Using AI Agents to Automate Repetitive Tasks in Construction Software

A different approach to construction software automation

A lot of work inside construction software is still repetitive.

Teams spend hours on tasks that are not complicated, just tedious: entering meeting notes, updating status fields, copying data between systems, filling out forms that should have been pre-populated. These are the kinds of workflows that drain time without adding much value.

Most automation solutions assume you have API access, integration budgets, or engineering support. In practice, many construction tools do not expose clean APIs, and many teams do not have the resources to build custom connectors.

This is where browser-based AI agents become interesting. Instead of trying to integrate at the API level, an agent can operate the interface directly, navigating the software the same way a person would. This opens up automation possibilities for tools that were never really built to connect cleanly.

Why we tested this

We wanted to see if an AI agent could complete a real construction workflow by using the website directly, not through an API.

For the test, we used a meeting transcript and had the agent enter it into Autodesk Forma (formerly Autodesk Construction Cloud). The agent needed to:

• Navigate to the correct project and module.
• Fill out the meeting form fields.
• Create the meeting content from the transcript.
• Move through the workflow step by step.

The goal was not to build a polished integration. It was to understand whether an agent could handle a practical, repetitive task in a natural way.

What we used

The setup was simpler than expected.

We used agent-browser, a tool that lets you direct an AI agent through browser tasks with more control than closed alternatives like ChatGPT Atlas or Claude for Chrome. The key difference is the level of flexibility it gives you over the workflow:

• Choose your AI model. You are not locked into a single provider. You can use the model that fits your needs, budget, or latency requirements.
• Restrict domains. You can define exactly which sites the agent is allowed to access. This matters in construction, where projects involve sensitive data and strict access controls.
• Save snapshots. The agent can capture evidence of its work along the way, which is useful for audit trails and verification.
• Keep clear records. You get a transparent log of what the agent did, when, and why.

We also used OpenCode, which made the setup easier through its Skills library. Installing the agent-browser skill gave the agent a much better understanding of how to navigate and interact with web interfaces, reducing the time needed to get the workflow running.

How the test worked

The agent was given a meeting transcript and instructed to create a meeting entry in Autodesk Forma. This is the kind of task that happens regularly on projects: someone takes notes, and those notes need to end up in the system of record.

Normally, this means manual data entry. Someone has to:

1. Open the project in Forma.
2. Navigate to the meetings module.
3. Create a new meeting.
4. Copy or type the details from the transcript into the right fields.
5. Save and confirm.

The agent handled this sequence directly. It opened the browser, navigated the interface, filled the fields, and completed the workflow. It took its time, but it worked.

The fact that it worked at all is what makes this approach worth considering. For tedious tasks that do not need complex judgment, an agent that can operate the UI directly is often good enough, and much faster than doing it by hand.

Quick walkthrough

Here is a short walkthrough showing the AI agent in action as it navigates Autodesk Forma, fills out the meeting form from a transcript, and completes the workflow step by step.

Why this matters for construction

Construction software has a specific set of constraints that make traditional automation hard:

• Limited API coverage. Many tools, especially older or specialized ones, do not expose APIs for the workflows teams actually use.
• Fragmented stacks. A typical project uses multiple platforms that were never designed to talk to each other.
• Changing interfaces. UI updates happen frequently, and custom integrations often break when they do.
• Security and access control. Teams need to control exactly what data an automation can touch and where it can go.

Browser-based agents sidestep some of these problems. They do not need APIs. They work with the interface as it exists today. And they can be restricted to specific domains and actions, which helps with security concerns.

This does not replace proper integrations where they are feasible. But it opens the door to automating workflows that would otherwise stay manual because the integration cost is too high.

Where this approach fits

Browser automation with AI agents is not the right tool for every job. It is best suited for:

• Repetitive data entry where the steps are clear and the judgment required is minimal.
• Tasks across tools without APIs where the alternative is manual copy-paste work.
• Ad-hoc workflows that change often and would be expensive to maintain as formal integrations.
• Quick experiments where you want to test whether automation is viable before investing in a deeper solution.

Some examples where this could be useful in a construction context:

• Entering meeting notes or site observations into project management platforms.
• Copying asset data between systems that do not sync.
• Updating status fields or checklists across multiple tools.
• Extracting and formatting data from web-based reports.
• Filling out repetitive forms for compliance or documentation.

The common thread is that these are tasks a person could do, but should not have to do dozens of times per week.

What this means for teams

For teams managing construction technology, this approach offers a few practical advantages:

• Lower barrier to entry. You do not need engineering resources or vendor API access to automate a simple workflow.
• Faster iteration. You can test an automation idea in hours rather than weeks.
• Better evidence trails. Snapshots and logs make it easier to verify what the agent did and debug when something goes wrong.
• More flexibility. You are not dependent on vendors building the integrations you need.

The tradeoff is that browser automation is generally slower and more brittle than a proper API integration. It is not a replacement for well-built connectors where they exist. But it is a useful layer for the gaps between systems that APIs do not reach.

A shift in how we think about integration

One of the more interesting side effects of this experiment is how it changes your perspective on software without APIs.

In the past, a tool that did not expose an API was often a dead end for automation. You either accepted the manual work or tried to work around the limitation with brittle scripts or data exports.

Agents that can operate the interface directly change that calculation. Suddenly, any web-based tool becomes automatable to some degree. The question shifts from “does this have an API?” to “is this workflow predictable enough for an agent to handle?”

In construction, where the software landscape is fragmented and API coverage is patchy, this is a meaningful shift. It gives teams more options for reducing repetitive work without waiting for vendors to build the connections they need.

References and tools

If you want to explore this approach yourself:

• agent-browser: https://agent-browser.dev/
• OpenCode: https://opencode.ai/
• Skills library: https://skills.sh/

These tools are evolving quickly, and the capabilities will likely expand. For now, they are already useful for the kinds of tedious, repetitive tasks that still consume too much time in construction software workflows.

Final thought

Construction software has made huge strides in functionality, but many of the workflows inside these tools are still slower and more manual than they need to be.

Browser-based AI agents are not a magic solution, but they offer a practical way to automate repetitive tasks without waiting for APIs or integration budgets. For teams willing to experiment, they can reduce friction in workflows that have been stuck in manual mode for too long.

If you want to explore other ways to streamline your BIM and construction data workflows, see our guide to connecting BIM data to Excel with Frame and our post on creating pivot tables with Frame and a 3D viewer. If you want to discuss how Frame can help with your data automation needs, contact us.