AI at Work: What Has Changed?

Six experts, six kinds of work. A video agency founder, the CTO of an AI visibility startup, a technical writer, two product leaders and a full-stack developer answered the same questions about how AI is changing their everyday work.

To the question “Has AI changed your everyday work?” István Szabó, a senior technical writer at Qdrant, answered: “Yes, it has changed my work completely, in almost every aspect. The job is not even similar to what it was a year ago.”

Anush Mnatsakanyan, founder of Burnwe, a video agency working with tech and AI companies, receives a brief of the day’s tasks with a priority list ranked by importance at the start of each working day. AI also organises her outreach and connects inbound leads, work that used to take a lot of manual follow-up.

The other four experts are Tatiana Maes, co-founder and CTO of Far & Wide; Boris Salapa, product lead at UNITH; Nick Wüsthoff, head of product at Ark Climate; and Cyrus Maharjan, a full-stack developer at SoftRankings. What follows is what they told me: how their working days have changed, what they still keep for themselves, what they expect next and where they see things differently. At the end, you’ll find the tools they use and what they would tell someone in their field who is starting out with AI.

Most sections are marked What they do when they describe the experts’ own practice, or What they think when they share a view or a forecast.

What they have in common

I studied the six sets of answers side by side. Six themes came up more than once: some in what the experts do, others in what they think.

What they do

A person stays responsible

Deciding what matters, checking the result, the fundamentals, client relationships and the final call stay with people.

6 of 6

What they do

Less doing by hand

AI takes over repetitive work and execution: sorting email, searching docs, taking interview notes, writing code and tickets.

6 of 6

What they think

The work starts before the prompt

Shared context, a set-up process, a clearly defined ask, a precise prompt.

4 of 6

What they do

Shared context and shared assistants

Company context in one shared place, or one AI platform that everyone works in.

2 of 6

What they do

Show it, don’t describe it

Prototypes and proofs of concept instead of tickets, Figma or explaining in words.

2 of 6

What they think

Easy content can lose value

Content is easy to make with AI. Some of it is hard to read, and it can feel like less effort went in.

2 of 6

Less doing by hand

What they doNick · Cyrus · Tatiana · Anush · István · Boris

All six experts describe AI taking over some repetitive work or execution; Boris’s engineering team, for example, uses AI daily to speed up code writing, while engineers remain fully responsible for understanding and controlling what’s written. Three experts explicitly say where the time goes instead: planning, deciding, checking.

Claude prototypes instead of Figma. Prototypes instead of tickets. AI notetaker in interviews updating product discovery hypothesis automatically.

As AI agents came by, I would spend more time on research and planning and AI agents like Claude Code would just implement the architecture I planned.

I spend much less time searching, sorting, and doing repetitive work myself. I spend more time deciding what matters and what to do next.

AI has let me remove almost all of my repetitive tasks and replace them with automated processes.

You need to make more high-level plans and think more about the tools you are using. You do significantly less manual execution and need less time from SMEs.

For a smaller team without a dedicated scrum master, getting help writing ISO-compliant tickets has also saved real time for everybody.

In a working day, that looks different in each job.

István Szabó

Senior Technical Writer, Qdrant

Before asking subject-matter experts

Try the feature first

Then dig deeper with AI, and ask them a few questions to clarify or double-check.

Less manual executionMore time for verification

Nick Wüsthoff

Head of Product, Ark Climate

Before an investment decision

An end-to-end prototype

A frontend with mock data for a potential new product, used in interviews.

AI notes update the hypothesis

Anush Mnatsakanyan

Founder, Burnwe

Every morning

A brief and a priority list

Her daily tasks, ranked by importance, so she knows where to focus.

Outreach organisedInbound leads connected

Cyrus Maharjan

Full-stack developer, SoftRankings

Before any code

A plan, checked with AI

He checks the approach with AI first. Then an agent such as Claude Code implements the plan.

Production-grade?Scalable?Best practices?

Tatiana Maes

Co-founder & CTO, Far & Wide

Instead of the inbox

An email digest

Sorted into three groups. Research on potential customers and competitors runs overnight.

Respond nowCan waitUseful to know

Boris Salapa

Product Lead, UNITH

From months of notes

A structured PRD

AI helps him pull scattered notes together and cross-check them against UX principles it can cite. He reads those sources himself.

Scoped for the next release

István’s interviews with subject-matter experts, mostly developers, are one example. “I try out the feature, use AI to dig deeper, and then ask a few questions to clarify anything that’s still unclear or double-check a statement that I can’t verify any other way.” Tatiana also works with agents: “I have a sales agent, and I use agentic systems to help build, test, and improve the product.”

Show it, don’t describe it

What they doNick · Boris

Nick and Boris, who both work in product, use AI to show an idea instead of describing it.

Nick Wüsthoff works with Claude prototypes instead of Figma and tickets. His recent example is an end-to-end prototype, a frontend with mock data, for a potential new product, to use in interviews before making an investment decision. Prototyping, he says, “works a lot better than assumed a few months ago”: he used to always need an engineer in the loop for actual business logic, and that is no longer the case. Even backends have become pretty easy to build, he adds, pointing to Supabase.

Boris Salapa uses AI at UNITH for quick proofs of concept, “especially when it’s easier to show something visually than explain it in words.” His larger example started with months of notes on where video synthesis is heading, what competitors are doing and observations from his reading. AI helps him pull those notes together and cross-check them against UX principles it can find and cite, “which I then go and read myself to verify.” Combined with the real constraints of the tech stack, that process let him turn months of loose notes into a well-structured PRD for scoping the next release.

The work starts before the prompt

What they thinkTatiana · Cyrus · Boris · Anush

Four experts put the effort in before the AI does its part, each in a different place.

Tatiana still prefers frontier models, but most of the quality, she says, comes from the context: memory, connectors, plugins, shared documentation and Claude.md files. “Recreating that context every time with prompts is incredibly inefficient.” At Shell, where she also works in Digital and IT, she uses AI to prepare for meetings: “I can give it a slide deck and ask it, based on the context of everything we’ve discussed before, to tell me the five things I should pay attention to in the meeting.”

I wish I had learned to provide the best prompt to get the exact result I wanted instead of letting AI take over and hallucinate.

Think about what you actually need and spend real time defining the ask and thinking it through by yourself, instead of just hitting enter and generating more and more.

I didn’t expect AI to save me time this quickly. To be honest, setting up processes takes a bit more time at first. But once you’ve built the process and taught AI to think and act like you, you can drop a huge amount of repetitive work.

Shared context and shared assistants

What they doTatiana · Nick

Two experts describe a shared AI layer. Tatiana and her co-founder give both of their AI systems the same company context, while Nick’s colleagues all work in one platform with various assistants.

We keep a shared company context in Notion, and both of our AI systems have access to the same information. It means we are not constantly explaining context to each other or starting from zero.

We are all in Langdock and have various assistants.

Nick’s examples: “sales collects all relevant data before a sales meeting, engineering uses Claude Code for everything they build and mostly review the results”. For Tatiana, the shared context is “a bit like having an internal Copilot, but with much deeper context about how we work, what we’ve decided, and what is happening in the company.”

A person stays responsible

What they doAnush · Tatiana · Nick · Boris · István · Cyrus

However much they hand over, all six experts mention a part of the work that stays with people. For Tatiana, it is deciding what matters. Nick describes engineers as mostly reviewing what Claude Code builds.

Our engineering team uses AI for code writing daily, no doubt about that. But we follow strict rules: given the complexity of the systems we run, engineers still need to understand what’s being written in detail, own it, and control it.

For motion graphics for tech companies, explainers, and detailed product videos, AI can’t yet understand a product deeply enough to create truly personalized content, so those remain manual.

You need to make sure that the text works for a human audience, too. What didn’t change? Merging a well-rounded, approved PR feels just as exciting.

What stayed the same is having to know all the fundamental and core concepts and not just learning the syntaxes to code.

Boris also describes a correction. In the early days, people leaned on AI’s judgement over their own: if two people were debating a research topic, a third would just ask AI to settle it. That only lasted a short while. The team course-corrected, so AI is used “for brainstorming between people, not for making the final call.” With clients, he puts in the time himself, because an AI-generated response “wouldn’t carry the same care we want to deliver.” Stand-ups, general team communication and sprint retrospectives “stay exactly as they were before AI.”

What is already being replaced

What they doAnush · Nick · Tatiana

Responsibility staying with people does not mean the old tasks survive unchanged. Three answers point to that displacement. Anush also shares a view: she believes some assisting roles are no longer needed, because AI can fully support the system. Nick says prototypes with real business logic no longer always need an engineer in the loop.

In video specifically, AI-generated content has already replaced live-action shots in some of our work, and we no longer use stock imagery because AI-generated visuals are more unique.

My approach is basically to automate myself out of my own job wherever I can. That doesn’t mean doing less. It means I can do much more with the same limited time and resources.

What they expect next

What they thinkBoris · István · Nick · Anush · Tatiana · Cyrus

Before the expectations, a look back. Three of the experts put a time on how much has changed, and each looks back over a shorter stretch: Boris to when he joined UNITH, István to a year ago, and Nick to just a few months ago.

NoteThese are expectations, not facts, and they point in different directions.

As my teammate regularly says, ‘What a time to be alive!’ I couldn’t agree more. AI is amazing and scary at the same time. I’m not sure whether it’s more amazing or more scary (we’ll find out soon). I don’t think we fully understand just how transformative this technology is.

The next stage, I think, is being able to give AI a single voice instruction and have it complete any task or organize any process for you. We’re waiting to see those capabilities arrive.

I think the future of computing may actually mean smaller screens and much more happening in the background. Instead of people navigating ten different interfaces themselves, AI agents will increasingly interact with those systems for them.

Even as AI gets more capable, I can’t imagine working on a complex design project without a professional designer in the room, or an engineer maintaining a microservice, or a tech lead making the final call.

Just curious how much better they will still get in terms of production-ready code, and how it will change jobs (especially engineering & product).

Tatiana draws a conclusion for her own field: “That’s also why I think AI visibility is becoming so important.”

Boris joined UNITH years ago, in what was then a two-person team. “The sector moves so fast that five years feels like a different era.” Back then, “transformer research wasn’t yet mainstream, GPT-2 was the state of the art, and most digital humans were only 3D-rendered.” He wouldn’t dare predict what comes next. He is open to how things evolve, “but I hope it happens in a sustainable, controlled way.”

The tools they named

What they doCyrus · Boris · Nick · Tatiana · Anush

Every expert who named their tools uses Claude or other Anthropic models. Four of these experts combine models from more than one provider.

Cyrus Maharjan

Full-stack developer, SoftRankings

Coding and brainstorming

ClaudeClaude Code

Everyday questions

Gemini

Scraping sites

Grok

Analysis

GPT

Boris Salapa

Product Lead, UNITH

Models

Mostly Anthropic and Google Models

Skills and MCP connectors into

JiraSigNozFigma

Nick Wüsthoff

Head of Product, Ark Climate

Prototypes

Claude

Chat

Langdock

Note-taking

Sally.io

Tatiana Maes

Co-founder & CTO, Far & Wide

Main tools

ChatGPTClaudeClaude CodeNotion

Meeting notes

Granola

Specific workflows

Small tools she builds with AI

Anush Mnatsakanyan

Founder, Burnwe

General

ClaudeChatGPTKaironSochClayKitHubSpotFathomAsana

Video production

Adobe FireflyArtlistElevenLabsRunwayHiggsfieldGoogle VeoSeedance 2.5Nano Banana

Where they would start

I asked the experts what they would tell someone in their field who is starting to use AI. Their advice draws on their own fields and experience.

Boris adds to his advice: “Maybe that sounds old-fashioned, but I believe owning the fundamentals has to come first.” Nick ends his with:

Thank you

This is the first article in our series of expert interviews. At a time when AI is changing how we work, I think it matters to stop and reflect on what is actually changing.

A big thank you to Anush, Tatiana, Cyrus, István, Boris and Nick for their answers and for sharing their experience. Their answers inspired us so much that one article became a whole project: Shift, where a new part of AI at Work comes out every two weeks.

If you’re an expert in your field and would like to take part, or you’d like to get the next parts by email, you’ll find both on Shift. You can also write to me on LinkedIn.

And thank you for reading all the way to the end.