Cosmin Rusu

Transforming Visions into AI MVPs & Products | Founder @Duty Ventures | Ex {Google, Apple and Amazon}

Hi, I’m Cosmin, the Founder and Lead Engineer of Duty Ventures. 

After some experience as a software developer in companies such as Google, Apple, and Amazon, I decided to launch my project.

With several former colleagues and college friends, we launched our own startup specializing in software development for entrepreneurs and customers needing software development services. 

Duty Ventures is all about developing software products from start to finish, thanks to the experiences we have all acquired in the past years.

One of our most successful services is aimed at entrepreneurs and early-stage founders, for whom we provide an MVP-builder service getting their product from 0 to 1 within weeks!

However, our services are not limited to that. The teams we create do much more than just create MVPs and are able to adapt to the different technologies requested, all of this in an agile way.

Our most recurrent services are the following:
- MVP creation
- Custom Software Development
- Blockchain Development
- Mobile Development
- Checkout pages
- E-Commerce

And our most used technologies are:
- Ruby on Rails
- React
- React Native
- Java
- NodeJS


If you want to learn more about how we can help you, head over to: https://dutyventures.com


Phone: (+1) 669 200 9568 / (+40) 754 973 314
Email: cosmin@dutyventures.com

Founder & Lead software engineer

I'm leading a software development agency focused on fast iterations. Most of our clients are shipping more features, responding faster to support tickets, and have accelerated growth because of us.

We focused on HealthCare a lot lately due to the accelerated digitalization in this industry. I am a strong believer that we can make healthcare more technically advanced via pure software. 

In terms of technologies, our previous projects involved Ruby on Rails, React, React Native, Scala, Java, Kubernetes, GCP, AWS (ECS, RDS, AKS), Stripe Connect, HIPAA compliance.

Some of our previous clients: 
- Spotify
- Colgate
- Go4Ellis (incl. HIPAA compliance)
- KindBody (+1 HIPAA)
- NXT Sports
- Noom (Healthcare)
- Mochila Fulfillment
- 3Step
- Mylk Guys (YCombinator)
- NLV Productions
- SH Technologies
- One Step Software (Techstars)
- Ergodic Labs
- Med Prep
- Testing Psych
- Abundance Digital
- Tookitaki

Everyone is cutting juniors.

AI writes code. AI reviews code. AI tests code.
So why would anyone hire a junior developer in 2026?

I'll tell you why. Because I do it. Every quarter.

Here's what the "AI replaces juniors" crowd doesn't understand:

AI doesn't build culture.
AI doesn't grow into your next team lead.
AI doesn't challenge your senior engineers with fresh perspectives.
AI doesn't stay loyal for 5 years because you gave them their first real shot.

The data says junior positions dropped 20-30% in some regions.
The industry says "juniors are obsolete."

I say that's short-sighted and dangerous.

If we stop training juniors today, who leads your engineering team in 2030?

We'll face a competence gap worse than any talent shortage.
A generation of engineers who never learned to think... because a machine did it for them.

At Duty Ventures, we pair every junior with a senior + AI tools.
They learn faster. Ship sooner. And they bring something no LLM ever will:

Hunger.

The companies cutting all junior roles to save money?
They're eating their seed corn.

undefined undefined

AI-forward engineers. Embedded fast. Delivering from day one. | Founder @Duty Ventures | Ex {Google, Apple and Amazon}

Anthropic just played the smartest marketing move in AI.

While OpenAI announced ads are coming to ChatGPT, Anthropic did the opposite - they ran a Super Bowl ad saying Claude will stay ad-free.

But that's not even the best part.

They started sponsoring Google search results for developer error messages and stack traces that return zero results.

Think about it: you're debugging at 2 AM, you paste an error into Google, nothing comes up - and right there is Claude offering to help.

That's not advertising. That's solving a real problem at the exact moment of frustration.

The marketing lesson here is massive:

Don't interrupt people. Show up where they already need you.

Most companies spend millions trying to grab attention from people who don't care. Anthropic spent it reaching developers at their most desperate moment - when no other solution exists.

This is the difference between push marketing and pull marketing done right.

What's the smartest developer-targeted ad you've seen?

Stop paying for ngrok. There's a better alternative.

Meet zrok - an open-source tunneling tool built on zero-trust networking principles.

Why I switched:

1. Free subdomains - no paying for stable URLs
2. Open source - self-host it or use their free SaaS tier
3. Built on OpenZiti - actual zero-trust security, not just marketing buzzwords
4. Supports public AND private shares
5. Auto HTTPS/TLS out of the box
6. Custom domains supported

If you're a developer who regularly needs to expose local services (demos, webhooks, testing) - ngrok's free tier feels increasingly restrictive.

zrok gives you what you actually need without hitting a paywall every 5 minutes.

The setup is straightforward. The docs are solid. And because it's open source, you can inspect every line of code handling your traffic.

In a world where we're tunneling more and more traffic for dev and testing, having a free and secure option matters.

Check it out:

zrok.io

Home - zrok by NetFoundry

Unpopular opinion: most people using AI to build apps have no business deploying anything to production.

A 20-year sysadmin just quit his job this week. Not burnout. Not pay. His company offered bonuses for the best "vibe coded" app and got a tsunami of shadow IT - unsecured apps, unvetted AI agents on corporate machines, and a team of THREE sysadmins drowning under "innovation" built by people who can't explain what their own code does. When he raised security concerns? Debated. Overruled. Approved anyway. Because a VP saw a demo and got excited.

Copying ChatGPT output into production doesn't make someone an engineer. It makes them a liability with a deployment pipeline. A CTO in the thread admitted his marketing team had secretly built 11 apps - one storing customer data in a public S3 bucket. Eleven apps. Zero reviews. This is where we are now.

The predictable response: "That's just gatekeeping. Engineers are afraid AI will replace them." No. Good engineers use AI daily. The difference is they can read it, debug it, secure it, and explain it to an auditor. AI didn't lower the barrier to building software. It lowered the barrier to building dangerous software. Most organizations can't tell the difference until something breaks, leaks, or lands on a regulator's desk.

That sysadmin isn't the problem. He's the canary. Companies will keep ignoring people like him until they're explaining to a regulator why customer data ended up on the open internet because Dave from accounting "had a great idea for an app."

Stop handing out innovation awards. Start asking who's going to maintain it, secure it, and answer for it when it fails.

reddit.com

From the sysadmin community on Reddit

A friend of mine had his first meeting with a potential client in Germany.

He didn't prepare a pitch deck. Didn't try to "understand the project." Didn't walk through his company's services.

Instead, he asked one question:

"If I fly from Romania to Germany on Friday, can we meet face to face?"

They said yes instantly.

That was 5 years ago. They're still working together.

Here's what most agencies get wrong: they lead with capabilities. "We do React, we do mobile, we do cloud."

Nobody cares.

What founders actually want to know is: will this person show up when it matters?

The best business relationships I've seen didn't start with a scope of work. They started with someone proving they gave a damn before the contract was signed.

If you're choosing between a partner with a perfect proposal and one who'll get on a plane for you — pick the plane.

Hot take on vibe coding:

If frontend is mostly about calling APIs, and those APIs should be secure by default - does vibe coding make the most sense on the frontend?

Think about it. The frontend is essentially a curl wrapper. A UI layer. Less security risk.

Here's where I think things are heading:

Senior engineers will migrate towards building secure backends. Modular, composable backend "components" that can easily interact with each other.

Open-source microservices you can plug and play.

Need auth? Spin up your own Auth0-like instance. Need user-generated content? Spin up a UGC microservice - posts, comments, likes, replies, threads. The same patterns we use on LinkedIn, Facebook, X.

You reuse what exists. Then you generate your own microservice for what's specific to your product.

And once you become big enough? You open source it.

The frontend becomes the place where AI can truly shine - generating UIs, connecting to secure APIs, iterating on design fast.

The backend stays the domain of engineers who understand security, data, and system design.

Vibe coding isn't replacing developers. It's splitting the role in two.

For years, I thought the best way to create more value for clients was to charge less.

Lower rate = more value for their money. Simple math, right?

Wrong.

Here's what actually happened when I charged less:

→ Clients treated the work as disposable
→ My team felt undervalued, so they gave B-level effort
→ I attracted price-shoppers who churned the fastest
→ We couldn't invest in better processes, tooling, or talent

The moment I started charging what the work was actually worth, everything changed.

Clients were more engaged. More committed. More respectful of timelines.

My team felt ownership because the stakes matched the effort.

And the clients who stayed? They became long-term partners, not one-off projects.

If you're a founder evaluating dev partners and your main filter is "who's cheapest" — you're optimizing for the wrong thing.

The most expensive mistake isn't paying too much. It's paying too little for a team that doesn't care enough to get it right.

Heroku just entered "maintenance mode."

Except this time, it's not your app - it's the entire platform.

Salesforce announced Heroku is transitioning to a "sustaining engineering model." Translation: no new features, no new enterprise contracts, just keeping the lights on.

If you've been a developer long enough, you know what this means. Standing still in tech is the same as moving backward.

Here's the thing though - Heroku's ideas won.

Every git push deploy. Every platform that "just works." Vercel, Render, Railway, Fly - they all carry Heroku's DNA.

The lesson for anyone building software products:

Your product can pioneer a category and still lose if the parent company stops investing. Innovation without sustained commitment is just a head start for your competitors.

If you're still running on Heroku - your apps are fine for now. But start planning your next move.

The platform that taught us what great developer experience looks like deserved better stewardship.

"We spend $XX,XXX/month on LLMs" is the new "we hired 200 people in 6 months."

Both sound impressive until the board asks: what did we actually get for it?

The smartest teams I see aren't spending more on AI. They're spending less, and ACTUALLY shipping more.

Stop hiring 10 developers when 2 + Claudes can do the same job.

I know. Controversial. Especially coming from someone who runs a dev agency.

But here's what nobody in outsourcing wants to say out loud:

The old model is dying.

1. Junior positions in outsourcing dropped 20-30% last year
2. Clients are asking: "Why pay for 8 devs when 1 senior + AI does 80% of the work?"
3. McKinsey predicts a 20-30% reduction in the global outsourcing market by 2030

And yet... most agencies are still selling headcount.
Still billing by the hour.
Still pretending nothing changed.

Here's the uncomfortable truth from someone who runs an agency:

If your only value proposition is "cheap developers" - you're already dead.
You just don't know it yet.

At Duty Ventures, we made a decision 2 years ago:
Stop being a code factory. Start being a product partner.

Don't sell hours. Sell shipped features.
Don't sell headcount. Sell outcomes.

The agencies that survive 2026 won't be the cheapest.
They'll be the ones who embraced AI, specialized in complex domains, and aligned incentives with their clients.

The rest? They'll be replaced by prompts...

Marc Andreessen on how to get people to join your 3-person startup:

You have exactly 2 tools - stock options and vision.

Stock options they'll tell you aren't worth anything. So it really comes down to vision.

The best founders can explain how the world is going to look in a way that's so compelling, people leave Google and Microsoft to build it with them.

It's essentially sales. Selling to employees.

But here's the part most founders miss:

Hiring is also a self-selection process. The best companies are polarizing on purpose. They give a stark picture of what they are and what they aren't.

"We work 18-hour days - if that's not you, don't come."
"We go home at 5 every day - if that frustrates you, don't come."
"We do yoga as a company - if that's weird to you, don't come."

The founders who try to be everything to everyone end up hiring people who leave in 3 months.

The ones who are radically honest about their culture? They attract true believers.

Running a staff augmentation agency, I see this daily. The clients who know exactly what culture they're building find the right talent 10x faster.

So I guess Claude is OOO today

https://status.claude.com/

So if AI will write code, and we treat that as a commodity, and as a pure blackbox inputs / outputs type of thing - what's next? Maybe we fully automate QA? Maybe we fully automate a simulation on how that code would interact with actual users? So we "simulate" a real world usage of an app / a website over time to find bugs / issues? Then, what next?

There are two types of early-stage technical decisions:

Type A: Decisions that get your product in front of users.
Type B: Decisions that prepare for users you don't have yet.

Most founding teams spend 80% of their time on Type B.

Kubernetes, microservices, auto-scaling, blue-green deployments... all Type B.

A $10 VPS with your app deployed on it? Type A.

And here's the uncomfortable math:

3,000 users paying $30/mo = $1M ARR.

A single server handles that without breaking a sweat.

Every week you spend on Type B before you have paying users is a week you're not learning whether anyone even wants what you're building.

Ship boring infrastructure.
Build interesting products.

Hot take: If your outsourcing partner can't deliver the same result with fewer people... they're the WRONG partner.

Let me explain.

I've been on both sides of this.

As a founder of a dev agency, I've seen clients pay for:
- 6 developers sitting in standups for 3 hours/day
- 4 QA engineers manually testing what AI can do in minutes
- A project manager whose only job was "translating" between client and devs

That's not a team. That's a payroll.

The smartest clients I work with in 2026 don't ask:
"How many developers will you assign?"

They ask:
"How fast can you ship this feature, and what does it cost?"

That's the shift:

OLD MODEL: Pay per seat. Vendor profits when project takes longer.
NEW MODEL: Pay per outcome. Vendor profits when they ship faster.

We moved to milestone-based pricing at Duty Ventures years ago.
Result? Our incentives match the client's: ship fast, ship right, move on.

When AI makes a 3-person team as productive as a 10-person team was in 2023 - the billing model has to change.

If your vendor is still charging you by the hour in 2026, ask yourself:

Whose interest are they serving - yours or theirs?

The outsourcing industry is worth $618 billion.
Most of it is still billed by the hour.
That's a $618 billion misalignment of incentives.

Someone replicated Monday.com's core features in 60 minutes using AI.

Cost: $15.

Monday's stock took a hit. People panicked.

But here's the thing nobody's talking about:

Code has been getting cheaper for 20 years. AI just made the trend impossible to ignore.

If your entire business defensibility is "we built an app," you were already vulnerable. AI just made it obvious.

The real moat was never the code. It's:

→ Deep understanding of a specific customer problem
→ Product decisions that reflect real usage, not assumptions
→ An architecture that actually scales when users show up
→ A team that iterates based on data, not vibes

I see founders still spending months planning features that an AI tool could scaffold in an afternoon. And I see Silicon Valley founders shipping AI-generated MVPs that fall apart the moment real users touch them.

Both are missing the point.

The winners will be the ones who combine fast tooling with sharp product thinking.

Code is a commodity. Product judgment is not.

taplio

The Battle

Cosmin Rusu

Munir Machmud Ali

Followers

Posts

Engagements

Likes

Performance ScoreCosmin Rusu's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Cosmin Rusu

Munir Machmud Ali

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Cosmin Rusu's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreCosmin Rusu's LinkedIn Performance