Questions, answered.

What is Adapt Progress Evolve?

We're a UK software studio, and we build and run our own products - SaaS tools, data explorers, open-source libraries, that sort of thing. Day-to-day it's a fleet of specialised Claude AI agents doing the work, but people direct it, and anything customer-facing, destructive or costly needs a person's approval.

Do you take on client work?

Yes, selectively. We do AI agent development, Claude/LLM API integration, and rapid MVP builds - the same work we do on our own products. We keep the client list small on purpose, though. Our own things need to keep shipping too.

What does "AI-native studio" mean? How is that different from an agency that uses AI?

Most agencies use AI the way everyone does now, a developer pings a chatbot for help and moves on. That's not what we mean. An AI-native studio is structurally different: agents actually hold responsibilities here, engineering, SEO audits, content drafting, design review, monitoring, and they coordinate through shared infrastructure, with a person approving anything customer-facing, destructive or costly. You can see the difference in the output too: our agents have commit histories, not marketing copy.

What can AI agents actually do for a business?

What works, right now, today: research and monitoring, content pipelines with quality gates built in, code review and testing, data auditing, customer-facing drafts that get routed through human approval, and the ops glue between systems that nobody wants to do manually. The pattern that actually works is agents doing the volume, people making the calls. What doesn't work is treating an agent like a full employee with no oversight.

How fast can you build an MVP?

A focused MVP is usually weeks, not months. The AI-accelerated pipeline compresses the build, and fixed-scope sprints keep everyone honest about it. But speed really depends on scope, so we'd rather give you a real date on a scoping call than some impressive-sounding number on a landing page.

What is your stack?

TypeScript and React up front, Supabase (Postgres) and edge functions on the back, Cloudflare handling hosting and delivery, and the Claude API family doing the agent work. Three.js gets a look in where 3D actually earns its place. We keep the tools boring and proven, and spend the innovation budget where it counts: the agent layer.

Is a human in the loop?

Always, at the points that matter. Anything customer-facing, anything destructive, anything that spends money, it all goes through human approval. Agents do the work. People own the decisions. That's the split, and we don't move it.

Which products do you actually run?

Quite a few, actually. Live right now: ScriptGrain, Launch Video Maker, visibility to ai, How to get AI to, Hilyt.it, Assets for three.js, What is GEO, Can You Use AI, Is There a Thing, reviewsnot, I Want AI To, Using AI As, ukspend, Paws of London, ReviewNudge, Cortex Map, Jarvising, Brands over time, AI Guides On, Dead Writing Theory, Product Photo Shots, Chartsmade, Decksmade, Seracu, Look Good Tools, Noctem Labs, Not Quite Folk and MarketFrame. In development: Forematter and Jarvis (our own private agent fleet). Paused: UKMWI. We'd rather say that than pretend every shelf is full. We build these and we run them, which is why we know what breaks.

How do you stop AI-written content getting things wrong?

Each product has its own checks, and a person makes the calls that matter. How to get AI to guides are drafted with ScriptGrain from a fact sheet per topic, then read in full and checked against their sources before going live. Launch Video Maker only puts facts on screen that the customer approved, each one with the page it came from. Paws of London publishes how it checks listings, and says plainly what a check does and does not mean. Anything customer-facing, destructive or costly goes through human approval.

Is any of your work open source?

Some of it, yes. Cortex Map is a React and three.js knowledge-graph library, released under the MIT licence. The code behind What is GEO is public on GitHub, also MIT. Then there's the Jarvis architecture reference, which is public too, with real numbers and real failures in it (the failures are the useful bit). The Jarvis system itself stays private, so that's the one line we hold. Everything else mentioned here you can go and read, which feels about right for a studio that talks so much about checking things.

What does "no lock-in" mean if we hire you?

It means the work is yours. Full code ownership, no strings. You get a fixed-scope written proposal after a short scoping call, then focused sprints with visible progress, and on MVPs a live URL from week one. At the end it's deployed, documented and handed over, so you aren't stuck waiting on us to explain our own code. After that you've got a simple choice: stay on a retainer for iteration, or take the keys and go. Both are fine.

Where are you based, and how do I get in touch?

London, UK. Email hello@adaptprogressevolve.com and we'll get back to you, usually within a day. We reply to every genuine enquiry.

Reception

Something we didn't cover?

Ask at the desk. A short message is plenty, and we reply to every genuine enquiry.

Write to

hello@adaptprogressevolve.com

Based in

London, UK

Replies

Usually within a day

Ready to build? See the stations