Why Benetics AI Is Launching an AI Agent for Construction Before It’s Ready

At Benetics AI, a new AI agent for construction has just gone live. Not as a polished, fully autonomous system. Not as a finished product. But deliberately early – closer to an AI assistant for construction than a true agent.

“We wanted to get something safe into people’s hands early,” says Johan Tibell, CTO & Co-Founder at Benetics AI. “Because the only way to understand what an agent should really do is to see how people actually use it.”

‍

Together with Paul Siermann, Product Manager at Benetics AI, he is shaping what could become a new interaction layer for construction – one that goes beyond software as a tool and moves toward software that actively participates in getting work done.

From AI Assistant to AI Agent in Construction

The distinction between assistant and agent is not semantic – it is fundamental. What Benetics AI has launched today is an assistant: a system that can answer questions about projects, tasks, reports, documents, and plans within the platform. It helps users navigate and understand their data faster.

A true AI agent for construction, however, goes further. It acts. It proposes changes, prepares tasks, restructures workflows, and eventually executes actions. “An assistant answers questions. An agent does things on your behalf,” Johan explains. “That’s where we’re heading. But to get there, we need to start by understanding what actually brings value to our customers” This progression – from assistant to agent to workflow automation – reflects a broader shift in how construction software will be used.

Why AI Agents Matter for Construction

Construction is one of the largest industries globally – with lots of potential in digitizing its operational workflows. Despite years of software adoption, much of the daily work is still shaped by friction: administrative overhead, fragmented communication, missing information, and constant coordination between site and office. 

This is where construction AI becomes relevant – not as a feature, but as a structural change. “For a long time, software has been a tool people use to do work,” Johan says. “Now we’re at a point where software can actually do parts of the work itself.”

That shift is particularly powerful in construction, where a significant portion of time is spent on repetitive, documentation-heavy, or coordination-driven tasks. Benetics AI approaches this from a unique angle: starting with the data that actually reflects what is happening on site. “The data in Benetics AI is fundamentally different,” Paul explains. “It comes from frontline workers. Before, it lived in Messenger messages, emails, phone calls or on paper. Now it’s structured and accessible.” This creates something most generic AI tools don’t have: real operational context from the field.

Why Launch Early – Before It’s Ready

Releasing an early-stage AI assistant is a deliberate decision. “There’s only so much you can figure out in the lab,” Johan says. “At some point, you need real people in real projects.” This is particularly true in construction, where teams – especially those on site – operate under very different conditions. Benetics AI wants to learn from that reality early.

“The fastest way to build a reliable agent is to get people using it as early as possible.” Paul says. That approach comes with trade-offs. Some users will be impressed. Others will be underwhelmed. Some will encounter limitations or frustration. “We know people will try it and be disappointed in some cases,” Paul adds. “That’s a cost we’re consciously paying. At the same time, we’re managing the rollout very closely by setting clear expectations upfront and guiding interested customers through the first interactions. We aim to learn in a structured and responsible way.”

What the AI Assistant Already Does – and Where It Falls Short

In its current form, the Benetics AI assistant can access and interpret core project data, including tasks, reports, documents, plans, and project structures. This allows project managers to interact with their projects conversationally instead of manually navigating through layers of software. That alone creates tangible value. But this is an early version.

The assistant does not yet take action. It does not access all available data. And like any AI system at this stage, it can produce answers that sound plausible but are not useful in context.

Safe by Design – Even in an Early Stage

Launching early does not mean compromising on safety. The current assistant is deliberately constrained. It can only access data the person already has access to. It cannot modify, delete, or create anything. It cannot act autonomously. “The assistant cannot do anything destructive,” Johan explains. “It can only read your data – nothing more.” This creates a controlled environment for learning without introducing operational risk.

What Makes an AI Agent Work in Construction

A useful AI agent in construction is not just a chatbot with access to data. It requires deep context, access to the right information, and the ability to act in a way that aligns with real workflows.

‍

“At a high level, an agent is a model plus tools,” Johan says. “Tools to retrieve the right information, and tools to take action. Both are essential.” Over time, this includes integrating construction-specific knowledge such as technical standards, materials, and regulations. The goal is not just answering questions, but enabling reliable, context-aware decisions and actions.

Voice, UX, and the Reality of the Job Site

If AI is to work in construction, interaction design becomes critical. The current desktop chat interface is only a starting point. “Text input will only get us so far,” Johan says. “We believe voice, image and video understanding will be essential.” Construction work happens on site, not behind desks. Tools need to adapt to that reality. “The agent is useless if people don’t understand what it can do,” Paul adds. “it’s critical we make that visible at the right moment.”

How Benetics AI Measures Early Success

“I want to see a few people who really love it,” Johan says. “That’s usually the strongest signal that something is working.” Usage patterns, engagement, and repeated interaction matter more than broad adoption at this stage.

The Bigger Vision: From Software to Execution Layer

If this approach works, construction software evolves from a passive system of record to an active execution layer. Users move from clicking through tools to expressing intent: rescheduling tasks, summarizing projects, or generating structured work from unstructured input. “We know why people chose construction as their profession,” Paul says. “They want to build. The more administrative work we can take away, the better their work becomes.”

About Benetics AI

Benetics AI is a Zurich-based construction technology company focused on improving communication and workflows between office and construction site. By combining structured project data with AI capabilities, the company is building a new interaction layer for construction teams, with a strong focus on frontline workers.

About the Authors

Johan Tibell is CTO & Co-Founder of Benetics AI. Prior to co-founding the company, he spent over 15 years at Google as a Staff Software Engineer, working on large-scale systems and infrastructure.

Paul Siermann is Product Manager at Benetics AI. He previously held product and customer-focused roles at companies including Beekeeper, notime, and Angle Audio, focusing on translating operational people’s needs into practical product experiences.

‍