Custom AI agent development

Three weeks to a fixed price and a fixed schedule, then you decide.

Three weeks of verification on your own data, ending in a scope, a price, and a delivery date that do not move. You see all of it before you commit to the build — and stopping there is a legitimate outcome.

Remote-first: requirements, data handover, and verification all run online. On-site visits are an option, not a premise.

What three weeks produces

  1. 1. Workflow inventoryWeek 1
  2. 2. Automation candidates, rankedWeek 1
  3. 3. Accuracy report on your real dataWeek 2
  4. 4. ROI estimate — hours saved, payback periodWeek 2
  5. 5. Architecture decision recordWeek 3
  6. 6. Fixed quote and delivery plan for the buildWeek 3
DurationThree weeks

Priced per engagement. If you go ahead with the build, the verification fee is credited against it.

How we price and commit

Fixed bid, not time and materials.

Most vendors quote a range, bill by the hour, and discover the scope as they go. We fix the scope, the price, and the schedule first, and carry the risk of having got them wrong.

  1. The schedule is fixed before we start

    Verification takes three weeks. The build is committed in days, in writing, the moment requirements are set. We never start on "we will know once we try" — if something is genuinely unknown, we verify it first.

  2. The total price comes before the first line of code

    The verification deliverables include a fixed quote and a delivery plan for the build. The number does not grow mid-project. Cut a requirement and we show you what it does to the total, on the call.

  3. Stopping after three weeks is a fine outcome

    If the return does not justify the build, stop there. A verification that ends with "production quoted separately" leaves you with nothing to decide on. Ours ends with the quote.

The rest, briefly

Where we are useful.

We work on the paper-and-spreadsheet processes that resisted the last round of automation: handwritten reports, faxed orders, inspection records, delivery notes. Dirty real-world forms become checked structured data, and that data lands in the production, sales, warehouse, or accounting system you already run — by API where one exists, by file exchange or screen automation where none does. Where the data cannot leave the building, the agent runs entirely inside your network. We start with one process in one team rather than a company-wide programme, we test on your real forms rather than clean samples, and we stay through adoption: trial running on the floor and the tuning it takes to make the new way stick are part of the engagement, not a change order. Work is priced per engagement.

Next step

Start with the rough numbers.

Talk to us before your requirements are settled — that is the normal case, not a problem. If the work is not a fit for us, we will say so on the first call.

A 30-minute call

Tell us what you want automated. We will tell you whether an AI agent solves it, roughly what it costs, and how long it takes.

Book a 30-minute call →

Data that cannot leave the building

If the work has to run inside a closed network, start there — that page covers the constraints and how a project runs around them.

See the on-premise option →

You would rather try a product first

Before building anything, it is worth checking whether an off-the-shelf product covers it. If the work lives in spreadsheets, it often does.

Ichiyomi Sheet →
Contact →