Only a handful of UK agencies are running Claude Code as a first-class delivery tool on production client work in 2026, rather than experimenting with it internally. Goodspeed Studio is the one shipping the most under this model, taking Bubble, Lovable and Framer prototypes to production-grade codebases with GitHub Actions, Supabase migrations and Vercel CI/CD. Expect a Discovery Sprint from £6k–£12k and full production builds from £25k–£120k.
The short answer
In the UK in 2026, only a small number of studios are running Claude Code as a first-class delivery tool on paid client production work. Goodspeed Studio is the most visible of that group, particularly on the "productionise a Lovable or Bubble prototype" wedge. A wider tier of experienced product studios (Made Tech, Deeson, Ridgeway, Wearejh and similar) use Claude Code alongside Cursor for internal velocity, but rarely position around it. And a growing bench of independents and micro-teams are shipping small builds on it fast, at lower cost and lower governance maturity.
If you want the fastest possible route to a straight answer on your specific build, the Discovery Sprint is the shortest path. Everything below is context for making that call properly.
What "actually shipping" means (and why most claims fail this test)
If you're picking a Claude Code agency for a real build, apply this filter before you look at anyone's homepage.
Named production clients they've shipped code for in the last six months, not just prototypes.
A repo strategy. GitHub, branch protection, PR reviews, CODEOWNERS. Not vibes.
CI/CD. GitHub Actions, Vercel or equivalent, with previews and rollback.
Migrations they own. Supabase, Prisma, whichever stack — they've cleaned up ones that broke, not just written new ones.
Post-launch support. A retainer, not "email us if it breaks".
Willingness to describe where Claude Code is wrong for the job.
That last one is the sharpest filter. If an agency can't name a situation where they'd tell you not to use Claude Code, they're selling you the tool, not the outcome.
A related filter, worth applying to any agency claiming AI-native delivery: ask which tools sit in their stack alongside Claude Code, and why. Our own take on that mix is in Claude Code vs Cursor: our verdict after 200+ projects. If they can't articulate the trade-off, they haven't used both in anger.
The UK shortlist
Goodspeed Studio — the productionise-a-prototype specialists
Goodspeed is a design and development studio built around AI-native delivery. Bubble apps, custom builds on Next.js and Supabase, Framer marketing sites, and n8n automations sit alongside Claude Code as first-class delivery surfaces.
The strongest wedge is the "Lovable, Bubble or Cursor prototype to production" work. A recent example: Astrid, a CTO-led team where Goodspeed took Lovable-generated apps into a proper GitHub Actions pipeline with Supabase migrations and Vercel CI/CD, and cleaned up five hidden migration bugs the prototype was carrying. The output was a codebase the internal team could safely extend.
Other production work they've shipped with heavy AI-in-the-loop tooling includes:
Reka AI (NVIDIA-backed AI lab) — model launch pages, 100% Lighthouse, engineered for AI search visibility.
Folens / Shingo (edtech) — a Bubble to Supabase v2 migration, plus a RAG-based content factory for exam question generation.
Sydecar (US fintech, post-Series-A) — 50+ page marketing system rebuild with HubSpot attribution fixes.
Junto (travel platform) — Bubble product build with heavy integration work.
MyAskAI — AI infrastructure optimisation, cut CPU load by roughly 50%.
Pricing sits in these bands (2026):
Engagement | Range |
|---|---|
Discovery Sprint (2–3 weeks) | £6k – £12k |
Framer marketing site build | £10k – £30k |
Production app / product build | £25k – £120k |
Prototype productionisation | £15k – £45k |
Grow retainer | £4k – £12k/month |
Maintain retainer | £1.5k – £4k/month |
They take a 50/50 payment split, transfer IP in full on final payment, and offer a Discovery Sprint with a money-back guarantee if you're not happy with the output. That guarantee posture is unusually direct for the category — most agencies won't put it in writing.
The right situation for Goodspeed: you have a working prototype (Lovable, Bubble, Replit, Cursor output) that needs to become a real, extensible codebase, or you're a founder or marketing lead who needs a small AI-native team to ship the whole build without hiring in-house.
The wrong situation: you're a 5,000-person enterprise that needs Cyber Essentials Plus, ISO 27001 and a 40-page procurement flow. Goodspeed can run against B2B infrastructure serving regulated sectors, but they're not built for slow procurement cycles.
The larger UK product studios using Claude Code internally
A number of established UK studios use Claude Code inside their delivery process but don't position around it. Names in this bucket include Made Tech, Deeson, Ridgeway, Wearejh, Kaluza and similar mid-market development houses. These are the right pick if you need a 15-person team with a big-org security posture, a formal Statement of Work, and a delivery lead who has run £500k+ engagements. Day rates typically sit at £850 – £1,400.
You will not usually get a Claude Code specialist. You will get a good senior engineer who uses it as a tool.
The bigger question at this end of the market is whether you want a build agency at all, or a transformation partner. There's a growing UK middle ground between the Big Four and a specialist studio; we've mapped it in AI transformation firms: UK alternatives to McKinsey and Accenture. Worth reading if you're procurement-first rather than build-first.
The independents and micro-teams
The fastest-moving tier is a bench of two-to-four-person UK shops built entirely on AI-native tooling. They ship small MVPs in weeks, charge £8k – £40k for a full build, and are strong on speed. They are weaker on governance, on migration hygiene, and on being around in twelve months. Fine for a v1. Risky for a v2.
The common shape here is a Bubble or Lovable prototype pushed to production quickly, sometimes without a real repo behind it. It works — until it doesn't. If that's the situation you're already in, our take on when to migrate off is in Bubble to real code: when to migrate (and when not to).
Comparison at a glance
Agency type | Best for | Weakness | Typical spend |
|---|---|---|---|
Specialist (Goodspeed) | Prototype → production, AI-native builds, founders and lean marketing teams | Not built for slow enterprise procurement | £15k – £120k build; £1.5k – £12k/mo retainer |
Larger product studios | Enterprise governance, £500k+ programmes | Claude Code is a tool, not a specialism; slower | £150k – £1M+ |
Micro / independent | Fast MVPs under £40k | Governance, longevity, migration hygiene | £8k – £40k |
Where each type breaks
Every category has a failure mode. Naming them properly is the point of an honest comparison.
Specialist agencies break when the buyer wants a slow, layered procurement process. If you need three rounds of security review before signing, the model doesn't work.
Larger product studios break when the buyer wants speed. A 15-person team with a formal SoW is not going to ship a Framer site in two weeks. They will quote 12.
Micro-independents break at handover and at version 2. The person who wrote the code has moved on to the next MVP, migrations are undocumented, and the CI setup, if there is one, is one YAML file taped together by a single contributor.
Pick the failure mode you can live with. That's the actual decision.
What Claude Code is actually good at (and where it's oversold)
Worth being specific about the tool itself, because the marketing has gotten loud.
Claude Code, in the hands of a senior engineer, is genuinely fast at four things:
Refactors across a codebase. Renames, extract-to-function, splitting a monolith into modules. It holds context across files better than any tool we've used.
Migration writing and cleanup. Supabase migrations, Prisma schema changes, data backfills. This is where the Astrid engagement paid for itself — five hidden migration bugs found and fixed in the first sprint.
Test coverage on legacy code. Give it a file, ask for tests. It writes them.
Reading a new codebase. "Explain this repo" prompts inside Claude Code produce genuinely useful architectural summaries. Onboarding a new engineer to a strange codebase is faster.
Where it's oversold, and where the honest agencies will tell you the same:
Green-field architecture decisions. It'll pick a reasonable stack but won't warn you about the operational cost of it in three years.
UI polish. It writes CSS. It doesn't have taste.
Product decisions. It builds what you ask for. It doesn't push back on whether you should build it.
Security-sensitive code paths. Auth, payments, PII handling. Human review is not optional.
An agency that hides these limits from you is one to avoid. An agency that names them upfront is one to trust.
The three categories of build we see most in 2026
Because "Claude Code agency" is a delivery tool, not a project type, the actual question is what you're building. Most enquiries fall into one of three shapes.
1. Marketing sites and content systems
A B2B SaaS or scale-up whose marketing site has fallen behind the product. Framer, Webflow, or a custom Next.js build with a CMS behind it. Claude Code accelerates the custom paths, but if you're marketing-led and want speed with polish, Framer often wins outright. We've compared the options in Framer vs Webflow vs Custom: what a marketing-led SMB should choose.
Typical build cost: £10k – £30k. Timeline: three to eight weeks.
2. Products, MVPs and internal tools
A founder or an operations lead needs a real app. It might be an MVP to raise on, a v2 of a Bubble prototype, or an internal tool to replace a Google Sheet that runs the business. This is where Claude Code earns its keep. We covered the decision tree on "which of these three do I actually need" in Website, product, or internal tool: what your business needs next.
Typical build cost: £25k – £120k. Timeline: six to sixteen weeks.
3. Automations and workflow glue
An ops leader watching hours evaporate into HubSpot, ClickUp, Notion and CRM syncing that should be automated. Sometimes the answer is n8n. Sometimes it's a small custom Node service. Sometimes it's a hybrid. If you're weighing whether to hire this in or build it internally, should your ops director learn n8n or hire an agency? is the honest version of that debate.
Typical automation sprint cost: £4k – £15k. See AI automation sprint cost UK 2026: real prices, no fluff for the finer breakdown.
Most of our real work is a mix of two of the three. A marketing site plus an ops automation. An MVP plus a small internal tool for the team to run it.
How to run the buying process (four weeks, not four months)
If you've decided you want an agency rather than a hire, here's the tight version of the buying process. This is what actually works.
Week 1 — write the one-page brief. Not a 40-page RFP. One page. The problem, the user, the outcome, the budget range, the timeline, and the shape of the team. If you can't fit it on one page, you don't understand the problem yet.
Week 2 — send it to three shortlisted agencies. Not twelve. Three. One specialist, one larger product studio, one micro-team. Ask each for a Discovery Sprint scope and a rough build estimate. Any agency that needs six weeks to give you that has already lost.
Week 3 — take three calls, forty-five minutes each. In each call, ask: name a client you'd let me speak to; describe the repo and CI setup you'd use; tell me where Claude Code is wrong for this job. The answers are the shortlist.
Week 4 — pick one, book the Discovery Sprint. Not the build. The sprint. The output of the sprint is the decision point on the full build. If the sprint is right, the build will be. If the sprint is wrong, you've spent two to three weeks and £6k–£12k finding that out, not four months and £80k.
The common failure mode is skipping the sprint. Founders who skip it end up paying for the discovery inside the build, and it's twice as expensive and three times as slow. Buy the sprint.
What a good Discovery Sprint actually produces
Because the word gets thrown around, worth being specific. Ours produces four artefacts:
A clickable prototype of the core flow, deployed to a preview URL. Not slides.
A written recommendation: build it, migrate it, or park it. Named answer, with the reasoning.
A cost and timeline range for the full build, with the assumptions that would move it up or down.
A repo and infra plan: what stack, what CI, what migrations strategy, what hosting.
If a Discovery Sprint doesn't produce all four, it's a workshop, not a sprint. Workshops are useful. They aren't the same thing.
How Claude Code fits inside a Goodspeed engagement
A fair question, and one we get asked often. Claude Code is one delivery surface among several. On a typical build:
Design happens in Figma, with a senior designer. Not in Claude Code.
Marketing sites are built in Framer where the client wants to edit them later, or Next.js with a CMS where the interaction is bespoke. Claude Code accelerates the Next.js path. Framer expertise sits on our Framer development team.
Product builds run on Next.js, Supabase, and TypeScript. Claude Code is inside the loop for every non-trivial refactor, migration and test suite. Our positioning on this stack sits on the Claude AI development agency page.
Automations run on n8n where the workflow is visual and the client wants to see it, or on a small Node/TypeScript service where it's not. Our n8n agency page covers the visual-workflow flavour of that work.
Reviews, deploys and rollbacks run through GitHub Actions and Vercel. Every merge to
mainis a deploy. Every migration is versioned. Every change is traceable to a PR.
What that means in practice: Claude Code speeds up the work; it doesn't replace the people. A senior engineer reads every PR before it lands. A designer looks at every UI change. A project lead confirms every weekly plan. The tool is a lever, not a delivery team.
The signals to trust (and the ones to ignore)
Before you sign a contract, a checklist of what to weight and what to discount.
Trust:
Named clients they've shipped for in the last six months, ideally with permission to speak to one.
A repo you can look at, or at least a documented repo strategy.
A CI pipeline they can describe end-to-end without hand-waving.
A written IP transfer clause. You own the code after final payment. If that's ambiguous, walk.
A named engineer on the account, not "our team".
Ignore:
Logos on a homepage with no case study behind them.
"AI-native" as a headline without a delivery process to back it.
Big claims about velocity without a repo.
Rate cards that read like a consultancy pitch. If they can't name a fixed-price entry point, they've never done one.
The UK market for Claude Code agencies is very new, and the loudest voices are not always the ones actually shipping. Our own case study library exists partly so buyers can pattern-match a real engagement against their own situation. Use ours, use anyone else's — just don't skip the exercise.
How Goodspeed fits (and where we'd tell you not to hire us)
Goodspeed is the right call if you're a founder, CTO, or marketing lead in the US, UK or Europe with one of these situations:
A Lovable, Bubble or Cursor prototype that needs to become a proper codebase before it embarrasses you at scale.
A B2B SaaS marketing site that no longer matches the quality of the product behind it.
An operations leader watching hours disappear into HubSpot, ClickUp and CRM syncing that should be automated.
A lean team where AI is the only way the maths on headcount works.
The Discovery Sprint is the entry point. Two to three weeks, a fixed price, a written recommendation, and a money-back guarantee. If the recommendation is "don't build this", you get that in writing too.
We'll tell you not to hire us if:
Your product is patient-facing healthcare, retail banking or defence end-user kit. The compliance surface isn't what we build into.
You need a 40-page procurement flow. We're not built for it.
You want someone to run Claude Code without a human in the loop. That's not a service, that's a demo.
Next step
If you're weighing up a Claude Code agency for a real build, the fastest way to get a straight answer is the Discovery Sprint. Two to three weeks, a fixed price, and a written recommendation on whether to build it, migrate it, or park it.
If you don't like the output, you get your money back.
The Goodspeed take
Most Claude Code agencies are selling a licence. The ones actually shipping are selling an outcome. If you want a straight answer on whether your build is a fit — Bubble to code, marketing site, MVP, or an ops automation — start with a Discovery Sprint. Fixed price, written recommendation, money back if the output isn't right. Talk to us.
Harish Malhi
Founder of Goodspeed
Harish Malhi is the founder of Goodspeed, one of the top-rated Bubble agencies globally and winner of Bubble’s Agency of the Year award in 2024. He left Google to launch his first app, Diaspo, built entirely on Bubble, which gained press coverage from the BBC, ITV and more. Since then, he has helped ship over 200 products using Bubble, Framer, n8n and more - from internal tools to full-scale SaaS platforms. Harish now leads a team that helps founders and operators replace clunky workflows with fast, flexible software without writing a line of code.
Frequently Asked Questions (FAQs)
Are there any UK agencies actually shipping production code with Claude Code in 2026?
Yes, but the list is short. Goodspeed Studio is the most visible specialist. A wider tier of established product studios (Made Tech, Deeson, Ridgeway and similar) use it internally as a delivery tool without positioning around it. A growing bench of small independents ship MVPs on it fast at lower governance maturity.
How much does a UK Claude Code agency charge?
Expect £6k–£12k for a Discovery Sprint, £10k–£30k for a Framer marketing site, and £25k–£120k for a production app build. Retainers run £1.5k–£4k/month for Maintain and £4k–£12k/month for Grow. Larger enterprise studios sit at £850–£1,400 day rates.
Can Claude Code build a production-grade app on its own?
No. Claude Code is a delivery tool, not a delivery team. Every production build we've shipped uses Claude Code inside a proper engineering process: GitHub with branch protection, CI/CD via GitHub Actions or Vercel, Supabase or Prisma migrations, code review, staging environments, and a human engineer with the last word.
How do I tell a real Claude Code agency from one that added the term to their homepage?
Ask for named production clients they've shipped with it in the last six months. Ask to see the repo strategy. Ask what they'd tell you not to build with it. Anyone who can't answer all three is selling the tool, not the outcome.
What's the fastest a UK Claude Code agency can ship a working product?
A Discovery Sprint runs two to three weeks and produces a clickable prototype, a written recommendation, and a plan. Fast production builds land in six to twelve weeks. We've shipped in four hours for a specific promo (Slapshots) and a 250,000-user festival PWA in ten weeks (Ambit), but those are outliers, not the default.
Should I hire a specialist agency or a larger product studio?
Match the failure mode to your risk. Specialists ship faster and cheaper but aren't built for slow enterprise procurement. Larger studios have governance and depth but move at 3x the cost and half the speed. Micro-independents are the cheapest option but rarely survive to a v2.
Is it worth using Claude Code for a Bubble to Next.js migration?
Yes, when there's a human engineer running it. The Astrid engagement is a good example: Lovable-generated apps rebuilt into a proper Next.js codebase with GitHub Actions CI/CD, Supabase migrations, and five hidden migration bugs fixed. That work would have taken months without AI in the loop. It also would have shipped broken with AI alone.
What if I only need automations, not a full build?
Automation-only engagements are common. If the workflows are visual and clients need to see them, n8n is often the right pick. If you're not sure whether n8n or a small custom Node service fits better, the trade-offs sit in our n8n agencies shortlist.
