For firms and brands where a wrong answer is expensive
Drafted from your records. Sent on your OK.
I build AI agents that draft quotes, client replies and status updates from your own records. Each draft waits in a queue for a person you name. Every step is logged.
I'm Eyal Gantz, one senior developer in New York, and I've been building software for 23 years. I've also been the AI and automation engineer for Salem Wine & Liquor since May 2026. Read the case study on eyalgantz.com.
Then a $950 review, credited toward the build. For the first five clients, the Build balance is due only when it's live.
A quote request, drafted and held for approval
- Form
A past client asks for a quote on the same work as last year
- AI
Pulled the last job, the rate sheet and open invoices
- AI
One line item has no current rate. Asked, did not guess
- AI
Drafted the quote, gap marked. Held in your queue
- You
You fill in the rate, approve and send.
Where the drafting backs up today.
Old quotes get copied forward
Someone copies last year's quote, misses the changed rate, and the client holds you to the number.
Status updates wait on one person
The update goes out late because the only one who knows where the job stands is out of the office.
AI tools that guess get switched off
The first time one tells a client something nobody approved, it is turned off and the drafting lands back on whoever has time.
What I build
An agent for one job at a time, with written limits on what it may read and what it may change.
Drafts from your own records
Quotes, client replies, status updates and intake summaries written from your files, your CRM and past work, in your wording.
An approval queue with a name on it
Each draft lands in a queue for the person you name. They approve, edit or reject it. A draft in that queue is not sent until they do.
Reading and changing, kept apart
A written list of what the agent may read and the few things it may change, such as adding a note or a tag. A send, a delete, a price or a promise waits for a person you name.
A log of every step
What it read, what it drafted, who approved it and when. Open any draft and you see the records behind it.
It asks when it is unsure
A missing rate, two records that disagree or a request outside its job stops the draft. It asks the named person a plain question instead of guessing.
A supervised start
It begins with every action held for review. You widen its limits only after you have seen its drafts on real work, and every step stays logged.
Works with what you already use
- Google Workspace
- Microsoft 365
- HubSpot
- Salesforce
- QuickBooks
- Slack
- Microsoft Teams
- Notion
Four kinds of draft. One approval queue.
Each agent reads your records, writes its draft and puts it in the same queue, with the log beside it.
Fixed prices. Stop after any step.
Written scope, nothing locked in. Every step stands on its own.
Review
$950
After a free 15-minute call: a 60-minute session and a written plan for your top three jobs, each priced. You bring the workflow. No logins, no client or customer data. Credited toward the build within 30 days.
Build
from $4,500
One job, built to a written scope at a price fixed before I start, and switched on only when you say. Every account stays in your name, and a handover doc ships with it, so any developer can take it over.
Run
from $1,200 /mo
Month to month, cancel any time. I monitor it, fix it, keep the wording and source documents current, and go through the logs with you monthly. AI and hosting fees are billed to accounts in your name at the provider's own price, so there is nothing for me to mark up.
Software
from $15,000
Your own internal tool, client portal or Shopify app, when one Build is not enough. Quoted after the review. Larger platforms are scoped with me directly at eyalgantz.com.
These prices cover one workflow inside the tools you already use. Work that spans a whole store, a custom platform, or the technical side of a company is quoted with me directly at eyalgantz.com, where it starts with a $3,500 store audit or a sprint from $7,500. The first step is the same free 15-minute call.
If it fits, the $950 review comes next, credited toward the build.
Questions I get on the first call
What does the agent do on its own?
It reads the records, writes the draft and logs what it did. Everything else, including anything with a price or a promise, waits for a person you name.
What can it read, and what can it change?
Only what the written access list says. I write that list with you before I build, and the access the agent gets is set to match it.
How is this different from giving our staff a chat tool?
A chat tool knows only what someone pastes in, and its answers are not tied to your records. An agent reads your records directly, works inside the limits you set and logs each step.
Which job should the first agent take on?
The one where drafts pile up and the rules are already written somewhere, like repeat quotes or weekly client updates. I pick it with you on the review call, and the agent, the queue and the log run in accounts in your name.
Tell me which drafts pile up. I'll scope the agent.
- First. I read it myself and write back with a time for a free 15-minute call. Keep it general. No client or customer data needed.
- On the call. We find the one job worth building. No charge.
- After. If it fits, the review: 60 minutes and a written plan for your top three jobs. $950, credited toward the build within 30 days. You decide what, if anything, to build.