Sometimes, artificial intelligence isn’t all-knowing and can actually make jobsites less safe.

AI is, inarguably, changing everything, and has proven to be remarkably effective for certain tasks in certain industries. But there remain some areas where total reliance on AI is far too risky: Construction jobsite safety is one of them.

When used to achieve general safety goals, such as real-time monitoring with video-based AI to detect risks, it can be a valuable tool if there are frontline supervisors available to respond to alerts. The inconsistent nature of a construction jobsite, where new problems and dangers can happen at any time, and workers often make split-second decisions to save themselves and their colleagues from injury or in some cases take unnecessary risks, makes relying on video-based AI highly impractical, and due to often very complex work conditions, prone to inaccuracies.

Here are five reasons why video-based AI has limitations in construction safety:

1. The Full Time Power That is Needed to Support Video-Based AI Isn’t Available in 90% of the Highest Hazard Construction Settings

AI is very energy consuming and in order to process video-based AI, full time power is required. Given that 90% of the work areas that are associated with the highest hazard construction work do not have full time dedicated power available, the battery-powered fixed-point camera systems that are needed in the highest risk work zones would require changing batteries frequently, which isn’t practical or even viable in some work areas.

2. Environmental Erraticism and Data Quality and Consistency

High-performing video-based AI depends on large volumes of consistent, high-quality data. Manufacturing naturally produces uniform datasets, improving model performance. In healthcare, hybrid-AI and human-intelligence approaches help ensure accuracy in complex situations.

Construction video data is often fragmented and inconsistent: Camera angles change regularly and are often far away from work activities; different weather patterns and lighting can impact clarity; obstructions are common; and high hazard work behaviors are unpredictable. These factors reduce data quality and increase the likelihood of missed or inaccurate video-based AI detections.

3. Roaming Video-Based AI Systems Such as Robots or Helmet Cams Miss 90% of the Story and Cannot Access the Highest Hazard Work Zones

As with almost any work environment, worker behaviors are heavily influenced when a manager is present. Roaming video-based AI systems inherently change worker behaviors when they are present but are unable to evaluate or impact 90% of the work activities when such roaming systems aren’t present. Fixed-point camera systems that record 100% of the work activities and which are then “smart sampled” by expert human video annotation construction specialists are much more impactful. Robotic camera systems are also unable to access the highest hazard work areas especially on the top floors of buildings as they are being constructed, and helmet-fixed camera systems are mostly worn by staff who aren’t permitted to access these highest risk work areas.

4. Unsafe Worker Behaviors Often Occur in a Just a Few Seconds Making Video-Based AI “Real-Time” Alerts Ineffective and Can Overwhelm Frontline Supervisors With Too Many Alerts, a Subset of Which Will Be Inaccurate

Many of the extremely dangerous worker behaviors in construction take place in 1-5 seconds, making real-time video-based AI alerts extremely impractical. Overwhelming a single safety professional on a construction project who oversees safety for 100+ workers with a barrage of real-time video-based AI alerts will almost never result in a worker being notified in real time of the high-risk action they just took.  Without human video annotation specialists being used to ensure that all videos sent to frontline supervisors are accurate, there are going to be many inaccurate videos sent which will quickly erode the confidence in the video data integrity.  Furthermore, “smart sampling” and sending a small subset of the overall coaching clips can drive safety performance to extremely high levels without overburdening safety professionals with too many alerts.

5. Video-Based AI Solutions Do Not Provide Construction Clients With the Consultative Video Coaching Services Needed to Truly and Significantly Impact the Safety Culture on Projects

There are countless examples over the last 50 years of frontline supervisors in many industries being overloaded with automated data streams that are often underutilized and wind up not materially improving safety, quality or productivity. Building a safety culture that can drive the 97%-100% compliance levels needed to significantly reduce the risk of workplace injuries requires a highly consultative coaching approach, which video-based AI alone cannot accomplish.

Over time, video‑based AI can contribute meaningfully to construction, but it should complement versus replace time-tested, human‑led consultative video coaching methods. Its greatest value lies in supporting human teams by:

• Providing additional visibility across sites in the small number of work areas where there is full-time power available
• Capturing rare, but potentially high-risk video-based AI events with fully powered cameras that can be reviewed first by human video annotation specialists before the video events are sent to frontline supervisors

A balanced approach of combining AI capabilities with experienced human oversight offers the most practical path forward.

While video‑based AI is often presented as a compelling solution for construction safety, its effectiveness is limited by the dynamic and unpredictable nature of jobsites. Unlike industries such as manufacturing and healthcare, construction lacks the consistency and stable power infrastructure needed for AI to perform reliably. As a result, the greatest impact will come from selectively combining video‑based AI with human video annotation specialists and proven consultative video coaching programs.

SEE ALSO: THE NEWEST MEMBER OF YOUR FACILITIES TEAM? YOUR BUILDING.