Talk numbers
Work

Client stories: the question each business couldn't answer, what we built and what changed.

All work
Built

Tools we built, for clients and for ourselves, when nothing we could buy did the job.

Open to anyone

Runs our work

For clients

See all builds
AI

AI moves fast. Here's how we use it in our own work, and where to start in a business your size.

The AI hub
Approach

Every business is different, so we always start with a conversation.

How we work
  1. 15 minA quick callYour goals, your numbers, and whether we're a good fit.
  2. 1 hourDiscoveryWhere you're losing enquiries, and what's already working.
  3. ThenA clear planPaid ads, a custom build or both, as a one-off project or ongoing.

What we connect

About

We're a small London team of marketers, and we use AI to go faster.

About us
  • Cam BowenCam BowenCo-founder
  • Ben PowerBen PowerCo-founder
  • ElectraElectra
  • MahirMahir
  • NatashaNatasha
  • RahabaRahaba
  • ZobaerZobaer
Tool · AI

Anthropic

Claude does the planning, the hard builds and the reviews, through Claude Code. Opus for the hardest jobs, Sonnet for routine build work, Haiku for simple runs.

Talk numbers All tools
Why we use it

Why Claude gets our hard jobs

  • Right model, right job

    Opus for the lead session when a job needs it, Sonnet for sub-agents, Haiku for simple runs.

  • Big jobs, split up

    A lead session hands parts of a job to sub-agents, each in its own copy of the repository, then reviews what comes back.

  • Reviews the other agents

    Claude Code reads every pull request Cursor's agents open before anything merges.

  • Written-down limits

    Model limits sit in our agent rules, and in our operations repository a hook blocks email and publishing unless someone asked for them there and then.

  • Claude stays in Claude Code

    Cursor's agents run Cursor's own models, so each plan covers its own work.

  • A person signs off

    Claude writes and reasons well and still gets facts wrong, so someone checks before it ships.

Claude is the family we trust with our code, and we match the model to the job. Tokens cost money and bigger models cost more, so the biggest model only gets the work that needs it. Claude runs through Claude Code, which sits inside Cursor.

Which model for which job

  • Opus: the lead session, only when a job needs it. Planning a big build, design-heavy work, anything with open questions.
  • Sonnet: the most a sub-agent can use. Sub-agents build the parts of a job in their own copy of the repository.
  • Haiku: simple, repeatable runs.
  • Cursor's own models: Cursor's agents take the clear-cut jobs (fixes, small features, docs) on Cursor's Auto, Composer or Grok, never on Claude. Claude work stays in Claude Code.
  • Gemini: sits behind some of our tools, like the TechStack Analyser.

Why the split

We pay for Claude and for Cursor separately, and some days we hit the limits on the AI tools we use. So Claude's kept for the work it does best, and none of our work depends on one tool. A subscription is a good way to manage cost, as long as your workspace can run several agents at once without losing context. The right plan depends on the person and what they do.

Where Claude runs for us

  • Claude Code: plans the work, builds the hard parts, briefs Cursor's agents and reviews every pull request they open.
  • Claude Desktop: laptop automation outside our client repositories.

Rules it follows

Our model limits are written into our agent rules. In our operations repository, a hook blocks email and publishing from Claude Code unless a person asks for that action there and then. A person decides what goes live.

As of 10 October 2026: 37 of the 45 repositories in our GitHub organisation have Claude Code set up, and 130+ pull requests have merged on this website's repository since 29 September 2026.

There's more on how it all fits together on our AI page. If you want the same setup for your team, have a look at AI setup.

Wires

What we connect it to

On its own

Models write and reason well, and they still get facts wrong. So a person checks anything before it leaves the building.

Claude CodeCursorGoogle GeminiGrokGitHubElevenLabs
Services

Services that use it

We don't sell it on its own. Each of these services uses it alongside the rest of your setup.

How we start
About the tool

What Anthropic is

Anthropic is the AI company that makes Claude. Its models come in tiers, and Opus, Sonnet and Haiku are the ones we use. They power the Claude chat app, Claude Code, Claude Desktop and an API for building them into other tools.

Questions

Questions about Anthropic

Straight answers on how we work, what it costs and what to expect. If yours isn't here, just ask.

Still have questions?

Talk to us
Which Claude model do you use for what?

It depends on the job. Opus for the hardest jobs, Sonnet for routine build work, Haiku for simple runs. Bigger models cost more, so we match the model to the task.

Why pay for Claude as well as Cursor?

Because they do different jobs, and some days we hit the limits on the AI tools we pay for, so none of our work should depend on just one. Cursor's agents run Cursor's own models, and Claude runs through Claude Code. A subscription is a good way to manage cost, as long as your workspace can run several agents at once without losing context. The right plans depend on the person and what they do.

Can I trust what Claude writes?

Not blindly. Claude writes and reasons well, and it still gets facts wrong. So a person checks anything before it leaves the building.

Where next

Start with your numbers.

We'll tell you which tools you need and which connections are missing.

Trusted by teams across the UK and beyond

Next step

Bring your numbers.

We'll tell you if the spend makes sense.

  1. A 15-minute callYour goals, your challenges, and whether we're the right fit.
  2. A 1-hour discovery callWe dig into your numbers. Worth a hundred emails.
  3. A concrete planAds, a custom build or both, as a one-off project or ongoing.
Talk numbers
15 minutes · No pitch deck · A straight answer