Engine room · 30-day sprint
In 30 days I take one to three workflows your senior people repeat every week and turn them into AI agents that research, draft and route the work, with a human approving anything that matters.
Typical project €7.5k–15k. Fixed price per sprint.
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 digest, and stops spending at a hard monthly cap the firm sets.
observed 2026-10-05 · engine config default
03 Fit
Built for you if
Not for you if
04 The problem
Look at a week in a senior person’s calendar: researching a lead before the call, answering the same client question again, digging through drives for a document someone wrote two years ago, chasing a client for onboarding details. None of it needs their judgement until the very last step.
That is exactly the shape of work agents are good at: gather, draft, check, route. It’s also why searches for “AI agents for business” grew +210% year on year (Truelogic/DataForSEO, June 2026).
The risk is the opposite failure: an agent that sends, spends or promises something without a person. Every workflow I build has a gate where a human approves, a budget it can’t exceed, and a log of everything it did.
Signs you need this
Not sure which workflow to automate first?
Book 20 minutes and bring your busiest week. You leave knowing which workflow I’d start with, whether or not we work together.
06 Where I suggest starting
Your sprint can be any workflow the audit ranks highest. These three are where I suggest starting, because the hours are obvious and the risk is easy to gate.
Knowledge
Staff ask questions in plain language and get answers from your own documents, contracts, manuals and past emails, with the source linked under every answer.
Onboarding
A guided, autosaving intake that collects everything a new client engagement needs, lets clients pre-fill it from their own website or documents, and hands your team a complete, structured brief.
Speed to lead
Every new enquiry researched before you open it: who they are, what they need, whether they fit, and a drafted reply for you to edit and send.
07 What you get
Every workflow goes into production on your accounts, not a demo on mine.
Your repetitive work mapped and ranked by hours saved, risk and data access, before anything is built.
Running on live work by day 30, chosen by you from the audit.
Anything that sends, spends, publishes or promises waits for a named person.
A hard monthly budget per workflow. When it’s reached, the agent stops; it doesn’t overspend.
Input, output, cost and who approved, kept where you can search them.
Automated tests against captured real cases, so a model update can’t quietly change behaviour.
Internal data exposed to your team’s AI assistants through scoped, logged tools.
How each workflow works, what to do when it fails, and how to switch it off.
08 How it runs
Fixed price, fixed timeline. You choose what gets built and where the human gates sit; the agents never get to decide that.
Agents at work Human gate: nothing moves on until a person approves
Step 1: Days 1–5
Interviews and a look at the real inboxes, drives and tools. Each candidate workflow scored on hours saved, risk and data access.
Human gate You choose which workflows get built.
Step 2: Days 6–10
For each workflow: triggers, data sources, the agent’s job, what it is never allowed to do, where the human gate sits, and its budget cap.
Human gate You approve the design and the “never allowed” list.
Step 3: Days 11–22
Built on your accounts and tested against real captured examples. A working demo every week.
Step 4: Days 23–27
Running on live work with every output approved by your team. Prompts tuned on their feedback.
Human gate Your team approves every output during the pilot.
Step 5: Days 28–30
Runbook, recorded walkthrough and an access review.
Human gate You decide which gates stay manual for good.
09 Proof
The pattern is the same everywhere: agents do the work, a person approves what matters, and a budget guard stops anything from running away.
5 min heartbeat
A daily approval digest with one-click links, an inbound-email research agent that turns names into drafted articles, 649 automated tests and a €100 monthly breaker.
8 edge rate limiters
A budget guard pauses sign-ups, then uploads, before a cloud bill can run away, and emails a daily ops report.
And Cuckoo, my own distribution engine in pilot: agents draft a week of posts, clients approve each one from a phone.
10 Price
Starts at €7,500 · $8,300 per sprint; most land at €7.5k–15k. Fixed price per sprint.
Costs more when
Costs less when
Paid by you, directly to each provider, in your own accounts. I add no markup and no hosting fee.
| What | How it works | Paid to |
|---|---|---|
| AI model usage | Billed by the provider in your account, capped per workflow by a hard monthly budget. | Anthropic / OpenAI |
| Hosting and schedules | Cloudflare Workers, queues and cron. Small volumes fit Cloudflare’s free plan; Workers Paid is budgeted when volume or CPU needs it. | Cloudflare |
| Your existing tools | Agents 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 spend cap. Both are in the runbook, and only you can change them.
11 Handover
No licence fees to me, no lock-in, no proprietary platform. If I disappeared tomorrow, any competent developer could pick it up from the runbook.
12 Questions
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.
Text extracted from the HTML this server sends — no JavaScript executed, the way GPTBot, ClaudeBot and PerplexityBot fetch pages. It is also the Markdown twin I publish for AI agents.
Loading…
<!doctype html> <title>Custom Web Tools & Cookie-Free Websites | AIWebRevolution</title> … <body> <div id="root"></div> <script type="module" src="/assets/index.js"></script> </body> → 0 words of page content. Beautiful to people. Blank to ChatGPT.
Two minutes · personal reply