Every client job, planned, approved and proved.

An AI agent for your agency, each client and each project. You brief it in plain words, approve the plan, and it ships the work with the proof attached. We install it and train your team.

Join the waiting list Watch a client job run

For web agencies, freelance developers and in-house web teams. Set up for you, not handed over.

Built inside a working agency.

We have run our own client work through the Orchestrator since April 2026. It is not a prototype: these are our delivery records.

  • 6,600+agent runs

    Jobs planned and finished by an agent since April 2026.

  • 190+client projects

    Across 26 clients, all on one Orchestrator.

  • 861mistakes turned into rules

    Each written up with what to avoid next time.

Measured on our own Orchestrator, 29 October 2026.

One agent per client, one per project.

Each agent has its own memory and its own sandbox. A project agent can only read and change its own project's files, and that limit is enforced in code, not left to good behaviour.

Big jobs are split between helper workers on a cheaper model, inside the same project, and the project agent checks their work before it reports back.

From brief to proof, the same way every time.

Brief it like a colleague.

Paste the client's email, drop in the screenshot, or dictate it on the way to a meeting. The agent reads the client's history, the platform playbook and your house rules before it answers.

  • Attach files, screenshots and voice notes
  • Real email in and out, with attachments, where you allow it
  • Two AI vision models read every screenshot, not one

Plans you approve before anything changes.

Anything non-trivial starts as a short numbered plan with the risk and the way back spelt out. You approve it, change it or reject it. Nothing touches a client site before you say so.

  • Risk and how to undo it, on every plan
  • Approve from the board, the chat or your phone
  • Every plan and approval kept per client

Proof on every job.

Every job ends with a written validation: what changed, what was tested and anything still open. Visible changes come with screenshots or a short video, checked point by point.

  • Screenshots on desktop and phone
  • Checks listed with pass or fail
  • A cost for every job, so you know what to delegate

Watch a client job run.

Pick a brief. Read the plan, approve it, and watch the proof arrive.

Acme client · Magento

4. Size guides on 1,240 products

Add size guides to 1,240 product pages.

PLAN:

  1. Read the product template and the size data.
  2. Add the size guide block to the product template.
  3. Screenshot five products on desktop and phone and check each one.

RISK: low · template change, no schema change

REVERSIBILITY: reversible-in-place

VALIDATION: 3 checks passed. Nothing else changed.

NEXT:

QUESTIONS: none.

RISK: low

Evidence3 items · 6 of 6 checks passed

      Lesson saved: size data lives in the size_chart attribute.

      Fictional clients, example data. The orchestrator never changes a client site before you approve.

      Join the waiting list See how it learns overnight

      Let clients see the work, not chase it.

      Give a client their own workspace for their project, signed in with Google, inside your website. They can ask for changes, approve plans and see the proof. They see their project and nothing else.

      Join the waiting list

      Can we add size guides to the product pages?

      Plan ready. Nothing changes until you approve.

      PLAN: size guidesApprove

      Proof attached

      Signed in with Google. This project only.

      Your agency learns overnight.

      Most AI tools forget you the moment the chat closes. The orchestrator writes down what every job taught it, and three night shifts turn those notes into rules.

      1. 03:00 Review.

        It audits the last 24 hours of work and ranks what could be better.

      2. 04:00 Consolidate.

        Lessons from every client and project are merged into playbooks and principles.

      3. 05:00 Promote.

        Well-evidenced rules go live for every agent. Anything big waits for your sign-off.

      4. 09:00 Your team starts.

        Every agent, on every client, starts the day knowing what the whole agency learned yesterday.

      And you inherit ours.

      Every orchestrator ships with our core: 32 platform playbooks, 262 principles and 861 lessons from real client work, and it keeps receiving our updates as we learn.

      Join the waiting list

      Your clients' data and rules stay in your orchestrator. What we share is the craft, never your clients.

      What it works on today.

      Counted from our own delivery records, 29 October 2026. Each platform has its own playbook, corrected every night.

      Beyond code, the same agents write SEO changes, Google Ads fixes, reports and client emails. Shopify and Craft CMS work is light so far.

      Standing jobs that run themselves.

      Some work is the same every week. Set it once and let it run, inside limits you choose.

      • Schedules. Daily or weekly jobs, such as a Monday report or a monthly check, run at the time you set.
      • Autopilot. A standing instruction works through a queue in batches, and stops on its own when something needs a person.
      • A summary after each batch. You get an email with what was done, what was skipped and why.

      Set up for you, not handed over.

      The Orchestrator is installed by us, not downloaded. Real developers from our team in Bromley set it up and stay on hand.

      1. Scoping call

        We map your clients, platforms, access and the way you work.

      2. Install

        We install your Orchestrator and load your clients, access and house rules.

      3. First jobs together

        Your team runs real client jobs with one of our developers alongside.

      4. Your team runs it

        You approve the plans. We keep the playbooks and models current.

      See how setup works

      Questions agencies ask first

      • Does it replace my developers?

        No. It takes the repetitive work off them and turns them into reviewers of more work. Most agencies use it to take on clients they would otherwise turn away.

      • Can it break a client site?

        Anything non-trivial starts as a plan you approve, runs inside a sandbox for that one project, and ends with screenshots and checks. You can reject or change any plan.

      • Which AI models does it use?

        Claude does the work, and GPT and Gemini cross-check anything visual. We move to better models as they arrive, so you do not have to track them.

      • What does setup involve?

        A scoping call, then we install it, load your clients, access, platforms and house rules, and train your team on real jobs.

      • Who owns what it learns?

        You do. Your clients' knowledge stays in your Orchestrator. Our core playbooks keep improving and reach you as updates.

      • How much does it cost?

        It is priced on a call, because setup depends on how many clients and platforms you bring. Join the waiting list and we will be in touch.

      • Something else on your mind?

        Ask us directly and a developer will reply.

        Ask a question

      Bring us your client list.

      Join the waiting list. When your place comes up, we will show you the Orchestrator running on one of your real jobs in the first call.

      Join the waiting list Watch a client job run