Product and engineering partner for founders building with AI
Build fast with AI. Build the right product with us.
Golux turns ideas, Lovable prototypes and AI-built MVPs into products people can depend on — product strategy, design and engineering under one accountable person. Whatever stage you are at, including nothing but an idea.
For founders building with Lovable, Cursor, Claude Code, Replit, Bolt, v0 — and for founders who have not written a line yet.
Start where you are
Where are you in your product journey?
All three of these are normal places to start, and the first one is not a lesser version of the others. What changes between them is what the first month looks like, not whether we can help.
I have an idea
You know what you want to build and nothing exists yet
The most expensive mistakes on this list get made here, before anything is built — the wrong user, a data model that cannot carry the second feature, three months spent on the part nobody needed. This is the cheapest possible moment to have someone argue with you.
- What the product actually is: the user, the one job, the metric
- The smallest version that proves it, and what to leave out
- User flows and the screens that carry the risk
- Architecture and the data model, decided before they are expensive
- Built — by us, by your team, or by you with an AI tool and our review
- Launched, with the things that break on launch day already handled
I am already building
You started with Lovable, Cursor or Claude Code and hit the wall
The tool got you further than anyone expected and then stopped being the bottleneck. What is slow now is deciding what to build next, and the features that need someone who has shipped this before — payments, permissions, anything with an edge case.
- What to build next, and what to stop building
- Design that holds together past screen nineteen
- The features the tool cannot finish: payments, permissions, integrations
- Row-level security, auth, and the data model underneath them
- Architecture review before the codebase becomes unchangeable
- The bugs that only appear with real data
My product is live
You have users, and now reliability is a business problem
Everything that was fine at ten users is a decision at a thousand. This is also where the audit is most useful, because the things that hurt here are visible from the outside before they are visible from inside.
- Production readiness: secrets, backups, monitoring, ownership
- Performance — usually one query, not the framework
- Security and access control that survives an enterprise questionnaire
- Technical debt, ranked by what it is actually costing you
- Integrations that fail quietly and should not
- Scaling, and knowing which part will break first
Not sure which one you are? Start with the free audit if something exists, or the review call if it does not. Both are free and neither needs you to have decided anything.
The actual problem
AI made it easier to build software. It did not make it easier to build a good product.
Almost nobody is stuck because they cannot generate more code. They are stuck on questions no generator answers, and every one of them is a decision rather than a task:
- What should we build next, and what should we stop building?
- Is this actually good UX, or does it just look like software?
- Is the architecture sound, or is it fine until the second feature?
- Should we rebuild this or improve it?
- Why are users dropping off on the screen that tests fine?
- Can this handle real customers, and how would we know before they arrive?
- Is our database actually refusing the wrong reader?
- How should this integration behave when the other side is down?
That is the work. The code is the easy half now, and it has been for about a year.
What this is
Not another freelancer. A product partner.
Product thinking, design and engineering in one arrangement, so you are not the integration layer between four suppliers who each did their part correctly. We help you decide what to build, and then help you ship it.
To be exact about what you are buying: Golux is Nikola Golubović, accountable for every decision, plus collaborators he has worked with for years — a designer for production design, additional engineers for capacity — named before they start and never a rotating bench. That is smaller than an agency, and it is why these arrangements are built to run for years: the same person is on your call in month nine, and by then he knows your data model better than a new hire would after a quarter.
Who you would actually be working withWhat we do
Four things, and they run in this order
- Strategy
- Design
- Build
- Improve
Product strategy
The decisions that make everything after them cheaper.
- Product direction and positioning
- MVP definition and scope
- Roadmap and prioritisation
- User journeys
- Product review
- What not to build
UX and UI design
Decided first, drawn second. The other order is how products end up looking like software.
- UX flows and wireframes
- Interface design
- A design system your AI tool can follow
- Onboarding
- The states nobody designs: empty, loading, error
- Conversion work
Engineering
Built to be changed by whoever comes next, including you.
- Frontend and backend
- Database and data model
- APIs and integrations
- Authentication and permissions
- Payments
- AI features that do something
Production and growth
The half that decides whether the product survives its own customers.
- Launch support
- Monitoring and alerting
- Debugging what only happens in production
- Performance
- Analytics that answer a question
- Technical debt and scaling
If you are building with AI
Built with Lovable? Keep using it.
You do not need to throw a product away because a generator wrote it. In most cases improving what exists is both cheaper and faster than rebuilding, and we will tell you which case you are in before you spend anything. What we deal with is the part the tool was never going to do:
- Supabase and the data model under it
- Row-level security that actually refuses the wrong reader
- Authentication, sessions and the login loop
- Stripe, webhooks, and access that is granted by the payment rather than the redirect
- APIs and third-party integrations
- Transactional email that arrives
- GitHub, deployment and an environment you own
- Security, secrets and backups you have restored at least once
- Performance with real data rather than demo data
- Responsive UX past the happy path
Also building with Cursor, Claude Code, Replit, Bolt or v0? Same answer. The tool changes what the code looks like, not what production demands of it.
What changes
Where founders usually are
- An idea, or a prototype that demos well
- Features chosen by whoever had the thought
- A UI that is different on every screen
- Generated code nobody has read
- Bugs that only appear with real data
- No roadmap, or one nobody believes
- Nobody accountable for whether it works
Where this gets you
- A product definition you can say no with
- A prioritised roadmap, and a reason behind each item
- Consistent UX, including the states nobody designs
- Engineering built to be changed
- Integrations that fail loudly instead of quietly
- Enough confidence to send real traffic at it
- One person accountable, and documentation so you are not trapped
From something that works to something people can depend on.
How it works
A product team without building one yourself
01
Show us what you are building
A URL, a repository, a Figma file, or a paragraph describing an idea that does not exist yet. All four are a fine starting point.
02
We review it together
Forty-five minutes on the product, the design and the technical side, and what the biggest opportunity actually is. Free, and you leave with the answer whether or not you continue.
03
Agree how we work
Which arrangement fits, what the first quarter covers, and the terms — all settled on the call rather than guessed from a page. The recommendation comes from what you need, not from what we would rather sell.
04
We build
One prioritised thing at a time, visible to you in Golux Club — tickets, approvals, decisions and files in one record you can read without asking.
05
Improve continuously
Ship, measure, learn, change. The documentation is written as we go, because the intended ending is that you hire a team and we hand over.
How to work together
One rate card, whatever stage you are at
These are ongoing arrangements, not a job you buy once. Hours are stated on every tier because they are a cap, and a cap is the only honest way one person can promise availability. Moving between tiers is an email, not a renegotiation.
Start here
Product review — Free
A call on what you are building and what it needs. If something exists, run the automated audit first and we will go through the findings together. You leave with the biggest product, design and technical opportunity named, whether or not there is a second conversation.
Advisory
$2,500/month
6 hours a month
Senior judgement in writing, for a founder who is building and needs someone to argue with.
- 6 hours a month — ask as often as you like
- Written answers within one working day
- Product, design and architecture decisions reviewed before they are expensive
- Review of what your team or your AI tool shipped
Most chosen
Product partner
$6,000/month
16 hours a month
A weekly call and real work, for a product being actively shaped.
- 16 hours a month across calls and work
- A weekly call on what is being built and why
- Product definition, flows and the risky screens
- Architecture, data model and the security decisions
- The features your tool cannot finish, built
Embedded
$12,000/month
36 hours a month
The design and engineering function itself, for a launch or a rebuild.
- 36 hours a month across direction and production
- Production design by a named collaborator
- Hands-on engineering on the parts that carry the most risk
- Launch, monitoring and the first weeks after
Startup partner
$20,000/month
60 hours a month
The whole technical side, under one accountable person. One company at a time.
- 60 hours a month across product, design and engineering
- Two calls a week, plus availability between them
- Hiring: we write the roles and run the technical rounds
- Investor and enterprise due diligence
A single piece of work rather than a month: $300–$400 an hour, 10 hours minimum, scoped and agreed before anything starts.
We take a small number of companies at a time so the answers stay fast. If there is no room this quarter you will be told, not queued.
These arrangements are for companies that can fund senior engineering properly — funded startups, profitable businesses, and teams with a deal or a raise on the line. If you are pre-revenue and self-funding, the free audit and the guides on this site are genuinely yours to use, and that is the honest answer rather than a discount.
The alternatives
Why founders choose this over hiring separately
All three of these are reasonable answers and two of them are the right answer some of the time. The difference is mostly about who holds the whole picture.
| Freelancers | In-house team | Golux | |
|---|---|---|---|
| Product strategy | Sometimes | Yes | Yes |
| UX and UI | Sometimes | Yes | Yes |
| Engineering | Yes | Yes | Yes |
| Architecture decisions | Varies | Yes | Yes |
| Capacity you can change monthly | Yes | No | Yes |
| Time to start | Days | A quarter | A week |
| Hiring and management overhead | You | High | None |
| One person accountable for the outcome | No | The CTO you hired | Yes |
Hiring is the right answer eventually, and when you get there we write the roles and run the technical rounds. A good freelancer is the right answer for a defined piece of work. This is for the gap between them, which is most of the first two years.
Honest qualification
This is a good fit if
- You have an idea you are serious about and can fund
- You built an MVP with Lovable, Bolt, v0, Cursor or Claude Code
- Your prototype works and needs to become a product
- You have users and need the engineering to hold
- You are tired of coordinating several freelancers
- You want product thinking, not only more code
- You want one technical partner for the next couple of years, not a series of suppliers
It is probably not a fit if
- You are looking for the cheapest developer available
- You want a small one-off job — an hour here, a tweak there. Run the free audit and read the fix guides instead; they are genuinely free and they cover most of it
- You want a supplier for a single deliverable rather than a partner for the next couple of years
- You want unlimited development for a small monthly fee
- You want someone to execute a spec without questioning it
The second list is not a sales technique. Saying it here costs a few enquiries and saves both of us a call that was never going to work.
Why take the advice
We do not only advise founders. We build and operate our own products.
Fifteen products where this team made the product calls, drew the flows, chose the architecture and then lived with the consequences for years. That is a different teacher from a portfolio: every bad decision came back as a support ticket or a schema we had to change.
- 15
- products built and still operated by this team
- 15+
- years shipping production software
- 208
- AI-built apps audited, and published
- 1
- person accountable, collaborators named
LifeCopilot
AI planning platform — goals, habits and routines
Merlin
A platform for building AI products
Server Inspector
Infrastructure and website monitoring SaaS
RetreatBooker
A booking marketplace
MojaFarma
Digital farm management
CoinMarketNotifier
Portfolio monitoring and alerts
Questions
Before you write to us
- I only have an idea. Is that too early?
- No, and it is the cheapest moment to get the decisions right — the user, the one job the product does, the smallest version that proves it, and a data model that can carry the second feature. What matters is not how far along you are but whether you can fund the build once it is decided.
- Do you only work with Lovable?
- No. Lovable is one of the tools we see most, along with Bolt, v0, Cursor, Claude Code and Replit, and we also work on ordinary codebases with no AI involved. The tool changes what the code looks like, not what production asks of it.
- Can you work with the app I already have?
- Usually, and usually that is the better answer. A rebuild is occasionally right — when the data model cannot carry where the product is going — and we will say so plainly rather than quoting for one. The free audit is the fastest way to find out which case you are in.
- Do you build from scratch?
- Yes: product definition, design, architecture, build and launch. The difference from an agency is that the first weeks are spent deciding what to build, and the most valuable ones are where the answer is 'buy it', 'wait', or 'that is two weeks, not two months'.
- Are you an agency or an outsourcing company?
- Neither. The model is a fractional product and engineering partner: one accountable person, named collaborators for design and capacity, month to month. An agency's revenue depends on the build, which makes it structurally unable to talk you out of one.
- How long do engagements last?
- These are built to run for years, not weeks. Knowing a product properly takes a couple of months, and the value compounds after that — the decisions get faster because the person making them already knows your data model, your users and what you tried last spring. Terms are agreed on the call, and they assume both sides are in it for the long run.
- Who owns the code and the product?
- You do, from the first commit — your repositories, your accounts, your Figma. No proprietary layer, no licence, and nothing that requires us to stay. The documentation exists so your next team does not need us.
- What if I cannot afford the monthly arrangements?
- Then you get the honest answer rather than a discount. The audit, the benchmark and the fix guides on this site are free and are written to be usable without hiring anyone — that is not a teaser, it is most of what a good first month would have told you anyway.
The next step
You do not need another development vendor. You need someone who cares whether the product works.
Tell us what you are building. If you have a URL we will run the audit before the call so we are looking at something real rather than at a description of it.
One person reads every one of these, usually the same day. Both of us are deciding whether this is worth a couple of years, so if it is not a fit you will be told, and told why.
