Reality capture first, AI claims second: a cautious pilot for corridor and complex site teams

Project managers and project controls leads often face the same problem: too many site records, too many manual handovers, and too little confidence that everyone is working from the same current condition. Based on these two September 2026 vendor items, my editorial advice is to pilot a standard capture baseline first, then test any AI-led workflow claims against that baseline rather than buying the promise up front.

Proposed pilot: pick one linear package or one MEP-heavy area for eight weeks. Inputs: one existing programme update cycle, one recurring design or document-control workflow, and one agreed capture cadence. Checks: whether the team can retrieve the same location/date record quickly, whether planners and field leaders use the same source during reviews, and whether manual document collation time changes from your current baseline. Baseline: your own current turnaround time, current rework in information handling, and current number of separate repositories. That is proposed editorial advice, not a tested result.

What changed

On 8 September 2026, OpenSpace published a vendor blog built around product manager Wesley DuBose’s field-to-tech background and his view that construction complexity has outgrown many traditional workflows in document control and coordination. The article says routine processes such as document control could take hours of manual work, and presents AI and agentic workflows as a way to compress some tasks from hours to minutes. The piece is not a product release note, does not state any region-specific rollout, and does not mention Australia, beta status, or general availability limits. It is best read as a vendor position on workflow direction rather than a verified deployment result in market. OpenSpace: Wesley DuBose on building technology for the complexity of construction

On 2 September 2026, DroneDeploy published a vendor case study on Stacy Witbeck, a California-headquartered US rail contractor, describing a standardised reality-capture workflow across its transit portfolio. The article says the program covers 86 sites and 5,594 flights, using Part 107 pilots, DJI Mavic 3 Enterprise drones with RTK, and combined aerial and ground capture in one platform. It reports 507 flights and 72,678 uploaded images in the previous 30 days. It also lists 1,974 ground walkthroughs without specifying a period. The story is specific to US operations and references Part 107 licensing, so it should not be treated as evidence of Australian regulatory fit or delivery conditions. No Australia availability statement, beta restriction, or region-wide product limitation is provided in the source. DroneDeploy: Stacy Witbeck's standardised reality capture

Why it matters on a project

Taken together, the practical decision is not whether AI is exciting. It is whether your team has a repeatable, trusted capture record that planners, designers, supervisors and controls staff can all reference before layering automation on top.

The DroneDeploy case study is useful because it describes an operating model, not just a future ambition: simple hardware, a licensed pilot model, combined air-and-ground capture, and a shared historical record. For a corridor, spread-out utility package, road job or large services installation, that matters because the controls team needs a consistent date-stamped record before it can reliably compare progress, explain change, or support design queries.

The OpenSpace item matters differently. It is a reminder that manual information workflows still absorb hours, especially where teams are trying to make sure everybody has the same information. But because this source offers vendor commentary rather than measured project outcomes, it should push you toward disciplined testing, not broad assumptions about benefit.

Try this

Use a small proposed two-step pilot.

Step 1: standardise capture. For one live workfront, define who captures, how often, and where records are stored. Keep it deliberately simple. If your work is linear, test whether one historical visual record reduces the time needed to answer “what was there on that date?” If your work is vertical or MEP-dense, test whether the same record improves weekly coordination reviews.

Step 2: test one manual workflow against that capture baseline. Good candidates are document-control collation for site reviews, evidence gathering for design queries, or preparing progress review packs. Measure your current turnaround first, then compare after the pilot. Do not set a savings target from these articles; neither source provides an Australia-specific benchmark you can safely adopt.

Limits and checks

Both sources are vendor publications. The DroneDeploy article contains customer-attributed outcomes, but it still sits inside vendor marketing. The OpenSpace article is more directional again: it discusses complexity, manual work and AI potential without publishing measured before-and-after project data.

There is also no basis here to claim the products integrate with each other, or that either workflow is approved, available, or proven in Australia. The DroneDeploy article is explicitly grounded in US rail-corridor practice and US pilot licensing. The OpenSpace article names a former MEP project manager in Texas and makes no region-specific rollout statement.

So the check is simple: keep your pilot narrow, require a before-and-after baseline from your own project, and separate “easier retrieval of current conditions” from “actual time reduction in a named workflow”. If you cannot evidence both on your job, do not generalise from these vendor stories.

Sources

OpenSpace: Wesley DuBose on building technology for the complexity of construction

DroneDeploy: Stacy Witbeck's standardised reality capture

Next action: nominate one package this week where a date-stamped shared record would remove the most argument, then document your current turnaround before any pilot starts.

Correction, 18 September 2026: the 30-day period applies to flights and uploaded images only. The source does not specify a period for its 1,974 ground walkthroughs.