Fixing a Years-Old Pothole with Claude: How AI Solved a Municipal Problem in Two Days

A pothole that had plagued an office in the Netherlands for years was finally repaired within two days after a user leveraged Claude to push the fix through. According to a Reddit user's account, the individual uploaded 6 photos of the pothole to Claude, and the AI reconstructed a 3D model and simulated the impact force of a vehicle driving over it at 30 km/h. The resulting video was submitted to the city council. As a result, the pothole that had gone unresolved for years was patched within two days.

From Photos to 3D Model: How Claude "Sees" a Pothole

Traditional complaints are usually just a text description or a flat photo, making it hard for municipal departments to assess the actual hazard. This user took a completely different approach: they fed 6 photos of the pothole from different angles into Claude, letting the AI reconstruct the pothole's three-dimensional structure based on visual information.

The key here is that Claude can not only identify the pothole in the images, but also infer its depth and the extent of edge damage, building an interactive 3D model. Compared to manual measurement, this approach eliminates the need for on-site inspection and avoids the information loss caused by subjective descriptions.

Reconstructing 3D structure from multiple 2D photos falls under the category of Multi-View Stereo (MVS) and Neural Radiance Fields (NeRF)-type techniques in computer vision. The basic principle is: when the same object is photographed from different angles, parallax causes corresponding pixels to shift, and algorithms match these feature points to infer camera positions and scene depth information. In the past two years, methods based on diffusion models and large-scale vision-language models have further lowered the barrier—no longer requiring professional photogrammetry software or precise camera calibration. A few photos taken with an ordinary phone can be fed into the model, and the AI infers the geometric structure. It should be noted here that 3D models inferred by large vision models are "reasonable estimates" rather than measurement-grade precision. Their depth data may contain deviations—sufficient for demonstration and persuasion, but still requiring human verification if used for engineering acceptance.

30 km/h Impact Simulation: Making the Case with Physical Evidence

Simply showing a 3D model wasn't persuasive enough. The user further had Claude simulate the forces acting on a car driving over the pothole at 30 km/h. This speed is a common urban road speed limit and also the condition under which potholes most easily cause tire damage, suspension failure, or even accidents.

The simulation results visually demonstrated the potential threat of impact forces to vehicles and pedestrians. What the city council received was no longer a complaint letter saying "please fix the road," but a risk report with physics simulation. This level of evidence was clearly more effective than years of accumulated complaint records.

The impact force when a vehicle drives over a pothole is essentially the instantaneous load caused by the tire colliding with the pothole edge. A simple estimate: a 1.5-ton car traveling at 30 km/h (about 8.3 m/s) through a pothole 5 cm deep—the tire is forced to change direction of motion in an extremely short time, producing vertical acceleration that can reach several times the acceleration of gravity, with corresponding impact forces potentially exceeding several times the vehicle's weight. This impact is partially absorbed by the suspension system, but repeated or high-speed traversal can cause tire bulging, wheel rim deformation, shock absorber leaks, and in severe cases, blowouts or loss of vehicle control. For two-wheeled vehicles (bicycles, motorcycles), the same pothole can directly cause riders to fall and be injured. This is precisely the core basis for municipal departments when assessing road maintenance priorities—the danger of a pothole lies not just in "uneven pavement," but in the quantifiable safety risks it poses to different road users.

Two-Day Repair: How AI Compressed the Municipal Process

According to the original post, after the video was submitted to the city council, the pothole was repaired within two days. This speed is quite rare in routine municipal maintenance—something that years of complaints couldn't push through was accomplished by an AI-generated simulation report.

The underlying logic isn't complicated: municipal departments face a large number of complaints every day, and prioritization often depends on hazard assessment. When a report clearly demonstrates specific risks (rather than vague grumbling), decision-makers can more easily flag it as urgent. Claude's role here is translating vague "dissatisfaction" into quantifiable "danger."

Municipal maintenance prioritization typically follows a risk management framework: first assess the probability and severity of hazards, then rank them combined with repair costs and pedestrian/vehicle traffic density. In the Netherlands and other European countries, local governments generally have road condition grading standards and statutory response time limits—for example, high-risk defects must be addressed within days, while low-risk ones can be included in annual plans. The problem is that ordinary citizens' complaints often lack structured hazard evidence and can only enter the "pending assessment" queue. When a report provides specific physics simulation data, clearly identifies risk levels and potential accident consequences, it effectively completes the preliminary risk assessment work for the municipal department, making it easier to trigger a fast-track response channel. In other words, AI doesn't bypass bureaucratic processes—it re-expresses the same demand in a language that decision-makers are familiar with and trust: data and visualization.

A New Paradigm for Citizen AI Applications

This case sparked discussion on Reddit and X, with the core not being how complex the technology is, but how it's used. The original poster joked: "Petitions are for people who don't have Claude." This statement captures the key point—AI is lowering the barrier for ordinary people to participate in public affairs.

In the past, pushing for municipal repairs required organizing signatures, repeated complaints, or even legal action. Now, one person, one phone, 6 photos, plus an AI model, might accomplish what previously required collective action. This is not just an efficiency improvement, but a shift in how citizens take action.

Of course, this approach has limitations: it depends on municipal departments being willing to receive and value digital evidence. But at least in this Dutch case, AI-generated 3D simulation successfully penetrated bureaucratic inertia. For more similar scenarios—damaged sidewalks, dangerous intersections, repeatedly flooded road sections—this method may be replicable.