MISTER CONTECH
ANALYSIS · ALI EL HARIRI · 13 SEPTEMBER 2026

Construction AI:
potential is not realised output.

Anthropic’s March 5, 2026 study distinguishes theoretical capability from observed exposure in Claude usage. It does not measure the share of work actually performed by AI inside engineering firms.

What this means for construction leaders

An occupation-level indicator does not establish your savings. Reading documents, calculating quantities and approving technical choices require different data and controls. Start with a real process and measure before and after, including review time.

A practical evaluation

  1. Choose a representative project and retain document revisions.
  2. Define the expected output and errors that prevent its use.
  3. Measure preparation, processing, review and corrections, with and without the tool.
  4. Decide the usable scope from observed results.

Read the original Anthropic study ↗

YOUR FIRST USE CASE

Which process should
we start with?

In 30 minutes, clarify the expected outcome, the people involved and the conditions for a first scope.

Book 30 minutes ↗