Free · 60 s Free AI-readiness audit See your site the way ChatGPT does.
Personal reply within one business day.

Runs for you · 30-day sprint

AI Agents & Automation Sprint

In 30 days I hand one to three tasks your senior people repeat every week to AI agents. They research, draft and pass the work on; a person approves anything that matters.

from €7,500$8,300 30 days

Typical project €7.5k–15k.Typical project $8.5k–16.5k. Fixed price per sprint.

Get my instant estimate

Proof on file

CLIENT · NAME CHANGED

€100

hard monthly AI budget breaker

Corvane Legal’s engine runs on its own every few minutes, sends a daily approval email, and stops spending at a hard monthly cap.

observed 2026-10-05 · engine config default

03 Fit

Who this is for, and who it isn't.

Built for you if

  • Senior people in your firm spend hours each week researching, drafting, searching and chasing.
  • Leads wait hours, or until tomorrow, for a first reply.
  • Answers to staff questions sit in old emails, drives and manuals, and nobody can find them.
  • Onboarding a new client takes weeks of questionnaires and reminder emails.

Not for you if

  • You want a chatbot on your homepage for its own sake.
  • The decision legally requires a professional. Agents draft; a qualified person decides, always.
  • There’s no settled way of working yet. We’d map it first: automating chaos makes faster chaos.
  • You need a simple two-step automation. A tool like Zapier does that; keep the money.

04 The problem

Your best people are doing your most repetitive work.

Look at a senior person’s week: researching a lead before a call, answering the same client question again, hunting for an old document, chasing a client for onboarding details. None of it needs their judgement until the last step.

That’s the kind of work AI agents do well: gather, draft, check, pass on. Searches for “AI agents for business” grew +210% in a year (Truelogic/DataForSEO, June 2026).

The real risk is an agent that sends, spends or promises something on its own. So every workflow I build has a person who approves, a budget it can’t exceed, and a record of what it did.

Signs you need this

  • Only one person knows where anything is.
  • You tried a chatbot once and it made things up.
  • Monthly reports are copy-pasted together by hand.
  • Partners approve things by forwarding emails.

Not sure which workflow to automate first?

Book 20 minutes and bring your busiest week. You leave knowing where I’d start, whether or not we work together.

05 Where I suggest starting

Three workflows worth building first.

Your sprint can cover whatever the review ranks highest. I suggest these three: the hours saved are obvious and the risk is easy to control.

Onboarding

AI client onboarding

A guided form that collects everything a new client engagement needs. Clients can pre-fill it from their website or documents.

  • Saves at every step
  • Documents uploaded in the same place
  • A complete brief for your team
  • Reminders sent for you, approved by you

Speed to lead

Lead briefs and reply drafts

Every new enquiry researched before you open it: who they are, what they need, whether they fit, and a reply drafted for you.

  • Researched from public sources, with links
  • A fit score against criteria you set
  • Draft reply in your voice, never sent automatically
  • Also: inbox replies, reports, approval emails

06 What you get

What you get, item by item.

Every workflow runs on real work in your accounts, not a demo on mine.

  • A review of your repetitive work

    Ranked by hours saved, risk and the data it needs, before anything is built.

    Helps with: Runs for you
  • One to three working agents

    Running on live work by day 30, chosen by you from the review.

    Helps with: Runs for you
  • Approval and a spending cap

    Anything that sends, spends, publishes or promises waits for a named person. Each workflow has a hard monthly budget; at the limit it stops.

    Helps with: Runs for you
  • A record of every run

    What went in and out, the cost, and who approved, all searchable.

    Helps with: Runs for you
  • Tests on real examples

    Built from your real cases, so an AI update can’t quietly change how it behaves.

    Helps with: Runs for you
  • Manual and handover

    How each workflow works, what to do when it fails, and how to switch it off.

    Helps with: Everything

07 How it runs

Thirty days, from review to running on live work.

Fixed price, fixed timeline. You decide what gets built and where a person approves, never the agents.

Agents at work Human gate: nothing moves on until a person approves

  1. Step 1: Days 1–5

    Review your work

    Interviews and a look at your real inboxes, drives and tools. Each task scored on hours saved, risk and the data it needs.

    • agent://researcher
    • agent://architect

    Human gate You choose which workflows get built.

  2. Step 2: Days 6–10

    Design

    For each workflow: what starts it, what data it uses, the agent’s job, what it may never do, where a person approves, and its budget.

    • agent://architect

    Human gate You approve the design and the “never allowed” list.

  3. Step 3: Days 11–22

    Build and test

    Built in your accounts and tested on real examples. You see a working demo every week.

    • agent://builder
    • agent://tester
  4. Step 4: Days 23–27

    Pilot

    On live work, every output approved by your team, instructions tuned on their feedback.

    • agent://watcher

    Human gate Your team approves every output during the pilot.

  5. Step 5: Days 28–30

    Handover

    A manual, a recorded walkthrough, and a check of who can access what.

    Human gate You decide which approvals stay manual for good.

08 Proof

Agents with limits, already running on real work.

Same pattern everywhere: agents do the work, a person approves what matters, a budget guard stops anything running away.

CLIENT · NAME CHANGED

Runs every 5 min

Corvane Legal: approval email and research agent

A daily approval email, a research agent that turns a name sent by email into a drafted article, 649 automated tests and a €100 monthly spending cap.

  • Agents
  • Approval email
OWN PRODUCT · LIVE ↗

8 automatic traffic limits

Filyhost: budget guard

A spending guard pauses sign-ups, then uploads, before a hosting bill can run away, and emails a daily report.

  • Automation
  • Spend control
Open the live product ↗
~€33/mo✓ AI spend (throttled to 15 rewrites/day) observed 2026-10-05 · configured throttle
649✓ automated tests observed 2026-10-05 · test suite count
4✓ AI providers (Claude, GPT, Gemini, Grok) observed 2026-10-05 · codebase

And Cuckoo, my own distribution engine in pilot: agents draft a week of posts, clients approve each one from a phone.

09 Price

What moves the price, and what it costs to run.

Starts at €7,500$8,300 per sprint; most land at €7.5k–15k$8.5k–16.5k. Fixed price per sprint.

Costs more when

  • Workflows that touch many systems.
  • Sensitive data that needs stricter access rules and records.
  • A third workflow in the same sprint.
  • A custom app instead of email, Slack or Teams.

Costs less when

  • Data already in one place (Google Workspace, Microsoft 365 or a CRM).
  • Approvals by email or chat rather than a new app.
  • One workflow done properly first.
  • A named person on your side with time for the pilot.

Monthly running costs

Paid by you, directly to each provider, in your own accounts. I add no markup and no hosting fee.

WhatHow it worksPaid to
AI model usageBilled by the provider in your account, capped per workflow.Anthropic / OpenAI
HostingSmall volumes fit Cloudflare’s free plan; its paid plan is budgeted if volume grows.Cloudflare
Your existing toolsAgents work inside the email, drive and CRM you already pay for. No new subscriptions unless we agree one.Nobody new

Every workflow has a written “never allowed” list and a spending cap. Both are in the manual, and only you can change them.

Instant estimate · no email needed

What would yours cost?

Answer up to five questions about your AI Agents & Automation Sprint. You get a price, a range whose top is a binding ceiling for that scope, and my next free start weeks. A 30% deposit reserves one, and comes back in full if we don’t agree on the scope.

10 Handover

What you own when we're done.

  • Code, AI instructions and tests your GitHub account
  • The running system, its data and its keys your Cloudflare account
  • AI provider accounts and access keys in your name
  • The record of every run your database
  • Manual and “never allowed” list your GitHub
  • Recorded walkthrough shared drive

No licence fees to me, no lock-in, no proprietary platform. If I disappeared tomorrow, any competent developer could pick it up from the written instructions.

11 Under the hood

For your web developer, the technical detail.

How it’s built, for whoever looks after your systems.

Show the technical detail
Where it runs
Cloudflare Workers, Queues and Cron Triggers in your Cloudflare account; code, prompts and tests in your GitHub organisation; run logs in your database.
Models and integrations
Anthropic or OpenAI business APIs on your keys; Microsoft 365, Google Workspace and CRMs through their official APIs. Deterministic code where rules suffice, a model only where judgement is needed.
Guards
Per workflow: a “never allowed” list, human approval before anything is sent, spent, published or promised, and a hard monthly spend cap that halts the run.
Logs and tests
Each run logged with input, output, cost and approver. Regression tests from captured real cases catch drift after a model or prompt change.
MCP (optional)
Internal data exposed to your team’s AI assistants as scoped, logged MCP tools, within existing permissions.

12 Questions

Straight answers to fair questions.

Is our data used to train AI models?

Not by default. The major AI providers’ paid business services don’t train on your data unless you opt in. The workflows run under your own accounts, so you can check that setting yourself, and I’ll show you where.

How is this different from Zapier, Make or n8n?

Those tools move data between apps by fixed rules, and do it well. An agent reads, judges and drafts. I use fixed rules where they’re enough and AI only where judgement is needed. If Zapier solves it, I’ll say so.

What happens when an agent gets something wrong?

A person catches it before it matters: anything that leaves the building waits for approval. Every run is recorded, and tests built from real examples catch new errors early.

What exactly is an internal knowledge agent?

A private assistant that answers staff questions from your own documents, with the source under every answer and your existing permissions respected.

Does it work with Microsoft 365, Google Workspace or our CRM?

Usually, yes, through their official connections. I check access in week one, before anything is promised.

What happens after the 30 days?

Run the workflows yourself with the manual, start a second sprint, or keep me on through the AI Department for monitoring and changes.

Rather talk it through first?

20 minutes with me, not a sales rep. You leave with a clear next step — whether or not we work together.