Every project engineer knows the Friday afternoon dread. Four thirty rolls around, the site has emptied out, and somewhere between wrapping up RFIs and pretending the week is over there is a ten to fifteen page owner report waiting to be built by hand: pulling last week's photo angles, matching them to this week's, writing captions nobody will read closely unless something goes wrong, then formatting the whole thing into a deck due first thing Monday. Most owners skim it in ninety seconds. The one week they read every page carefully is the week you really did not want them to.

That specific chore, and the reality capture cameras that quietly document it, are exactly what OpenSpace spent its annual Waypoint customer event trying to make obsolete. On September 10, OpenSpace unveiled the next generation of its Visual Intelligence Platform at Waypoint 2026, its virtual customer summit, introducing what the company is calling spatial agents: AI agents built on top of its reality capture data that do not just show teams what happened on site, but act on it. CEO Jeevan Kalanithi framed the shift plainly, saying construction does not need another disconnected tool that asks teams to manually input information into a form, and that the next phase of the industry will be defined by platforms that understand the physical world and make that intelligence available wherever teams already work.

The mechanics behind that pitch are built on what OpenSpace calls its Spatial AI Engine, which organizes years of 360 degree images, drone captures, and BIM data by time and place and understands what is actually in them. Two features announced at Waypoint make that concrete. AI Autolocation 2.0 with Live Location turns any smartphone into a real time indoor positioning system on an active jobsite, solving what OpenSpace argues is the most important and most commonly missing piece of field data: exactly where a given photo was taken. AI Voice Notes lets a superintendent talk through a punch list walk instead of typing it, with the Spatial AI Engine logging each issue into a Field Note and filling in fields like assignee and due date automatically. Both are aimed at the same target: getting information off a person's head and a photo roll and into structured, usable data without anyone opening a form.

The owner report problem is not hypothetical, and OpenSpace has a customer case study to prove it. At NOVO Construction, a project engineer who used to spend three to four hours every Friday assembling that weekly report now spends about thirty minutes polishing an AI generated draft that arrives eighty five to ninety percent complete, built from the site's own tagged photo history rather than from scratch. That is the pattern spatial agents are meant to generalize: OpenSpace demoed agents that can revisit a previously flagged issue in newer imagery to check whether it was actually resolved, and, more ambitiously, compare a subcontractor's pay application line by line against verified OpenSpace Track progress data before queuing it for a human to approve. The agent still does not cut the check. It just makes sure the person who does is looking at a pre-checked number instead of a raw claim.

That last detail is the one worth sitting with, because it is the same shape of problem BidForgeAI works on the bidding side: an AI system's real job is rarely to make the final call, it is to do the tedious cross-checking so a human's final call takes minutes instead of hours. OpenSpace is applying that logic to pay applications and punch lists using data it captures itself. What makes the bet interesting is the "open" framing OpenSpace has attached to the announcement, describing the new capability as an interoperable Agent Ecosystem rather than a closed feature set, which matters a great deal if reality capture data is going to actually feed scheduling, cost, and procurement agents built by other vendors instead of staying locked inside one company's dashboard.

Whether that openness holds up once OpenSpace has more agent workflows to sell is the fair question to ask a year from now. For today, the more immediate tell is simpler: a decade into reality capture being sold as a way to see the jobsite, the pitch has quietly shifted to a platform that also does something about what it sees, starting with the report nobody wanted to write on a Friday afternoon.

Sources