AI-powered software development tools – an interview with Head of Frontend

The CTO vs Standing Quo collection research how CTOs problem the present state of affairs at their firm to push it towards a brand new peak … or to put it aside from doom.

“Nice coders who perceive prompting can most likely do 10x the rate of somebody who’s simply studying to make use of it.”

You received’t get a working Fb clone out of AI with a single immediate – or many prompts, for that matter.

However for open-minded software program builders who’re prepared to study, AI can enrich their careers in astonishing methods.

We mentioned this subject with Dimitris Kountanis, the Head of Frontend at Native Groups. He shared his experience-based perspective on:

• how AI permits builders to generate extra worth for the end-user,
• why profitable tech groups are shifting their focus from software program engineering to product growth,
• what’s the secret to boosting creativity and velocity for builders,
• the place to get real-time updates on every thing happening with AI.

About Dimitris & Native Groups

Native Groups’ imaginative and prescient

Piotr Urbas: Hello Dimitris, thanks for becoming a member of us! Congratulations on an wonderful yr at Native Groups – your crew grew to 270+ staff and reached £10M ARR. What do you assume made this doable?

Dimitris Kountanis: Hi there, Piotr! To be sincere, issues are transferring so quick that I nonetheless haven’t absolutely processed these milestones.

At first of 2024, we had been about 150 individuals, so our headcount nearly doubled in only one yr. 

We had a terrific foundational crew that was capable of construct quick, a sensible advertising and marketing strategy, and a very good market positioning. Beginning in Balkan international locations and increasing to different locations in a while turned out to be a terrific choice.

How essential was the position of the frontend crew in reaching these milestones?

Our frontend crew is break up. One half works on advertising and marketing, the opposite on our software. Though they complement one another, the on a regular basis job calls for are completely completely different.

The first focus is advertising and marketing – that’s the place the frontend crew has the strongest influence on the underside line. It’s about creating content material, constructing a weblog, creating touchdown pages, and creating all of the brochures for promoting.

The applying facet’s aim is to make an interface that works nicely and enhances the backend.

AI enters the stage

I do know that you just’re captivated with AI. What instruments do you and your builders use?

We built-in AI instruments into our customary workflow in November 2024. The reason being easy: it’s unattainable to code as quick as we would like with out AI. Plus, the insights and steering it supplies are actually useful.

It started with me encouraging everybody to obtain Cursor. I advised all frontend engineers to make use of the free trial for every week, see what occurs, and report again to me in every week.

Everyone instantly obtained higher and quicker. Each process went smoother. At that time, it turned apparent that you just couldn’t ignore AI instruments for coding. It simply didn’t make sense to work with out them anymore.

We switched to GitHub Copilot for a second, however now we’re absolutely migrating to Cursor as a result of it appears to be quicker.

I’m not a software program engineer, however I not too long ago constructed a Swift app with Cursor and ChatGPT, so I absolutely agree that these instruments are useful.

Did you calculate how a lot quicker your crew has been coding with AI?

I haven’t crunched the numbers but, nevertheless it seems like we’ve not less than doubled our velocity because of AI instruments. One thing that used to take a complete day to construct, I can now create in an hour – together with the Asana process, Figma file, and all different components of the method.

With the suitable setup and context, Cursor can end nearly all of a process with a single immediate. All that’s left is to refine it manually and ensure it really works as supposed. We’ve got a design system in place and Storybook integration, so builders can decide the constructing blocks and create precisely what they want on the frontend.

When AI eliminates a lot guide work, engineers can deal with logic and be extra inventive. Would you agree?

Undoubtedly. Used accurately, AI turns into a pressure multiplier. With our inner API documentation, we’d should learn it and join the dots ourselves to construct capabilities. Now, we simply feed our docs to Cursor and generate the calls we’d like. We simply have to grasp the stream and interior workings of the code. AI permits us to focus extra on the precise person expertise.

I preserve telling my crew that we’re shifting from software program engineering to product engineering. We will lastly deal with buyer issues and person expertise as an alternative of spending our days determining the right way to wrangle an API.

I’m beginning to assume that for those who’re good sufficient, you may construct something you need with AI. However let’s not get carried away – my subsequent query is about drawbacks.

For instance, some individuals say that AI limits creativity. Have you ever seen this occur amongst frontend devs?

I believe it boosts creativity. It takes away many of the boring stuff, the mundane duties that no person ever actually loved. You’ve extra time to consider the precise web page or view you’re constructing, and essential points like its efficiency and UX.

AI brings you nearer to your prospects. You may deal with what actually issues. There’s much less interested by code and extra interested by the product and the person. Product work is extra inventive than software program engineering. 

Let me provide you with one attention-grabbing instance: one in every of our new options features a massive web page for bulk invites. 

Bulk actions are at all times a wrestle for the frontend as a result of you might want to do loads of stuff in a very small window. Think about importing a spreadsheet with a thousand rows. On each single row, you may have empty fields that you could be must pop up an error for. However you may’t simply drown the person in a thousand errors without delay –  that might be horrible UX. What to do?

To unravel this, you might want to get inventive and discover a good strategy to make it work for the person. With AI, you may present a pleasant template for filling within the knowledge or use AI to fill within the knowledge for them.

The AI toolset

Let’s dig deeper into the precise AI instruments. How do you utilize them responsibly? How a lot hand-holding is concerned in coding with AI?

The guide half principally includes overseeing the AI’s work and testing the code on the finish. Nevertheless, I might by no means use the agent mode in our codebase. It principally means you let the AI do no matter it desires. I don’t suggest utilizing it.

It’s important to ensure that AI is aware of what you’re constructing, what the top consequence ought to seem like, what your code and formatting requirements are, and even the way you need it to answer to your prompts.

A very powerful tip I may give to individuals who need to use AI coding instruments responsibly is to at all times set guidelines for AI. It’s important to give it restraints. Our course of for coding with AI is principally this: write a small immediate, focus on the output, and repeat.

On the finish of the day, it’s nonetheless your job to grasp what you construct.

That makes loads of sense. What about further instruments for error detection and overview, like CodeRabbit? Are you able to depend on them an excessive amount of?

We don’t use these but. UI makes up nearly all of frontend duties. AI isn’t but subtle sufficient to grasp how and why you need to type your format.

However AI can perceive the API or any flows on that facet of issues, which is why I’d prefer to work with an AI assistant for merge request creation. I imagine it may assist us keep on with greatest practices. Fairly than deciding to merge code or not, it may maybe present a abstract of the request with any points that require trying into.

What about code optimization? This instinctively feels just like the most secure space for utilizing AI instruments in the long run. Would you agree with that?

Earlier than you optimize something, it’s a must to first determine what precisely is suboptimal in your code. With AI, you are able to do efficiency profiling quicker.

Debugging manually is time-consuming. Now, you may simply ask AI to do profiling and provide you with all the info you want. You are available on the finish and deal with the right way to optimize a selected loop or repair a faulty perform.

You may’t optimize one thing you may’t see. AI-powered debugging and efficiency profiling provide you with that context. Nevertheless, you want critical prompting expertise to interrupt your code into small components and discover what wants fixing. 

The extra context you give, the higher reply you’ll get, appropriate?

The most effective recommendation I preserve listening to from AI energy customers is to keep away from massive duties. No matter job you might want to do, break it down into small, easy steps. 

If you give AI an excessive amount of context, it can not think about every thing, and the output will disappoint you.

I observed the very same factor when working with Cursor. I began out pondering that I may one-shot a complete app with an excellent lengthy immediate. I shortly realized that it really works significantly better with small duties.

Precisely. You may’t simply open Cursor and say, “Construct Fb.” 

I’m curious in regards to the logistics of switching to Cursor as a crew. How do you implement greatest practices for AI utilization?

My strategy is to set guidelines for Cursor and thoroughly clarify them to the devs.

We focus on these guidelines internally and ensure everybody understands why they’re vital – why immediate X works, why add particular configs, why specify the packages we’re utilizing, and so forth. This helps keep away from incorrect replies, messy code, and unacceptable errors like altering the programming language.

I’ve seen individuals complain that AI provides them Python code for a JavaScript codebase, and it doesn’t shock me. In the event that they haven’t specified the principles and the context, the AI might be simply going for the simplest resolution doable.

Guidelines are a precedence. If I outline guidelines for Cursor about the right way to name APIs and add a constraint like “we use Tailwind,” the crew can have a better time working with AI and understanding its outputs.

Is AI already a necessary a part of the developer’s toolkit? 

Sure. I’m very assured about that. When you nonetheless don’t use a code editor built-in with AI, you merely can’t compete with people who do. As soon as you start utilizing these instruments, the speedy distinction is night time and day – nice coders who perceive prompting can most likely do 10x the rate of somebody who’s simply studying to make use of it.

Some corporations already refuse to rent devs who don’t use AI. Working within the business and figuring out nothing about it’s a bit like being illiterate. You may’t ignore it, even for those who’re a senior with years of expertise. A wise junior who’s an influence person of AI can already do what you may, most likely much more.

I believe the one companies that also wrestle to undertake AI code editors are these with essentially the most delicate knowledge. As soon as they determine the right way to use AI instruments with out the necessity to join with third-party servers, that constraint will go away.

AI & distant work

Let’s change gears for a second and speak about distant work. I perceive that Native Groups loves it. 

Do you assume it’s tougher to implement AI utilization requirements on distant groups?

Distant work is troublesome to handle at scale. The larger you develop, the extra layers there are in your group. The construction of a 30-person firm is way completely different than that of a 300-person one, and the stream of knowledge will get very difficult.

It’s important to think about completely different time zones and work hours. This requires a tradition the place data can unfold round freely and for processes that be sure that essential insights at all times attain the suitable particular person. 

It’s important to implement some requirements in several methods. I may give you an instance that’s not associated to AI – we’ve made it obligatory to make use of Prettier with Husky for pre-commit hooks. Every thing must be prettified earlier than it’s pushed to our codebase. Our setup is such that any commit that doesn’t adjust to that will probably be blocked routinely.

However in the case of AI, I believe it’s pretty easy: the usual we require is to make use of it as a lot as doable and at all times test if every thing is ok earlier than pushing it. That’s it.

Basic recommendation

What’s your recommendation to a developer who’s simply beginning to implement AI instruments, maybe simply making an attempt out Copilot or Cursor? How ought to they strategy these instruments to really develop into higher builders?

AI is evolving too quick to actually sustain with it proper now. Greatest practices change earlier than you may absolutely perceive them. You must determine a strategy to keep on prime of all of it.

Other than that, you want a foundational understanding of what you’re utilizing – how prompting works, how brokers work, and many others. Then, you additionally must preserve testing issues, like going by means of completely different fashions and seeing how they reply. Perceive why some individuals want Anthropic’s Sonnet over OpenAI’s GPT4o and why Anthropic’s fashions are higher for code than OpenAI’s for complicated reasoning. 

My subsequent piece of recommendation is to make use of AI to transition to product engineering, even for those who’re a backend developer. AI provides you the additional time you might want to lastly deal with the top person and never simply the intricacies of software program engineering. 

And what would you advise managers who need to foster a pro-AI tradition?

If Peter Thiel’s traditional “Zero to One,” it’s simple to see that we’re within the midst of a zero-to-one second – which means that the business won’t ever be the identical because it was earlier than AI.

As somebody who nearly missed its departure, I can confidently say that you just don’t need to be late for the AI practice. If AI wasn’t an enormous a part of our system proper now, I might need ignored it till subsequent yr or so – and I imagine that might have been too late to catch up.

As a supervisor, you play an essential half in protecting your builders within the loop. Present time and sources to your crew to attempt issues out, empower your self to do the identical, and watch what occurs. 

When you get an concept, do a tiny proof of idea. It won’t work out, however you’ll positively study so much within the course of. And if it does work out, it’ll inspire the entire crew to maintain making an attempt much more.

I believe that’s the most important factor I’d inform my youthful self – check extra issues, preserve making an attempt new tech, mess around with it, and cease being so afraid.

Assets

It’s time for our customary closing query: what ought to managers learn, watch, or hearken to in the event that they need to study extra about AI?

Scorching take alert: I don’t like studying engineering books or listening to engineering podcasts. I simply scroll Twitter or X.

All of the consultants and essential AI gamers are there like Guillermo Rauch or Sam Altman, asserting stuff and discussing the most recent developments. It’s the quickest, most dependable strategy to keep updated. 

Positive, books and podcasts have plenty of worth, particularly in the case of studying fundamentals. Nevertheless, they’re not up to date in actual time – Twitter is. You see bulletins as quickly as they’re posted. 

Plus, you will discover loads of stuff to copy-paste and instantly check out. Any individual who’s already mastered Cursor may put up a listing with guidelines that you may check with out having to provide you with your individual from scratch.

Take into account this: we had a 4-hour-long AI workshop at Native Groups someday round November. The primary three hours consisted of greater than 2,000 slides, all with data from Twitter bulletins. Anybody who wasn’t utilizing Twitter needed to course of all of it proper there after which. In the meantime, I barely had to concentrate as a result of I noticed nearly all of it whereas it was taking place. 

What’s subsequent? 3 ideas for CTOs to extend velocity with AI

Dimitris believes that, when used accurately, AI is usually a pressure multiplier, a creativity enhance, and a means for engineers to spend extra time including worth for the person as an alternative of typing out code. To attain this growth course of enchancment, he recommends the next:

• be sure that AI is aware of precisely what you’re constructing – what the top consequence ought to seem like, your code requirements, formatting, and even the way you need it to answer to your immediate,
• break down massive jobs into small duties – you may describe them in brief prompts and get appropriate outputs of excellent code high quality out of AI extra simply,
• keep open-minded – be prepared to check out new instruments shortly as an alternative of letting your self be slowed down by anxiousness.

Use Dimitris’ recommendation to extend the rate of your software program growth course of with AI, however keep in mind that it’s nonetheless your crew’s job to grasp what they’re constructing.