Where Grok Bot fell short
An honest list, updated as we go. These came straight out of our project files.
- It couldn't read the native CAD drawings. The architect's DWG file was in a format the bot couldn't open properly, so it needed a PDF export. Ask your designer for PDFs as well as CAD.
- The early budget used an assumed extension size. Before measured drawings existed, the estimate assumed about 85 m². The drawings showed 125.6 m². The method was right but the input was a guess, and the bot labelled it as an assumption. The lesson is to treat any early €/m² figure as a placeholder.
- Floor areas didn't match. The brief's existing ground floor area didn't match the drawing schedule. The bot flagged it, but couldn't resolve it. Only a site measure can.
- Grant figures went out of date. A heat pump grant figure from September was superseded when SEAI changed the scheme later that month. The bot re-checked when asked, but grant figures need re-checking on the day you apply.
- Scaled, not dimensioned. The steel lengths and some cladding areas were scaled from PDFs because the drawings weren't dimensioned. That's fine for checking quotes, but not for fabrication.
- Spreadsheet formulas. One exported spreadsheet showed blank totals in some viewers because formulas weren't pre-calculated. We now check exports in more than one app.
- It doesn't replace sign-off. The bot can't act as Assigned Certifier, PSDP, PSCS, engineer or QS, and it shouldn't pretend to.
Seen something we should add? TO CONFIRM: contact route