AI automation with a person in the loop, built for a fixed price.

I'm Johny Patrick. I build AI workflows, browser extensions and internal tools for SaaS and service businesses: support, billing, onboarding and sales follow-up. Anything that reaches a customer or changes a record waits for a person to approve it. A first look at one workflow is free, and paid work starts at $1,500.

Free, no slides. You leave with a scope and a price, or a plain "this isn't a fit".

I built Support Coach, a reply checker that is live on the Chrome Web Store, and I run my own work through the same approval pattern.

What I build

Five kinds of build, one way of working. Each one can start with a free teardown of a single workflow.

Support AI

Pre-send reply checks, chat reviews and AI drafts that an agent approves.

  • Flag a risky reply before it is sent.
  • Score finished chats and write the coaching note.
  • Draft answers from your help center for an agent to approve.

Support Coach, my own reply checker, is live on the Chrome Web Store.

The support page, with three demos

Browser extensions

Tools that sit inside the apps your team already uses, so nobody switches tabs.

  • A side panel inside your help desk or CRM.
  • One-click actions on the page your team is already on.
  • Checks that run in the browser, so private text stays there.

Support Coach works inside Zendesk, Intercom and Zoho SalesIQ.

How the extension is built

Agent workflows with approval

Agents that sort, draft and route work, then wait for a person before anything goes out.

  • Triage incoming requests and draft the reply.
  • Prepare refunds, follow-ups or record updates for approval.
  • One approval queue you can clear from your phone.

I run my own work this way: scheduled agents, one approval queue, every decision logged.

The approval queue project

Dashboards and internal tools

Web apps for the queues and numbers your team works from, with sign-in and access per client.

  • A QA dashboard that scores every chat.
  • An internal tool that replaces a shared spreadsheet.
  • Per-client access, so each client sees only their own data.

Support Coach Dashboard is live at supportcoach.io.

The dashboard project

Access and AI-spend reviews

A review of an app you already have: who can read what, which AI features anyone can call, and where spend could run away.

  • Check that one client cannot see another's data.
  • Find AI features that work without signing in.
  • Add spend limits and usage counters.

I ran a security and tenant-isolation audit on my own products before launch.

Ask about a review

Newer services: agents that do real work, with you in charge

These come from the system I run my own work on. Each keeps a person in charge of anything that leaves the building. They are built and tested on my own work and on sample data. None has run for a client yet, so you would be one of the first, and the proposal will say so.

Fixed price, set in writing after a short call.

Core services

Built for you, on your tools.

Agent work queue with approvals

AI agents do real work in steps. You say go before anything starts, and again before anything ships.

Who it's for

Teams that want agents doing real work, not just chatting, without losing track of who did what.

What you get

  • Each job split into steps, picked up by one agent at a time, so nothing runs twice.
  • A go card before work starts and a final card before anything leaves.
  • Work lands as a draft you can preview, checked by a second, independent reviewer.
  • A record of every step, retry and approval.

Proof: I run my own builds through this pattern.

Ask about this: Agent work queue with approvals

Next-move finder for leaders

Finds the most useful next moves in your own records, shows the evidence for each, and is honest about what it does not know.

Who it's for

Founders and CEOs of small teams who have the data but not the time to read all of it.

What you get

  • A short list of opportunities, each with the records behind it.
  • Every point marked as fact, estimate, guess or unknown.
  • The first safe step for today, plus what is blocked and who can unblock it.
  • Read-only on your data. Nothing is sent or changed.

Proof: working today on a sample company.

Ask about this: Next-move finder for leaders

AI capability map

You say what you want done. The map shows what your tools can already do, who may use them, and the safest way there.

Who it's for

Operations leads with AI tools spread across teams and no clear picture of who may do what.

What you get

  • A map of what your AI and tools can already do.
  • Access in steps: show, read, draft, then act only after a person says yes.
  • What is blocked and why, in plain words. A job title or a clever prompt grants nothing.
  • The cheapest safe route to each goal, on a board your team can work from.

Proof: built and tested on sample organisations.

Ask about this: AI capability map

Add-ons

Short reviews, alone or next to any build.

AI operations audit

A read-only look at how your teams work, ranking where money and time leak before you spend on AI.

Who it's for

SaaS and service companies with 10 to 200 staff that want to use AI but do not know where to start.

What you get

  • A ranked leak list in dollars, each with its data and a confidence grade.
  • An automation roadmap: what to build, what it connects to, how long it takes.
  • A one-page decisions sheet, three options at most per decision.
  • Fixed prices from $1,500 for one team.

Proof: the full method on an invented sample company.

See the sample report Ask about this: AI operations audit

Launch readiness review

A check of a new product or offer across everything a launch needs, not just the software.

Who it's for

Founders about to launch, or teams adding a new paid offer.

What you get

  • Eight areas checked: demand, brand, website, content, sales, delivery, pricing and trust.
  • A plain ready or not ready, with the reasons.
  • Nothing counts as done on a plan alone. Each item needs proof.
  • A short list of what to fix first.

Proof: I use the same checklist on my own launches.

Ask about this: Launch readiness review

Agent execution review

A review of how your agents carry out work that takes many steps, so a person can read the plan before it runs.

Who it's for

Teams already building agents that call many tools in a row.

What you get

  • A review of how your agents run tasks today.
  • A plan-first design: the agent writes the whole plan, a person can read it, then it runs.
  • Permission checks on every step, with time and cost limits.
  • A written report with the fixes in order.

Proof: a prototype I built and tested on sample tasks. This is a design review, not a product.

Ask about this: Agent execution review

Who these are for

  • Operations leads. You want several AI agents working together on real work, without losing track of who did what or who approved it.
  • Founders and CEOs. You want the best next move from your own records, with what is fact and what is a guess kept apart.
Can the agents act on their own?

No. They prepare the work: they sort, draft, research and plan. A person approves anything that leaves your business or changes a record. More freedom comes one step at a time, and only after we have measured how often the agents were right.

Not sure which one fits? A 15-minute call is enough to tell.

Book a 15-minute call

How the pieces fit, with a person signing off

Most of what I build follows one pattern, with a person at the centre. Hover or tap any part to see what it connects to, drag it around, or play a sample task through it.

Approval queue

Every proposed action waits here as a card: what the agent wants to do, the draft, and why. A person approves, edits or rejects it, on a phone or at a desk.

Used in: Support AI, Agent workflows with approval, Browser extensions

Where to start, and what it costs

Every price is fixed in writing before any work starts. Where you see a range, the size we agree decides where in it you land.

Packages, typical duration and price
Package Typical time Price

One-Workflow Teardown

48 hoursFree

Send me one repetitive workflow: what it is, how often it runs, and which tool it lives in, with screenshots or a short recording. Within 48 hours you get a written verdict. No call needed. I take four a week.

You get:

  • A written verdict on that workflow.
  • Where AI would help, where it would hurt, and what I would build first.
  • No obligation to buy anything.
48 hours Free

5-Day Automation Blueprint

5 working days$1,500

Two 45-minute audio or screen-share interviews, then five working days on your workflow. I rank what is worth automating using your own numbers, and build a working proof of the top pick.

You get:

  • A map of the workflow.
  • Three automations ranked by hours saved and difficulty.
  • A working proof of the top pick, on sample data.
  • A fixed-price build quote with written acceptance tests.
  • A one-page decision memo.

If the Blueprint does not name at least one automation worth building and show a working proof of it, you get your money back. The fee counts in full toward a build signed within 90 days.

5 working days $1,500

First Agent Live

About 3 weeks$3,500 to $9,500

One workflow running with a person in the loop. For example, a pre-send check on agents' drafts, a drafting agent that waits for approval, or a browser extension inside a tool your team already uses.

You get:

  • The workflow, live, with a person approving before anything reaches a customer.
  • Acceptance tests written first, so you can see it pass.
  • The code in your own repository.
  • A runbook and an audio handover.
  • 30 days of fixes and support after launch.

Small $3,500, standard $6,500, large $9,500. Billed 50% at start, 25% at the midpoint, 25% when the agreed tests pass. If they have not passed 10 working days after the agreed date, that last 25% is waived.

About 3 weeks $3,500 to $9,500

Keep It Running

Month to month$1,500 to $2,500 a month

After a build, I keep the workflow healthy. I read the approval log, adjust prompts and rules when your policies change, and fix what breaks when a connected tool changes. Stop any month.

You get:

  • A monthly review of approvals, edits and rejections.
  • Prompt and rule updates as your policies change.
  • Fixes when a connected tool changes.
  • A short note each month on what changed and why.
Month to month $1,500 to $2,500 a month

Agent Operations Build

About 8 weeksIn your proposal

Two to four connected workflows for one team, one operating dashboard, and 90 days of Keep It Running. Built on the same queue, approval and permission pattern I run my own work on. The price is set in a written proposal, usually after a Blueprint or a first build. I take one a quarter.

You get:

  • Acceptance tests for each workflow.
  • One approval queue across them.
  • Runbooks and a handover for your team.

See the system it is built on

About 8 weeks In your proposal

If the scope changes, we agree a new price before I continue. Keep It Running is billed monthly.

Book a 15-minute call

How it works

  1. A 15-minute call.

    You describe the problem. I tell you whether I can help, which package fits, and roughly what it will cost.

    15 minutes

  2. A scope and a fixed price, in writing.

    One page: what you get, what it costs, what I need from you, and when it ships. You decide with everything in front of you.

    Within three working days

  3. Build, review, handover.

    You see it working on your own examples before it goes live. At the end you get the code, the documentation and a walkthrough.

    Five days to about eight weeks, depending on the package

Things I built and run

  • Support Coach

    A Chrome extension that checks a support agent's reply before it is sent.

    Live on the Chrome Web Store

  • Support Coach Dashboard

    Chat reviews for support teams, scored against a written rubric.

    Live at supportcoach.io

  • Approval queue and agent router

    A person approves before an agent sends, updates or spends anything.

    My own use, not a product

  • An operating system for AI agents

    Several AI models working as one team, with a queue, permissions and a person approving.

    My own use, private repo

Fifteen minutes, and you will know if it's a fit.

Tell me what your team is dealing with. I will tell you whether I can help and what it would cost.

Book a 15-minute call

Prefer to write?

Send a short brief