WE 
Based on a talk by Boris Strandjev, CEO of Qualifast, at ISTA 2025
At ISTA's 15th anniversary edition, Boris Strandjev returned to a theme he knows well from running Qualifast: what actually happens to software quality once AI enters the picture. Rather than cheerleading or panicking, he laid out a grounded, role-by-role view of where AI genuinely helps — and where it quietly erodes the things that make products good.
Setting the Context
Boris framed his argument around building digital products from scratch — his own home turf, and, as he noted, the space most shaped by current VC ecosystem strategies. But the reasoning, he was clear, travels well beyond startups.
He also drew a distinction worth keeping in mind throughout: there are two separate lines of LLM-based AI effects on software projects. One is AI as an engine for business in products — AI baked into the product itself. The other is AI as a facilitator for the development process — how teams build. His talk focused on the second.
The State of Affairs
AI is genuinely everywhere now — from conversational bots to art generation, vaccine research, and running the machinery of software projects. It's been around a long time, but LLMs and AI agents have made it radically accessible. Naturally, some organizations have taken the hype to the extreme.
Boris placed AI on the classic trend adoption curve: we've just passed the peak of inflated expectations, on the way down into the trough of disillusionment. He's convinced, though, that LLM-based AI will ultimately settle into a real, lasting trend rather than fade as a temporary hype cycle.
That leads to what he called the AI frontier: AI adoption should never be the objective in itself. You can improve your outputs, but only by applying AI sensibly — and different situations call for different mixes of AI and human effort. Maximum AI usage doesn't mean maximum ROI; past a certain point, over-reliance on AI actually drags returns down.
AI's Effect on IT, Concretely
Boris pointed to three shifts he's seeing directly:
- The market now demands shorter delivery times and lower budgets.
- Product versions are increasingly treated the way proof-of-concepts used to be — throwaway.
- Seemingly everyone now considers themselves an expert in everything.
The common thread is speed. It's arguably become the most important KPI in software development today — dominating conference talks and roadmaps alike. Boris offered an unpopular opinion here: cutting costs is really a byproduct of speed, not the goal, and the urge to adopt AI is pushing teams toward haste based on a flawed cause-and-effect assumption.
His counter-principle, and the whole philosophy behind Qualifast: "Be smart in balancing quality and speed." Make haste slowly.
Quality in the AI Era: Where the Opportunities Are
Rather than treating AI purely as a threat to quality, Boris argued for using the very same tool that created the market pressure to compensate for it. He broke the opportunities down by role: product people, developers, and QAs.
For Product People
- Vibe-coded POCs — Useful only for proof of concept. You might not be able to fix the problems that come up, and it's great for demonstrating ideas — but don't get too excited, this isn't a production-ready app. The quality gain isn't in the code itself; it's in reduced dependencies on other people to validate an idea.
- Requirements structure — One of the areas with the most room to experiment, especially since requirements are usually scarce at startups. AI can draft a first version of requirements straight from a conversation, track how grooming discussions change those requirements over time, and generate variants from wireframes or mockups.
- Communication facilitation — An image is worth a thousand words: mockups and wireframes can now be created with verbal commands, and conversations can be transcribed and summarized automatically.
- Improved insights — Better analytics and user-behavior insight, more efficient research, user interviews at scale, and predictive analytics.
The summary for product people: you can finally know what you're building, get yourself a thought transcriber and librarian, become more independent and self-sufficient, and make better-founded product decisions.
For Developers
Boris was direct about the right way to use AI tools:
Do:
- Use AI to shortcut boring, low-complexity work
- Use it for time-consuming tasks you'd otherwise avoid
- Prepare your architecture and design concepts before letting AI fill in the details
- Keep your local setup tightly integrated with AI models — minimal friction is essential
Don't:
- Use AI for code you don't know how to validate
- Treat AI-facilitated coding as a substitute for actually learning the underlying concepts and technologies
On pair programming with AI specifically, Boris compared it directly to human-human pairing. AI pairing wins clearly on velocity and on exposing more known features, but it tends to produce worse architecture, doesn't reliably reduce bugs, and falls short on team communication and knowledge sharing — things human pairing still does better.
Other developer-facing gains: AI helps with code inspection — orienting in unfamiliar code, recognizing good and bad patterns, suggesting improvements right in the IDE, and even spotting AI-generated code fragments. It also nudges developers who are typically reluctant to write tests toward writing more of them — though those tests still need review to make sure they're sensible and actually add value. And it can improve requirement traceability, the often-weak link between product, dev, and QA, by helping create and attribute tasks — something Boris's team has built custom tooling around.
His summary for developers: stop avoiding these opportunities, mind the general rules of usage, don't forget to balance speed against quality — and no, AI can't be the friend you never had.
For QA
Boris opened with the market's misconceptions about QA in the AI era: the "throwaway effect" ("why should we test at all"), the push for speed causing business owners to cut corners including QA, and the "expert in everything" mentality of "I'll do my own testing." QA teams are being forced to rethink how they present their value.
Where AI genuinely helps:
- Functional testing — Blackbox testing remains the way to go. LLMs can construct test selectors (guided or unguided), canvas and image assertions get much easier, and responsive UI testing benefits from added semantic understanding. LLM-generated tests do suffer the same architecture and reusability problems as AI-generated code — but that's fixable with proper architectural guidance.
- Performance and load testing — A growing area of concern for AI-built apps, but business owners are more open to investing here. Tests can mimic real user behavior more closely, partly because AI is good at analyzing that behavior in the first place.
- Security and penetration testing — Probably the one quality practice actually thriving in the AI era, and definitely the topic of greatest concern for business owners. Unguided blackbox tests benefit significantly from AI, which can also help detect adversarial user behavior.
- Vibe testing — An emerging concept distinct from vibe coding: rather than just generating automation from words, it focuses on testing sentiment and feel — smoothness, naturalness, satisfaction.
- Report and defect tracking — Report automation is easier and reports more understandable; defects can be tracked through verbal or written instructions and LLMs, with context like prerequisites and reproduction steps extracted automatically. AI is a strong aid in initial defect analysis too.
QA's summary: it's getting less attention even as its importance grows; automation is already second nature to many QAs; verifying AI-proposed solutions is the right mindset to adopt; and there are new bridges available to both product and development.
The Conclusion: Quality, the AI Way
Boris closed with a straightforward call to action:
- Nothing is outside your capabilities.
- Always keep an eye on what can be done better.
- Never compromise on basic principles — Festina Lente, make haste slowly.
- Opportunities! Opportunities! Opportunities!
The throughline of the whole talk is simple: AI doesn't have to be the thing that erodes quality. Used deliberately — with the frontier of sensible adoption always in mind, not maximum adoption for its own sake — it can be the thing that helps every role in a product team hold the line on quality while still moving fast.
Boris Strandjev is CEO of Qualifast. You can find him on LinkedIn.
