Ask a construction CEO if their business is ready for AI, and most will say yes. Ask them how many pieces of equipment they own and watch the pause that follows. This gap between confidence and visibility is the subject of our conversation this week with David Gal, VP of Product and Engineering for Connected Equipment at Samsara, the fleet and asset management platform running underneath a large share of the country's construction operations.

Samsara's telematics and asset-tracking hardware sits inside vehicle fleets, heavy equipment, and small tools across thousands of construction companies, from regional contractors to some of the largest privately held builders in the country. David's team watches that data at scale, which gives his read on AI readiness a level of granularity most executives don't have: he can see, account by account, which contractors have digitized their operations and which are still running on paper.

This vantage point shapes his central claim: most of the industry is nowhere near ready for agentic AI because it hasn't finished the much less exciting work that comes first.

Construction's AI readiness problem is fundamentally a data problem: most contractors can't fully account for what they own. David told us:

It's a fair description of an industry-wide gap, built up over decades of doing things a certain way. A large contractor can carry hundreds of thousands of individual assets, everything from a bucket attachment to an excavator, and most of that fleet has never been digitized. Where an asset sits, whether it needs maintenance, and, in some cases, whether the company still owns it are all questions that go unanswered.

We push back on AI hype in this newsletter often enough that David's framing is worth taking seriously precisely because it avoids the hype. His argument is that AI agents need an operating substrate to act on, and most construction organizations haven't built one yet. Automating a decision about equipment you can't locate is simply not possible.

David put a number on the gap: he estimates roughly 75% of the construction industry is still in what Samsara internally calls "Phase 0." The rest of the field breaks down like this:

The broader theft and loss numbers make the case for digitization on their own. Industry estimates put annual U.S. construction equipment theft losses at $300 million to $1 billion a year, with fewer than a quarter of stolen machines ever recovered. David's own customer data backs this up at the account level: Samsara has documented that Maxim Crane saved $13 million in maintenance costs after digitizing its fleet, with Sterling Crane Canada saving more than $3 million combined across on-road and off-road maintenance and replacement.

David cited a similar figure on the maintenance side alone in our conversation, describing Sterling Crane's shift to proactive maintenance as saving "roughly a half million dollars" annually in that category specifically.

The pattern across David's examples holds up. Digitization typically starts as a safety or cost initiative, usually vehicles first, with the AI layer arriving later as a natural extension of data the organization already collects.

For firms still in Phase 0, the operational cost is concrete:

David's advice to leadership teams is to resist the instinct to fix everything at once. He recommends picking the single pain point causing the most damage, whether that's safety, fuel, or theft, and solving it before expanding into other asset classes.

Sequencing is important because Phase 2 insight compounds once an organization has Phase 1 visibility to build on. David described how stakeholders beyond the original safety or operations buyer start pulling value out of the same dataset once it exists: finance teams optimizing tax filings around asset location, procurement teams deciding what to buy next, maintenance shops running service schedules built on data that used to sit unused. The data infrastructure becomes useful to departments that never asked for it in the first place.

David also addressed a common worry among operators considering camera-based monitoring: driver resistance. His experience is that skepticism about being watched tends to flip within about a week, once a driver is exonerated from an at-fault accident by footage proving they weren't at fault. From that point on, drivers tend to see the camera as protection, and the resistance largely disappears.

The agentic layer David describes is already shipping in narrow form:

Both are examples of targeted automation sitting on top of a dataset that already exists, well short of full autonomy.

Full jobsite autonomy sits further out than warehouse or trucking autonomy, in David's view, because a jobsite carries at least two orders of magnitude more complexity and variability than a controlled facility. This lines up with what we heard from Craig Rupp on the physics case for autonomy a few months back: autonomy multiplies time, but only once the surrounding operation can actually absorb it.

David frames Samsara's role as an orchestration layer, the operating system coordinating autonomous and non-autonomous equipment alike, regardless of which OEM builds the machine. Whether or not that framing survives as autonomous construction equipment matures, the underlying logic holds: autonomy needs a coordination layer, and coordination needs data that most of the industry doesn't yet have. It's a theme running through the broader shift toward agentic scheduling tools across the industry this year.

David's closing suggestion for executives is a single diagnostic question: ask someone on your team how much you spent last year on small equipment. If the answer comes back uncertain, that gap is the real starting point, well ahead of any AI roadmap sitting in a slide deck. The firms pulling ahead in this industry are simply the ones who can already answer that question.

POWERED BY: