Welcome to another Applied Report, where I share insights on how companies are adopting AI.
Last time, I shared the 6-agent AI system behind Applied. If you missed it, it’s a great way to understand how our use cases and tools are researched, prioritized, enriched, and translated.
Today, I have a new AI use case to share, this time from my first client in the construction industry. I’ve built multiple agentic systems before, but construction was a first for me. I learned there is a lot of room to optimize processes and put AI to work in this industry.
Building permits should take 10 seconds…
The Problem: Scaling Permit Processing
SAR Facades is an architectural firm based in South Florida. Their core service is creating plans and technical drawings for glazing projects, a large market in the region due to its booming real estate industry, hurricanes, and extreme hot weather.
The firm wanted to expand its services, so the owner added permit processing to their catalog. However, the firm struggled to scale the number of permits it could manage. The process was not straightforward. It required many manual checks, constant follow-ups, and experienced managers to keep everything moving.
Why the Process Was Hard to Scale
For those who are not familiar with the construction industry, almost everything requires a permit. From small updates to individual homes to large public or private infrastructure projects, permits are a core part of how construction moves forward.
Municipalities are responsible for reviewing and approving in a highly manual process. It requires engineers, designers, architects, and regulatory experts to review documents, check requirements, and move each permit through the approval process.
On top of the paperwork and expert review, the technology layer is also fragmented. Most of the process happens through old websites with one thing in common: they are not exactly modern.
No APIs.
The UI/UX is outdated.
Notification systems are inconsistent — sometimes by email, sometimes by SMS, and many times there is no notification at all.
How Permit Approval Is Supposed to Work
In the simplest version, the process looks like this:
The permit requester goes to the municipality website, makes a payment, and submits the required designs, construction information, specifications, and supporting documents.
The municipality receives the documentation. An expert then reviews the submission and approves it.
The approved permit is then sent back to the requester.
Depending on the type of work, a physical inspection may also take place to make sure everything matches.
Once approval is granted, construction can begin.
It sounds simple, but it is not. In many cases, documents are missing, designs contain errors, or the submitted information is incomplete. When that happens, extra steps are required. The municipality may request additional documents, corrections, or clarifications before the permit can move forward. If the requester misses those notifications, the process stalls.
Now multiply that same issue across every permit needed for a project. For large developments like the ones SAR Facades works on, we might be talking about dozens or hundreds of permits for a single project.
Keeping up with the full process requires an entire team just to track permit statuses, catch updates, follow up with municipalities, and reduce delays.
The Solution
The agent was built in a similar way to the system I covered in my previous report. The tricky part was handling local differences. Each municipality has its own website, interface, login flow, notification system, and security requirements. Some have strict access controls. Others include anti-spam checks, bot detection, or location-based restrictions.
The best approach was not to build one generic agent. Instead, we built a specific agent for each municipality, trained to handle the requirements of that particular website. Each agent knows how to log in, navigate the site, find the right permit information, check for updates, and alert the permit manager when action is needed.
The result is a set of specialized agents that monitor permit statuses, reduce manual checking, and help the team react faster when something changes.
The Results
Instead of manually checking each website, agents alert the team when something when a permit needs attention. A human still reviews the issues and submits any required documentation or extra information, but they no longer need to check each permit.
This changes the operating model. Before, the team needed roughly one person for every 10 permits being monitored. With the agent, one person can monitor a much larger permit pipeline while focusing only on the cases that require action.
The main KPIs we are tracking are:
Permits monitored per person: x5 on average
Manual status checks avoided: trending to 0 per permit.
Average response time to municipality updates: from one month to 48 hours.
Missed notifications or delayed responses: trending to 0.
The most important metric is capacity. The goal is not to remove the human from the process, but to let one person manage a much larger number of permits without missing critical updates.
Expanding this use case further
Reducing processing time
The next opportunity, and arguably the biggest, is reducing the time it takes to review permits. This is a huge use case, but it requires cooperation from municipalities. With the right training, an agent could help identify missing documents, flag inconsistencies, check requirements, and surface potential issues before an expert reviews the submission.
The goal would not be to remove experts from the process. Instead, it could reduce the manual burden, speed up the first review, and let experts focus on the cases that actually require judgment.
So, if you work at a municipality, anywhere in the world, and are feeling this pain right now, feel free to reach out.
Kickstarting the process
There is still another bottleneck: starting the permit process itself. Someone still needs to go to the municipality website, fill out the initial forms, upload the required documents, and pay the permit fee.
In the future, this could change if municipalities adopt better digital infrastructure, APIs, or agent-friendly payment methods… But agent-to-agent payments and transactions is a topic for another article.
PS — If this use case, or a similar one, came to mind while reading this, feel free to get in touch. Applied is taking on client work.
Permits are the right shape of problem for an agent. The success criterion is binary, the permit was issued or it wasn’t. The failure mode is also well-known, status gets lost in someone’s queue. Most ‘AI agent’ demos pick creative tasks and then can’t say what done means. This one can.