AI implementation studio · Los Angeles
One workflow, one agent. Working in 30 days.
An AI agent is software that does a repeated task for your team, such as answering the phone or sorting the inbox. A $2,000 review finds the one workflow (a task your team repeats) worth handing to an agent. A fixed-price 30-day build puts the agent to work on it.
Three ways to work together
Review
$2,000
An AI Agent Architecture Review of one workflow. If you sign a build within 30 days of delivery, the $2,000 comes off the build price.
Build
$15K–$25K
The Agent Build: we make the agent and have it doing that one task within 30 days, at a fixed price.
Run & Improve
$3K–$6K/mo
A monthly plan after launch: we watch the agent, fix what goes wrong, and add new tasks one at a time.
A working example you can try

Live example
Harvest Plan Lab
Open the working example and try it.

Video
Harvest Plan Lab, explained
A short walk through what the example does.
Planning horizon
5 months
Harvest Plan Lab plans 65 grow zones five months ahead of the menu, then tests the plan against 2,000 simulated futures. The data is synthetic.
How it works
01
Review one workflow
A $2,000 architecture review of a single workflow. If you sign a build within 30 days of delivery, the $2,000 comes off the build price.
02
Build it in 30 days
A 30-day Agent Build puts the agent to work on that workflow, at a fixed price of $15,000 to $25,000.
03
Run and improve
The Run & Improve retainer covers monitoring, iteration and new workflows, one at a time.
Latest writing
Writing on AI systems, agents, and what actually changes when software starts doing the work.

November 21, 2026
Rules That Run, Part 3: Write the Rule, Then Wire It
How I turn a rule from an instruction file into a Claude Code hook: a gate with a logged override, a weekly count of how often the override gets used, and one dispatcher process that fails open.

November 20, 2026
Rules That Run, Part 2: The Agents Ate the Week
In the week starting October 1, 2026, subagents were 63% of my Claude Code usage and the weekly cap had been hit three weeks running. These are the subagent rules I shipped on October 4, and the hook that enforces them.

November 19, 2026
Rules That Run, Part 1: The 93,000-Token Hello
A plain claude -p call from a project folder loaded about 93,000 tokens of harness before it said a word. Here is the lean wrapper that cut a one-word reply to 438 input tokens, and why every unattended call now picks a model by role instead of by id.
Books by Ariel Agor
Three books on running a business with AI without handing it your judgment.
Frequently asked questions
What owners ask before the review.
What kind of workflow is a good fit?
How long does a typical project take?
What's the investment range for your services?
Do you provide ongoing support after deployment?
Start with the review.
A $2,000 architecture review of one workflow. If you sign a build within 30 days of delivery, the $2,000 comes off the build price. Not sure yet? Chat with the assistant to schedule a free strategy call.


