Thriving in Engineering

Roman Nikolaev is Head of Technology at Cambri, a Nordic research tech company whose AI-first platform helps global brands like Nestlé and Coca-Cola test product concepts before launch. He's also the author of High Impact Engineering, a Substack newsletter on practical, field-tested guidance for engineering leaders. Roman has been building and scaling software teams for over 20 years — and Cambri started working with AI and ML well before the ChatGPT moment made it mainstream.

In this conversation, we discuss:
  1. Why 65% AI tooling adoption only produced 10% more PR throughput — and what the GetDx data actually tells us
  2. The "building on a swamp" problem: why applying AI on top of a broken development process gives you almost no upside
  3. Whether the software engineering profession is going away — and the car mechanic analogy that reframes the question
  4. How Cambry is running semi-automated agent workflows right now, with agents picking up tickets, asking questions, and raising PRs
  5. MCP servers worth paying attention to: Figma, Chrome/Chromium, and what Roman would actually build with them
  6. Why every engineer is now effectively a manager — and the two constraints that stop anyone from being a one-person company
Roman's Details:
High Impact Engineering newsletter - https://highimpactengineering.substack.com/
LinkedIn: https://www.linkedin.com/in/roman-nikolaev-43465019/

What is Thriving in Engineering?

Software engineering isn't just strategy and sprint plans. It's ambiguity, pressure, trade-offs, and people.

Thriving In Engineering helps you navigate all of it.

In the newsletter, Alex Ponomarev distills 25 years of engineering leadership into practical frameworks to help you lead with confidence, stay calm under pressure, and build thriving teams.

The podcast goes deeper, talking to engineers, leaders, and builders across the industry about what AI transformation actually looks like in practice, past the headlines and hype. One guest at a time.

Hello and welcome to Thriving in Engineering podcast. In each episode, I talk to people across the

software industry about AI and software development. Uh we're not talking about

the headlines, the hype, or the doommongering. Instead, we focus on honest conversations about

how AI is actually showing up in day-to-day work. What's working, what's not, and what others can

learn from it. My guest today is Roman Nikolaev. He's head of technology at Cambri, a Nordic

research tech company whose AI first platform helps global brands like Nestle and Coca-Cola

test product concepts before launch. He's also the author of High Impact Engineering, a Substack

newsletter focused on practical field- tested guidance for engineering leaders. So, make sure to

check it out and subscribe. Roman has many years of experience building and scaling software teams

at Cambri. They started working with AI way before it became mainstream. Roman has a reputation

for writing about engineering leadership with honesty and depth. And like a lot of people in

the industry, he got here the way many of us did, falling in love with computers as a kid. So Roman,

we both started out making games in basic um and tinkering with DX Spectrum and uh you know other

early technology. Can you tell us how that turned into a career in engineering? Yeah, it's a yeah

very nice question. So I really like to reminisce on the old times. So started uh working with

computers when I was maybe 9 years old. So my my parents bought me my first first computer. It was

very bulky. So at times machine uh which had only 32 kilobytes of memory. After that I got the ZX

Spectrum the clone of British computer and then I got PC. So I started programming maybe when I was

12. Then gradually I moved from basic to C++ and then after that learned other different languages

and I graduated from uh applied mathematics and informatics in university and after that started

my professional professional career. I've been uh working as a professional developer

and manager for about 20 years now. Starting from 2003 maybe, but that's time flies. Yeah,

that's impressive. Quick question. Can you say that the fact that your parents introduced you to

to computers in general even though you know these bulky big machines that were very different from

what we have now? Is it fair to say that this is why you love software engineering and that's what

set you up for success in the career? Yes, I think that's was a pivotal moment for me. So I've been

always interested in mathematics and this kind of more focused introversial things rather than

you know playing football in teams. I prefer to spend time like reading and tinkering with some

mechanisms and Oh yeah I computer was a very very good computer was very good match for me. So yeah

and then uh went into into the programming and uh it was clear for me from age of maybe 12 or 13

what I want to be a professional programmer in the future. When my career developed I first went into

uh like started as a junior software engineer then worked as a senior and then suddenly I ended up as

a lead developer and then as a manager. So it was a an interesting turn of events when I was just

offered this position when when I didn't have any firsthand experience leading people and I needed

to learn these things on the fly and after a year or so I got promoted to the head of technology at

Camry which is maybe uh similar to more familiar title CTO position. the title was different that

the scope of responsibility is pretty much the same. So leading the whole technical organization

so it's a different level of responsibility, different level of different different viewpoint

on the business and technology. I think it was it has been very interesting journey so far.

It sounds like the memory of the transition from the senior engineer to the leadership role

is still very fresh. Looking back, what do you think about this decision about moving

to leadership? I think I I wish I would have made it earlier. So I think it's very rewarding

to work not only with technology but working with people more. The best part of working

in these leadership roles is that you have bigger leverage well one of the best things but the thing

what you can accomplish as a software engineer is limited to your personal productivity. But

when you work as a manager you can enable others to do great things and then as an organization you

can create something what is impossible to create by yourself. Another side of this is that working

with people is actually gratifying in the sense that when you see how people develop, how I grow,

you also get this kind of feeling of satisfaction from the job well done which is which lasts much

longer than uh satisfaction from feedback or implemented features. So it stays with me for

a long time. Yeah. I I can totally relate to that and to the I wish I did it earlier part for sure.

How do you feel like well let me rephrase this. You went through the full journey from the junior

engineer to the senior engineer to the leadership role and now you know you have a lot of impact

a lot of leverage as you said and now the whole industry is changing. So you went through a lot of

changes and transformations along your career. Now everything is changes uh changing. How do you feel

about what's coming and what's changing because clearly I I'm judging based on my experience. I

see that even leadership changes in this new AI era. What do you think about that? What is your

perception? Yeah it's very broad question but let me answer it from my perspective. So I think that

for engineering leaders the main change that has happened is that everything accelerated before the

the industry we are in is not static. It has been changing like we talked about this bulky computers

with 32 kilobytes of memory and then we went from this to supercomputers in your pocket in the in

just 20 years. So the industry is changing very fast but now the rate of change is even higher.

So how it personally impacts me is that I need to constantly rethink what rethink the things what

worked before. So let's say we let's take our typical software development life cycle. So we

have requirements then requirements are checked by development team then there is a development work

then there is code review then after that there is a validation then if validation passes then there

is deployment and then the cycle continues from the beginning. So how this is changing now for

instance code review there is a lot of talk now in the industry how to treat code reviews should code

reviews be done by humans should they be done automatically how to streamline code reviews

another part of the traditional development life cycle is working with requirements we used to use

Figma prototypes but now you can create a live application within minutes with code or cursor.

You don't need to draw in Figma for days. You can just have a working version what you can

demonstrate. You don't need PRD anymore. So these changes are coming very quickly and tooling is

changing, ways of thinking is changing and then we all as a leaders we always need to rethink how

we can get how we can keep up with the changing technology landscape. Yeah, absolutely. Here is an

interesting question for you. We do need I I again I think we're super aligned on how we're seeing

this change happening as leaders. I I think our day-to-day work changes also in a way where it's

not just meetings anymore. It's not just you know guidance architecture but we get to participate

in the software development um process and the the way we rethink processes is by constantly looking

at what's out there what's the next hot thing this week. Trying something ourselves, not just

assigning people on our team like you you go and experiment with this, you do this. No, now we have

to lead by example. That's my feeling. Can you say that um like your day-to-day work changed because

of the tooling and because of this rapid change? Do you feel the pressure of leading differently?

I think my day-to-day work like in in substance it hasn't changed much. So my

essence of my work is still about coordinating people, coordinating teams, providing clarity

from the top of the organization to the individual contributors and then getting the status from the

individual contributors to the to go to market teams to the rest of the leadership team.

So in that sense it remained the same but what is changing is the rate at which we are rethinking

how we work. So we cannot basically stop now that next month there will be another another

technology there will be another tool and then we need to start thinking are we on the cutting

edge of the innovation or are we behind so it puts pressure on me personally I always need

to be on checking the news checking the latest technology trends I saw a meme today on on X

I go to sleep, I wake up, there's new Claude release. That's how [laughter] that's how I

feel. Yeah. Yeah. Exactly. Yeah. There is a strong fear of missing out. And this fear of

missing out I'm sure you're not the one in your company who feels it. How do you how do

you think as leaders we can help our teams with this fear of missing out? Is there a way to help?

I think I don't know people are different but what I can say from for myself what helps me to stay

sane in this rapidly changing world is thinking about what in the end of the day we are producing

are we producing code or are we producing positive outcomes for our business or or society in general

if we are a mission driven company for example what do we create and what is the relation

between the tooling and the outcome of our work. So tooling is changing tooling has been changing

all the time while I have been in this software engineering industry but what we are doing and

what we are providing to our users with is the most important part. So for people who are working

individual contributors role I think it's helpful mindset shift or maybe not a shift maybe for many

it's a natural way of thinking but uh thinking about uh what we do from the standpoint of the

change we create rather than the output. Mhm. And the tool becomes a tool. Tools tool is a tool. A

tool is a tool. Yeah. Some some people think that you know the whole profession software engineering

profession is going to go away because hey Claude is so great at writing code. Codex is so great

at writing code and then they can write the PRD they can write tests let's put them together to

you know to scope to decide what needs to be built to to build and then let's build the feedback loop

and it actually works in in some cases. I'm curious what you think about the trajectory where

everything is moving. Is it reasonable to even um in your opinion is it reasonable to even imagine

that AI is gonna do everything and we won't need engineers? Yeah, there are maybe two things. Well,

I've been I I was very skeptical about AI agents when they was introduced. I was very skeptical

what it's going to work. But two years later now you have cloud agent we have cursor what

can perform task autonomously. So it's really a big shift from where it started. So I cannot tell

for sure what will happen in one year or two years time. So but where we are standing now and looking

at the trajectory I would say that the software engineering profession is not going away. I can

give you the an analogy. So if you if you have a yeah if you have a if your car break you can

fix your car yourself. So there are instructions on the internet. You can order the tools. You can

order spare parts and fix it yourself. And maybe if you need to do it twice, it will be cheaper

than going to a workshop. But most of the people they still prefer this job done by professionals

and they don't become a professional car mechanics. So that's how I see it for the software

engineering profession as well. So of course if you read manuals, you understand how systems work,

you understand underlying technology, you can probably create your own Asana or you can create

your own mini sales force. But should you for many people I think the answer is no. Um thank you for

this analogy. That's exactly what I'm telling both to my engineers and you know to other leaders who

ask me the same question. It's almost like you you use the car repair analogy. I usually use the

uh washing machine analogy like you can wash your clothes and fold them yourselves but you still buy

a washing machine. You can run from A to B. You're perfectly capable of doing that. there's a reason

why you're driving a car that needs repair and so on. So yeah, again I think we're 100% aligned.

I have this idea that uh one one shift that is happening is that everyone essentially becomes

a leader now. Only that like the only difference is that you as a head of engineering you probably

work with multiple teams of humans, but every engineer now is essentially working with a team

of AI engineers. You you have QA agent that uses browser and mobile app testing. You have another

agent that works with the database, another writes, ML models and so on. What what do you

think about this idea? Do you agree or not? And sorry, I I have broad questions. Um and

um and if the answer is yes, do you think that the skills that we create while managing human

teams are also applicable for the agent teams? Like basically can the readers of your newsletter

apply the same things to managing agents? Yeah, this is a broad question indeed. I think there

are two constraints that prevent one person to be like a one person company. It's it's possible

but in the larger organization there are two main constraints. I would say the first constraint is

that there is a context business context limit. So software engineers are not fully aware of or

we are not that deep in the business relations. We don't know all customer pain points. We don't know

how clients work. We don't know their workflows. So there should be someone still who interacts

with clients directly. If you have a very small business where you are a CEO, software engineer,

head of marketing, etc. then you can do that. But as you start scaling, it's not possible to

perform by one person. So you still need someone to interact with clients and understand their

needs. This is the first limitation and the second limitation. Teams of agents work very

well on well-contained small code. But when system starts to grow, you need to understand intricately

how it's um how it's structured, what are possible dependencies between the components,

what are side effects of this particular change you are doing. So it's when you have a system

running in production, it's very easy to break everything by just modifying one line of code

and in the end it would be a person who sign offs on the change who is responsible for the for the

for the for the change. It's not an bunch of agents. So we still would need the person in

the loop who would understand and sign off on the change. Keeping this complex systems together this

multicomponent system requires interaction between different team members what are specialists in

their parts of the system. I think that's kind of puts the limitation on the one main army idea.

Yeah. I 100% agree, but it probably works until a certain point just like you could get pretty far

away as a one person army pre-AI, you can just get way further now. But what about engineers

for example in in your organization as an example managing a bunch of agents working on spec not

trying to have AI roam freely and rebuild half of the system overnight on its own with open claw

Ralph Wiggum Loop or another version of the same approach but maybe just just an engineer assigning

tasks to an agent reviewing the output just like you know just like an engineering manager does

because that's what we do as well, right? We think through technical strategy. We assign things to

different teams or to to different people. We look we check back with them about what they're

doing will review the results or some metrics. That kind of works the same with with AI agents.

That's at least how I feel. Yeah, we are currently experimenting with semi-automatic workflows. So we

have an agent running or we would have an agent running on the virtual hardware and then we would

set it up this way that it checks our ticketing system. If there is a task but is marked as a

task for an agent, it can pick it up uh ask additional questions from a task owner and

then after that start the implementation, finish the implementation, write a pull request and then

that's where the job ends at least for for a while until it's reviewed. If there is no feedback what

the pull requests need to be reward then it is being merged by a human. So that's kind of

we can call it a semi-automated workflow where we have a person who is checking the work and signing

off. Yeah that is pretty advanced. I have been talking to someone just last week and they are

still using like copy pasting code to chat GPT and I think most of the teams are still like that. I

would say Camry is pretty advanced at adopting the technology. Yeah, we are now looking looking into

literally this week we plan to run an experiment. So part of the team is having offsite in Lapland.

So skiing and working at the same time. So part of the evening curriculum is to try out working with

working with open claw in another aspect of this is interesting aspect of working with this agent

is data security so that it doesn't break or it doesn't do what it's not supposed to do. That's

an interesting challenge as well. Yeah, we are currently looking into this. Let's see. Maybe it's

if the experiment is successful, we would probably have it running in a couple of weeks. Yeah, that

sounds like a top of- mind thing for for a lot of teams uh these days. Here is a question for you.

At Cambri, you started working with AI and ML way before it became a thing, right? way before the

ChatGPT moment when everyone realized that, oh, we should probably get on the AI train. As someone,

you know, spending that many time with AI and ML, what do you think about like the pace of

acceleration? Is it because of technology or it is it because just more people are finding

more applications for you know for LLMs for neural networks for ML. Yeah, I think it's both. So

there first of all there is a lot of interest and there is a lot of investment both from the effort

perspective but also from from the financial perspective. So it's a huge market but uh also

there is a there is a lot of interest and the tech technology itself is and the technology itself is

very interesting I would say. So the capabilities of technologies multiplied by the financial and

time investment create this kind of effect that it's advancing much faster than it's it was like

5 years ago for example. Yeah, that makes total sense. I wanted to ask you for advice. Not for

myself but for maybe engineering managers who are just uh thinking about AI transformation. I know

it's a buzzword but it's still happening whether we want it or not. Or maybe engineers who are

working in companies that are not as far ahead as Cambri. What kind of advice you can give to them

to to stay relevant and you know to somehow manage to to do your day job because we have our we have

to as software engineers we have to deliver value today, right? We have to have this impact but also

learn new things given the current pace of advancements. I think there are two things

here. First thing is that of course you need to keep up with the technology landscape. So if you

ask me today I would say just install Claude Code and play with it. The second thing if we

talk about the productivity perspective if you build your AI initiative on a swamp so to say

that it will it will drown. So you need to look at your software development life cycle. You need to

look at your at how you work with the business requirements, how you work with another other

stakeholders on in the company because if you just apply AI on top of unstable foundation, you will

have very very limited upside. It doesn't matter if you code quickly. If every other decision

takes a week then it doesn't matter. So from this kind of productivity perspective, what managers

can optimize in their teams is collaboration, visibility, team culture, software development

life cycle, question their rituals, question their habits, how they did things before. And then in

the broader context look at the organizational context how handovers are organized between the

product team. If there is a product team with the engineering team how business receives

requirements can we shorten this loop somehow. The biggest impact lies in the in shortening the loop

from getting requirements or getting understanding the user pain to uh getting this in production. AI

can help to shorten the journey but the impact of it will be will be very limited if other parts of

the journey are very long and complicated that that is very interesting. It it it sounds like

your advice is like fundamental because often times people recommend they they go deep into

tools. You just mentioned Claude Code and you know and the the fundamentals. Do you think

tools and specific techniques are not as important as the mindset like thinking more about the the

foundation and building from the bottom up? Or maybe I misread this. I think it also depends

on the size of the organization. If you are a a single person who runs their own business, so

the whole business context and customer relations are in your head. So tools become more important.

But if you work in a large organization, then the other parts of it, the other parts of the process

maybe are more important. So yeah, I think the concrete techniques way are important to collapse

the software development part, but if everything else takes still the same amount of time,

it will not matter. I by the way read recent study done by GetDx. It's a company that specializes in

developer experience and development metrics. So we run a study across 40 companies of different

sizes and what we found found out is that with 65% of AI tooling adoption we only have 10% of

PR throughput increase. So basically what what we are seeing here that the AI adoption is quite

high but the upside is very limited. So, and the comments that way the developers give us

basically that everything else is so slow that it doesn't really matter if we in four days or

three. That makes total sense yeah, because the speed of generating code was never an issue in

in my perspective. But looking at the software development piece, are there any tools or maybe

techniques you find specifically useful that you can recommend? You mentioned cloud code maybe.

Yeah, I think that like using MCP servers with MCP servers connections can be can be useful. So for

instance I personally tried out Figma connection. So it helped me to quickly translate the exact UI

requirements into the code when creating some kind of small front end heavy application.

Another MCP server I'm considering is connecting to Chrome or Chromium. I think that would be very

useful for running automated regression checks for instance in the early stage of development. If you

don't want to automate it with test automation framework at this phase but you want to do some

quick checking for instance that could be useful. Also you can theoretically build full workflows

where you feed requirements AI iterates on the requirements until the workflow fully works in the

UI. But from my perspective working in business to business environment with very high value

contracts it's too risky. So basically we would have a black box that we don't fully understand

how it works. So I wouldn't probably go to this fully autonomous development yet, but I think

it's an interesting idea and for smaller teams and maybe not mission critical applications, it could

be an interesting avenue to research. Yeah, I I support that. I would recommend agent browser by

Vercel. It's a CLI tool that uses Chrome dev tools under the hood and it connects without MCP server.

And what we've been doing and we're a smaller team and we're doing this for as a semi-automated

process to augment QA and software engineering. But basically what works is you ask Claude Code

to write a PRD and you ask it to write acceptance criteria. And of course you review them and make

sure they make sense. Then it implements and you tell it okay now use agent browser just go to the

browser and check that everything works according to acceptance criteria and of course it tells you

that everything works you go and check something doesn't as usual happens with humans as well right

but once you're happy with the results you can tell it okay now go write playwright scripts

based on browser testing you've done at the end of the session session. You have your implementation

and you have super fast and efficient playwright scripts to you know Yeah. Well, I I think our time

is running out today. I am extremely grateful for you joining our podcast today. Thank you so much.

I hope you enjoyed it and I think you gave amazing nuggets of wisdom to our viewers and listeners

and yeah, thank you for that. Thank you, Alex, for having me. It was a pleasure. Thank you.