From CAD Assistants to Engineering Agents

Engineering development is not one clean workflow. It is a chain of connected workflows across requirements, CAD design, simulation, reviews, documentation, and manufacturing preparation. Today, a lot of this work is still manual, fragmented, and dependent on expert knowledge spread across different tools and systems. In this landscape, I looked at the new generation of AI engineering copilots and workflow automation startups trying to change that. 

Below are the main things I learned about how traditional engineering workflows work, what these new products bring, and why the category is becoming interesting from an investment and acquisition perspective.

How do engineering workflows traditionally work?

• At a basic level, engineering work is not one clean workflow, but a chain of connected workflows. A team starts from an idea or a requirement, then slowly turns it into something that can be designed, tested, documented, reviewed, and eventually manufactured. Each step has its own tools and its own specialists. This is why the engineering software stack already feels so dense before AI even enters the picture.

• Engineering development is made of several workflow categories. First, workflows around product definition, like turning customer needs into requirements or tracking how a requirement changes over time. Second, workflows around design: creating CAD models, modifying parts, or reusing past components. Third, workflows around validation: preparing a simulation, reviewing a design, or comparing test results with earlier assumptions. Fourth, workflows around industrialization: generating drawings, preparing manufacturing data, or documenting what needs to be produced. 

• A lot of engineering workflows is still about moving information between tools. A design may live in one CAD system, while the simulation is prepared somewhere else and the product data is managed in another system. Engineers often need to export files, clean them, update formats, copy information, and check that nothing was lost in the process. This is not always the most visible work, but it is a big part of the real workflow.

• Many engineering tasks are repetitive but still require expertise. This is one of the tricky parts of the category. Some tasks are clearly repetitive, but they are not simple. Preparing a model for simulation, checking manufacturability, creating a drawing, or building a toolpath still requires engineering judgment. So the opportunity is not only to automate basic tasks, but to automate parts of expert workflows.

• The result is that engineering teams spend a lot of time around the work, not only on the work itself. Engineers do not only design. They search, prepare, document, review, transfer, and coordinate. The bottleneck is not only inside CAD or simulation, it is in the myriad of workflows between all the tools and decisions.

• This is also why replacing the whole stack is unlikely in the short term. CAD, CAE, CAM, PDM, and PLM tools are already deeply embedded inside engineering organizations. They contain years of data and internal workflows. So the most realistic startup entry point is probably not to replace these systems, but to sit around them and make them easier to operate, search, and connect.

What new generation engineering copilots and workflow automation products bring?

• The general direction is that AI is becoming an execution and intelligence layer around the existing engineering stack. Most companies on this landscape are not trying to rebuild CAD or PLM from scratch. They are trying to make the current stack more “intelligent”. The AI layer sits around the tools engineers already use and helps them move faster through the workflow.

• The first product pattern is workflow orchestration. Synera, Bench, and Aibuild are good examples of this direction. They are not only building chat interfaces. They are trying to automate engineering workflows that touch several tools at once. The promise is closer to “let the system execute the engineering process” than “let the system answer a question.”

• Synera is probably the clearest example of the agentic workflow automation angle. Synera started more as an engineering automation platform and has clearly repositioned around AI agents. What I find interesting is that the agent does not replace the workflow. It operates inside it. For example, the OpenAI add-in brings text extraction and image interpretation into deterministic workflows, while the Cambrion add-in turns documents into structured workflow inputs.

• Bench shows a similar idea, but from the angle of generating and executing design workflows across existing tools. The Onshape integration is a good illustration of this. Bench can read and update design parameters, then modify part features through FeatureScript workflows. That is the difference between a copilot that comments on the work and a system that can actually act inside the engineering environment.

• Aibuild is interesting because it connects the engineering workflow to manufacturing execution. The company started from additive manufacturing software, but its product story is now broader. Aibuild OS is positioned as an autonomous engineering platform that can move from design intent to production ready workflows. This is a good example of how the category can extend beyond design and into manufacturing preparation.

• A second product pattern is the engineering copilot embedded inside CAD and product data workflows. Leo AI, CADAICO, NEOCAD, and Hanomi all approach this from different angles. Leo AI is building a mechanical engineering copilot around CAD and PDM/PLM data. CADAICO has a chat-to-design product and a PDM assistant. NEOCAD focuses on searching and reusing CAD archives. Hanomi starts from the very concrete pain of turning CAD models into 2D drawings.

• One big use case is making engineering knowledge searchable and reusable. Companies already have a lot of engineering knowledge, but it is often trapped in old files and past projects. NEOCAD is a good example here. Its product is built around finding similar CAD files, reusing past work, and helping engineers avoid starting from scratch every time.

• Another use case is capturing design intent and engineering memory. Tandem seems less focused on generating geometry and more focused on remembering the engineering context. The product captures design sessions, requirements, reviews, and decisions inside the CAD workflow.

• Several companies started with narrow, painful workflows before expanding upstream. Hanomi is a good example. The starting point is very specific: generating 2D drawings from 3D CAD models. It may sound less spectacular than a fully autonomous AI engineer, but it is a painful and repeated workflow that can later expand toward a broader mechanical engineering intelligence layer.

• Another product shift is from passive copilots to active agents. The first version of many AI products is help and guidance. The next version is execution. This is what appears across the landscape: Synera launches engineering agents, Aibuild launches Aibuild OS, Leo AI talks about agents for PLM workflows, and Bench builds workflows that can act inside CAD tools. The direction is clear: the AI layer is moving from advice to action.

Investment & Acquisition environment

• The funding environment looks active, but less mature than AI native simulation. Compared with the simulation landscape, this category feels earlier. There is one more advanced company, Synera, but many of the others are still seed or pre-seed. Maybe buyers are still learning what they actually want from an AI engineering copilot.

• Synera is the most advanced funding signal in the landscape. Synera raised a €35M Series B in 2026 after earlier seed and Series A rounds.

• The rest of the landscape is still mostly early stage. Bench raised a £1M pre-seed. Leo AI raised a $5M seed. Hanomi raised a $3M seed. NEOCAD raised a €550k pre-seed. Aibuild is older and has already raised a Series A. So the market still feels like a category in formation or I have missed many startups (it’s possible).

• The investor mix already shows that this is not just a classic “SaaS like” category. You see classic software investors, but also industrial and strategic names such as BMW i Ventures or Nikon.

• Unsurprisingly, the founding teams are very “domain heavy”. This was one of the clearest patterns in the landscape. The founders often come from mechanical engineering, aerospace, robotics, additive manufacturing, CAD automation, or AI research.

• There are no acquisition events in this landscape yet, but I would expect acquisitions later. Traditional engineering software companies are used to buying important workflow capabilities. But I’m not sure of the real potential of AI copilot startups compared to AI simulation ones.