NTK26 Recap: Three Sessions on AI-Assisted Software Delivery

At NT Conference 2026, no one needed convincing to use AI. The audience asked a far more practical question: how do we turn it into a tool that helps our teams and organizations with real work?

Ana Roje Ivančić and Ognjen Bajić presented three sessions, each exploring a different part of AI-assisted software delivery.

Ognjen Bajić presenting at the NT Conference 2026 podium, with Ana Roje Ivančić at the speaker desk

Three sessions, one theme: AI-assisted delivery with people in the loop.

Trust the Code You Didn't Write: Leading Delivery When AI Outpaces Review

What evidence does your team need before AI-generated code can reach production? We presented four layers of automated evidence and control for reviewing and releasing AI-generated code:

BDD scenarios define the expected behavior.
Guardrails-as-code enforce the key rules.
Independent agents review every change.
Staged releases and observability reveal issues in production.

Our live demo started with business intent: AI wrote and reviewed the change, and a controlled rollout took it to production. People and organizations remain accountable throughout — they set policy, handle exceptions, accept risk, and approve the release.

Agent Skills in Practice: Governing a Product Backlog with GitHub Copilot

What happens when every team member starts using AI to act on your product backlog, but guides it differently? Curated and shared Agent Skills turn the team's refinement standards, review steps, and approval points into a consistent, repeatable way of working.

Starting from a product specification, we created an Epic–Feature–PBI backlog with refined descriptions and test-ready acceptance criteria. Agent Skills guided the sequence from refinement to review and approval, Azure DevOps MCP provided live context and actions, and people reviewed, steered, and approved every backlog update. From individual prompts to a shared team capability — that is the shift we demonstrated live.

Ana Roje Ivančić driving the live demo at NT Conference 2026

Every concept backed by a live demo — real tools, real context, people approving every step.

BDD Reloaded: AI-Assisted Collaboration from Business Intent to Executable Tests

One business rule can hide many assumptions, boundaries, and missing examples. Starting from business rules expressed in plain English, and with AI guided by practice-tested Agent Skills, we uncovered ambiguity, expanded concrete examples, shaped Gherkin scenarios, and connected them to automated verification.

Business intent stays visible throughout the flow: Product, Business, QA, and Development review the same examples and strengthen them with their domain expertise — AI assists the conversation without replacing it.

Do We Actually Use What We Show?

The same question kept coming up from the audience — and we do! That is why we talk about what works for us: how AI needs real context, clear rules, and checks that matter; how to handle unreliable output and boundaries; and which decisions require human judgment.

Our main takeaway from NTK26: the first AI hype is fading. Now comes the less glamorous, more important part — doing the work.

The "Trust the Code You Didn't Write" session was also presented at CollabDays Zagreb 2026 at Algebra Bernays University in Zagreb on September 25.

Put It into Practice

If you'd like to see any of these workflows in your own environment, book a short demo or invite us to run a hands-on workshop — from AI-powered backlog management to BDD testing with AI-assisted workflows. We'll tailor it to your setup and keep it practical.