlight what’s happening in document AI more broadly. The shift from static document processing to agentic document workflows — where AI doesn’t just read documents but acts on them — is a transformation that’s still underappreciated. And the infrastructure layer matters too: skills, MCP (Model Context Protocol), command line tools are quietly becoming the connective tissue that lets AI agents interact with real business data, stored in documents or otherwise. Moving forward, that’s going to be foundational for any business, not just software shops, which is why Nutrient will keep investing in this area.

Also Read: ​​AI-Driven Risk Intelligence: How FIs Are Predicting Systemic Shocks

[To share your insights with us, please write to psen@itechseries.com]