Eight AI agents run a painting contractor's office

Eight AI agents run a painting contractor's office

At J&B Tasoitus ja Maalaus Oy, cost estimating, contract documents and site messaging in three languages all run through AI agents. The office has one person in it.

TekoälySovellukset
Project

What we built

What the agents do

J&B Tasoitus ja Maalaus Oy is a painting contractor doing skim coating and painting for construction companies in the Helsinki region. The office work is the same as in a much larger firm. Invitations to tender have to be read, quantities measured, prices justified, contract documents written and the site kept informed. There is one person doing it.

We built the company eight AI agents, each with its own job and its own sources. They work from the same project folder: an agent opens the documents for a site, reads them in full and takes the work from there.

  • Cost estimating. The agent reads the building and room specifications in full, picks out the quantities, prices the work with the company's own hourly rates, adds a risk allowance and a margin, and writes the finished draft quote straight into the company's own estimating app. The owner opens the draft, checks the numbers and hits PDF and send.
  • Quantity take-off. A quantity from a single source is not good enough to price against. The agent takes the client's bill of quantities, measures the drawing itself, and then runs a sanity check on whether the number is physically possible. On one site the drawing title block was missing the scale entirely, and the wrong assumption would have given four times the real area. The spacing of the bench rows revealed the correct scale before a price had been calculated.
  • A check before the price. The agent compares the tender documents against each other and lists the contradictions, the missing documents and the places where the scope of work is left open. No price goes out before this has been run, because that is exactly where a contractor's risk comes from.
  • Contract documents. Variation notices, complaints, readings of the contract boundary, weekly progress notices and self-inspection lists all come out of the same project memory, so they carry the right dates, the right contract clauses and the right job number.
  • Payment schedules. The agent builds the payment schedule from the contract details of the site and pushes it into the time-tracking app, where the client signs off each instalment.
  • Technical solutions. Coating build-ups, substrate requirements and standard references are looked up in the company's own manuals. The agent does not invent a standard clause, a product code or a tolerance. If something is missing, it says so and flags the point for checking.
  • Site messaging. Messages to the crew are written in Finnish, Estonian and Russian. The foreman writes in the WhatsApp group during the week about what was done on which floor, and the agent turns that into the progress notice that goes to the client.
  • Post-costing. When a site is finished, the agent pulls the hours actually booked from the time records, compares them with the hours estimated in the quote and writes down what the work really cost per square metre and per working hour. The next quote for the same client is calculated from those numbers instead of from memory.

One day

In the morning an invitation to tender arrives from the client with a work specification attached. The owner opens the project folder and says one sentence to the agent. The agent reads the specifications, lists the contradictions, measures the quantities, prices the work and writes the draft quote into the estimating app. About ten minutes later the draft is ready to be checked.

Around midday the site calls: the Estonian crew needs instructions for tomorrow. The owner dictates it in Finnish. The message goes out in Finnish and Estonian in about ten seconds, with the same content in both languages.

In the afternoon the client asks whether one work stage is included in the contract. The agent opens the contract documents, finds the clause where the boundary was drawn and writes an answer that cites the page number. The owner reads it through and sends it.

What this replaced

The same workload used to need an office assistant, a cost estimator and a technical assistant. Now one person does it with the agents. The agents decide nothing. The price, the scope and the send button stay in human hands, and every draft is checked before it goes to the client.

Speed is the smallest of these benefits. The bigger one is that the same information is in the same place every time. The memory of a site survives even when the conversation changes or a month goes by in between.

Result

A quotation starts from the specification in about ten minutes, a message to a worker turns into Estonian and Russian in ten seconds, and the hours actually booked on every site calibrate the pricing of the next quote.

Next step

Do you have a similar project?

Tell us what you have in mind. We reply within one working day.

A painting contractor’s AI agents — Pohja Studio