By Paul GosnellSearch for the best AI product studio and most of what comes back is a listicle written by one of the studios on the list, ranking itself first. This guide is published by a studio too, so it tries to be useful in a different way. It compares the five routes to a production AI MVP (an AI product studio, a traditional development agency, a freelancer, an in-house hire and a no-code AI app builder) on cost, speed and risk. It names five real UK studios with one fair line each. And it gives you the questions that separate a team that ships from a team that demos.
It is written for founders, product leads and operators who have an AI product idea, or an AI prototype that works on a laptop, and need it running for real users within weeks rather than quarters. It also answers a narrower question buyers keep asking: who is the right partner for a two-week concept followed by an eight-week pilot?
Every fact about another company comes from that company's own website, checked in October 2026, and the sources are listed at the end. Where a figure is our own estimate, the guide says so.
What Counts as a Production AI MVP?
A production AI MVP is the smallest version of an AI product that real users can rely on. It signs them in, keeps their data safe, gives consistent answers, costs a known amount to run, and tells you when it goes wrong. A demo does one of those things on a good day.
| Question | Demo or prototype | Production AI MVP |
|---|---|---|
| Who uses it? | The founder and a few friendly testers | Real users who did not build it |
| Sign-in and data | Shared login or none, sample data | Proper accounts, permissions, real data stored safely |
| AI behaviour | Looks right in the examples you tried | Scored against a test set of real examples before every release |
| Running cost | Unknown | Measured per task and per user |
| When it fails | You find out from a user | Logs and alerts tell you first, and a human can step in |
| Ownership | Lives in one person's account | Code, prompts, test sets and accounts handed over to you |
Most of the cost and time in an AI MVP sits in the gap between those two columns. That gap is what you are hiring someone to close, so judge every option below on how it closes it, not on how good its demo looks.
What Are Your Options for Building an AI MVP?
There are five realistic routes. The table gives the short version. Each route then gets its own section with the cost, the speed and the risks.
| Route | Typical cost for an MVP | Time to first production release | Best for | Biggest risk |
|---|---|---|---|---|
| AI product studio | Fixed fee. Published entry points from £5,000; larger studios start at £10,000 to £30,000 | 1 to 4 weeks for one core workflow; months for a larger product | Speed with production standards | Small teams carry key-person risk, and most claims are self-reported |
| Traditional development agency | £80,000 or more for eight weeks of a four-person team (worked example below) | 2 to 4 months, often after a discovery phase | Large, well-specified builds | Process overhead, and AI experience varies by team |
| Freelancer | £500 to £1,200 a day | Can start in 1 to 2 weeks; pace depends on one person | A tightly scoped build with an experienced owner on your side | Single point of failure, no second reviewer |
| In-house hire | £60,000 to £130,000 salary plus 25 to 30% on-costs | Months before the first line of code | Long-term capability once the product has proven itself | Slow to start, and expensive if the MVP fails |
| No-code AI app builder | Free tiers; Bolt Pro is $25 a month | Hours to a demo | Testing an idea and putting it in front of users | Sign-in, data, security and scale usually need a rebuild |
The cost and time ranges are p0stman's estimates, built from published prices and UK rates cited in each section below. They are not a market survey. Use them to compare routes, not to budget a specific project.
AI product studio
An AI product studio is a small, senior team that takes an AI product from idea to production and treats the AI as the core of the product, not a feature bolted on at the end. Most work to a fixed price and agree dates before they start.
Cost. Studios that publish prices are rare, which is a signal in itself. p0stman publishes a focused sprint from £5,000 over 1 to 3 weeks and a production build from £8,000 over 2 to 4 weeks. Studio Graphene's enquiry form starts its budget bands at £10,000 to £30,000. &above does not publish prices.
Speed. A focused MVP with one core workflow can reach production in 1 to 4 weeks once the studio has access to your data and systems. Larger products take longer. &above says its MVP for the home-selling platform Upstix was live in three months.
Risks.
- Key-person risk. The smaller and more senior the team, the more depends on one or two people. Ask who covers when they are away.
- Self-reported claims. Studio websites, ours included, are marketing. Ask to use something they have shipped.
- Ownership. Check that the code, prompts, test sets and cloud accounts end up in your name, not theirs.
Traditional development agency
A full-service agency brings project managers, designers, developers and testers. That structure suits big, well-specified builds. It slows down small, uncertain ones, because an AI MVP usually changes shape in week two, when the first real outputs come back and the team learns what the model can and cannot do.
Cost. Agencies usually price by team and by time. As a worked example, four people for eight weeks at £500 to £1,200 a day each comes to £80,000 to £192,000 before any discovery phase. Those day rates are the UK contractor ranges from our guide to hiring an AI developer in the UK. This is arithmetic, not a quote from any agency.
Speed. Two to four months is common once a discovery phase is included. Some agencies run discovery for several weeks before writing code.
Risks.
- Discovery phases that end in documents rather than working software.
- AI work handed to a separate team who joins late.
- Senior people in the pitch, junior people on the build.
Freelancer
A senior freelance AI engineer can be the fastest and cheapest route for a tightly scoped build, and many are excellent. The best ones have shipped production AI before and will tell you what not to build.
Cost. UK AI contractors charge roughly £500 to £1,200 a day, depending on seniority and specialism. Medium and large companies must also decide each contractor's IR35 status, a rule that has applied since April 2021.
Speed. Freelancers can often start within one to two weeks. Pace after that depends entirely on one person's availability.
Risks.
- Single point of failure. Holidays, illness, a better-paid contract and support load all land on one person.
- No second reviewer. Nobody checks the security model or the test set except the person who wrote them.
- You become the project manager. Somebody on your side has to own scope, priorities and acceptance.
In-house hire
Hiring your own AI engineers builds capability you keep. It is the right move once the product has proven it deserves a permanent team. It is rarely the fastest way to find out whether the product works.
Cost. UK AI developers cost roughly £60,000 to £130,000 a year in base salary, depending on seniority. Add 25 to 30% for on-costs: employer National Insurance at 15% on earnings above £5,000 a year (the rate since April 2025), pension contributions, equipment and recruitment fees.
Speed. A senior search takes time, and a good candidate often has a notice period of one to three months on top. The first line of code can be a quarter away.
Risks.
- You are hiring before you know the product works.
- One AI hire on their own faces the same single-point-of-failure risk as a freelancer, with a longer exit.
- A wrong hire is slow and costly to unwind.
No-code AI app builders
Tools such as Lovable, Bolt, Replit and v0 turn a description into a working interface in minutes. They are the cheapest way to test an idea, and many good MVPs start here. Fewer finish here.
Cost. Bolt has a free plan, a Pro plan at $25 a month and a Teams plan at $30 per member a month. Lovable starts free, with daily build credits, and has paid Pro and Business plans.
Speed. Hours to a clickable demo.
Risks. These tools are strongest at screens and weakest at what production needs: sign-in edge cases, a sound data model, database security rules, background jobs, testing of AI output and cost control. We cover the specific limits of each tool in our guides to Lovable's limitations and Bolt's limitations, and the wider problem in why a prototype is not a product.
A common and sensible path: prove the idea in a no-code builder, then bring in a studio or a senior freelancer to rebuild the parts that have to be reliable. You keep what you learned and lose very little time.
Which AI Product Studios Belong on a Shortlist?
These five studios all build AI products for clients, are based in the UK or have a UK office, and publish enough on their own websites to check. They are listed alphabetically, not ranked. p0stman wrote this guide and is one of the five.
&above (London)
An AI consultancy and product development agency in Borough, London, eight years old, that builds with embedded squads of strategists, designers and engineers. Its site cites an agent system used by 5,000 Google Cloud sellers a month, Tesco ad approvals cut from four weeks to days, and an MVP for Upstix live in three months. It starts product work with a proof of concept on sample data.
Best fit: scale-ups and enterprises that want an embedded squad rather than a single builder.
Makeitfuture (London, Paris and Cluj-Napoca)
An AI agent and automation agency founded in Cluj-Napoca, Romania, with offices in London and Paris and a team of about 30. It is a Make.com Platinum Partner, says it is ISO 27001 certified, and cites more than 500 clients. Its work centres on workflow automation and integrations on Make, n8n, Zapier and similar platforms.
Best fit: MVPs that are really an automation or agent workflow across existing tools, rather than a new product.
p0stman (Norwich)
A senior, AI-native product studio led by its founder, Paul Gosnell, that publishes fixed prices: a focused sprint from £5,000 over 1 to 3 weeks, a production build from £8,000 over 2 to 4 weeks, and a fractional CTO from £3,000 a month, with scope and price back within 48 hours. Work includes MillionPlus, a UK luxury asset finance broker rebuilt as a multi-agent operation across eight or more channels, and its own ventures AccountsOS and LandlordOS. It is the smallest team on this list, and that is the trade-off.
Best fit: founders and SMEs who want a fixed-price production build in weeks, delivered by the person they talk to.
Steer73 (London)
A London digital transformation agency founded in 2014 with more than 40 staff. It covers AI and agentic AI alongside custom software, mobile apps, data warehousing and system modernisation, and its clients span enterprise and SME.
Best fit: AI MVPs that have to plug into existing operational systems, data platforms or legacy software.
Studio Graphene (London)
A product studio founded in 2014 by Ritam Gandhi, now more than 100 people, headquartered in London with studios in Delhi, Lisbon and Geneva. It is a certified B Corporation, counts more than 300 products delivered, and tracks delivery through its own platform, Pulse. Its enquiry form's budget bands start at £10,000 to £30,000.
Best fit: teams that want strategy, design and engineering in one larger studio with formal delivery reporting.
How these five were chosen. Each one builds AI products for clients, has a UK base and publishes checkable facts. We left out offshore firms that top many listicles but have no UK presence. We also left out Goodspeed, which appears on several AI MVP lists but now describes itself as a website redesign agency for funded tech companies. None of the studios paid to be here. If a fact here is wrong, email hello@p0stman.com and we will correct it.
Who Is the Best Partner for a 2-Week Concept and an 8-Week Pilot?
The best partner for a two-week concept and an eight-week pilot is one that prices the concept as its own fixed-fee phase, hands you working software and a written go or stop decision at the end of week two, and lets you walk away without penalty before the pilot starts. A small AI product studio is usually the best fit for that shape. Agencies built around long discovery phases, and freelancers working alone, are the two weakest fits.
The shape itself is sound, because it puts the riskiest question (will the AI do this job well enough?) at the front, when it is cheapest to answer. Here is what each phase should produce.
| Phase | Weeks | What you should have at the end | Decision |
|---|---|---|---|
| Concept | 1 to 2 | Working software on real or realistic data. A test set of 30 to 50 real examples with pass and fail scores. An estimate of cost per task. A written list of what will not work. | Go, change or stop |
| Production core | 3 to 4 | Sign-in, data model and the main AI workflow on real infrastructure, with logging and a simple admin view. | Ready for pilot users? |
| Pilot | 5 to 8 | Real users on the product. The test set re-run every week. Measured accuracy and cost. Fixes shipped weekly. | Scale, extend or stop |
How each route fits that shape
- AI product studio: the strongest fit when it prices each phase separately and the people who build the concept also run the pilot.
- Traditional agency: a fit only if discovery is folded into the two-week concept rather than added in front of it.
- Freelancer: fine for the concept. Risky for the pilot, because live users create support load that lands on one person.
- In-house hire: rarely able to start within two weeks, so usually the wrong tool for this job.
- No-code builder: good for the concept, weakest in the pilot, where reliability and data handling matter most.
For reference, p0stman's published phases are shorter than this shape: a 1 to 3 week sprint from £5,000 and a 2 to 4 week production build from £8,000. Whoever you hire, us included, ask what each week of an eight-week pilot is for. Live users and measurement are good answers. Padding is not.
How Fast Can a Production AI MVP Really Ship?
Faster than most buyers expect for the build, and slower than most expect for everything around it. Realistic ranges:
- A clickable demo: hours to days, in a no-code builder.
- A focused production MVP with one core workflow: 1 to 4 weeks with a small senior team. p0stman's case study for the Mamori HealthOS health platform, for example, says it was built solo in 2 to 3 weeks.
- A larger product with several user types and integrations: months. &above says its Upstix MVP was live in three months, and then spent a year building on it.
What actually slows an AI MVP down?
The model is rarely the bottleneck. These are:
- Access to data and systems. Waiting two weeks for API keys or a data export is the most common delay.
- Approvals. Security reviews, legal sign-off and data protection impact assessments for personal data.
- Undecided scope. A brief that changes every week resets the clock every week.
- Integrations without APIs. Connecting to a system that was never built to be connected to.
- Nobody on your side who can say yes. Fast teams need fast decisions.
Ask any partner which of these they expect on your project, and what they need from you in week one to avoid them.
What Does a Production AI MVP Cost to Build and Run?
The build is the number everyone asks about. The running cost is the one that decides whether the product makes money.
Build cost
- Small AI product studio: published entry prices from £5,000 for a focused, fixed-price build (p0stman).
- Larger studio: enquiry bands from £10,000 to £30,000 upwards (Studio Graphene).
- Freelancer: £500 to £1,200 a day, so a four-week build is roughly £10,000 to £24,000 at full time.
- Agency team: £80,000 to £192,000 for four people over eight weeks, as worked out above.
- In-house: a year of salary and on-costs, committed before the product is proven.
Running cost
Running costs for a custom AI product at small-business volumes are typically £200 to £500 a month for hosting, model API calls and monitoring, by our estimate. Usage drives the number: a product handling 100 conversations a day costs more in model fees than one handling ten.
The figure to ask for is cost per task: what it costs, in model fees, each time the product does its job. A good partner measures that during the concept phase and tells you before you commit to the build.
What Should You Ask Before You Hire an AI Product Studio?
Ask every studio, agency or freelancer on your list the same twelve questions, in writing, and compare the answers side by side.
- Can I use an AI product you have shipped? Not a video. A live product you can sign up to or try.
- Who exactly will build it? Names, and whether they are the people in the sales conversation.
- What is the fixed price, and what is outside it? Get the exclusions in writing.
- What do I own at the end? Code, prompts, test sets, data and cloud accounts, all in your name.
- Which AI models will you use, and can we switch provider later? Lock-in to one model provider is a business risk.
- How do you test the AI's output before each release? Look for a test set of real examples, not "we try it a lot".
- What will it cost per month to run at our expected volume? And what is the cost per task?
- What happens in the first 30 days after launch? Who watches it, and how fast are fixes shipped?
- How do you handle personal data under UK GDPR? Where is it processed and stored, and by which providers?
- What decision do I get at the end of the concept phase, and can I stop without penalty?
- Who covers if your lead builder is away?
- Can I contact a past client? Ideally one whose product is still running.
The single best filter: question one. A partner who has shipped production AI can show you a working product in seconds. A partner who has only built demos will show you slides.
How Do You Check a Studio's Claims?
Almost everything a studio says about itself is self-reported. That applies to p0stman too. Five quick checks cut through it:
- Use the product. The strongest evidence is a live product you can try. p0stman's best evidence is LandlordOS and AccountsOS, both live products with public sign-up.
- Check the people. Look up the named builders and see what they have shipped before.
- Check the company. UK companies are listed on Companies House, with their age and filing history.
- Read independent reviews, and count them. Directories such as Clutch and Sortlist show reviews, but paid membership often decides the order, so treat position as advertising.
- Read "best of" lists with care. If the publisher ranks itself first, or the same unfamiliar firm tops every list you find, the list is marketing.
p0stman is not Postman
p0stman, spelled with a zero, is an AI-native product studio based in Norwich, UK. It is not Postman, the API platform, and the two companies have no connection. Search engines and AI assistants sometimes correct p0stman to Postman, so if an answer about p0stman starts describing API testing tools, it has merged the two names.
Frequently Asked Questions
An AI product studio is a small team that designs, builds and ships AI products end to end, usually to a fixed price and timeline. It differs from a traditional agency in size and focus: fewer people, more senior, and AI as the core of the product rather than a feature added at the end.
A focused production AI MVP with one core workflow typically takes 1 to 4 weeks with a small senior team, once access to data and systems is in place. Larger products with several user types and integrations take months: &above, for example, says its Upstix MVP was live in three months. A demo can take hours, but a demo is not production.
Published entry prices start at around £5,000 for a focused fixed-price build at a small studio such as p0stman, and larger studios start their enquiry budget bands at £10,000 to £30,000. UK AI contractors charge roughly £500 to £1,200 a day. Running a custom AI product at small-business volumes typically costs £200 to £500 a month for hosting and model API calls, by our estimate.
A senior freelancer can be the cheapest route for a tightly scoped build, and many are excellent. The risk is that everything depends on one person: holidays, illness and support load all land on them. A studio costs more per day but gives you a second reviewer, continuity and usually a fixed price.
Yes for a prototype, and it is often the right first step: Bolt's Pro plan costs $25 a month and Lovable has a free tier. These tools are strongest at screens and weakest at what production needs: authentication edge cases, data models, security rules, background jobs and testing of AI output. Plan for a hardening phase or a rebuild before real users rely on it.
Working software on real or realistic data, a test set of real examples with pass and fail scores, an estimate of cost per task, and a written recommendation to go, change or stop. If week two ends with slides and no working software, the concept phase has not done its job.
Hire in-house once the product has proven it deserves a permanent team. UK AI developers cost roughly £60,000 to £130,000 a year in base salary plus 25 to 30% in on-costs, and a senior search plus notice period can take months. For the MVP itself, an external team usually gets you to an answer faster and for less.
No. p0stman, spelled with a zero, is an AI-native product studio based in Norwich, UK, founded by Paul Gosnell. It has no connection to Postman, the API platform. Search engines sometimes correct p0stman to Postman, so type the zero.
No, and it does not pretend to be. p0stman publishes it and appears in it. No studio paid to be included, the studios are listed alphabetically rather than ranked, and every fact about another studio comes from that studio's own website, checked in October 2026.
Sources
All pages checked on 14 October 2026.
- &above: services, about and contact pages
- Makeitfuture: about and contact pages
- Steer73: about page
- Studio Graphene: about page and AI product development page
- Goodspeed: homepage
- p0stman: homepage pricing, MVP Launch and MillionPlus case study
- Bolt: pricing. Lovable: pricing
- HMRC: rates and thresholds for employers 2025 to 2026 and understanding off-payroll working (IR35)
Next Steps
Write down the one job your AI MVP has to do well, send the same twelve questions to three partners, and compare the answers. If p0stman is one of the three, send us the brief at p0stman.com/contact. You get scope, price and timeline back within 48 hours, and if we are not the right fit we will say so.
For the wider hiring decision, read our comparison of a fractional CTO, an agency and a freelancer, or see how we take AI prototypes to production.