Spend enough time at a trade show and you can tell when an industry has crossed from curious to ready. At IMTS 2026, the engineers who stopped at Cosmon’s booth to talk weren't asking theoretical questions about AI. They had specific problems already in mind, and they wanted to know if an agent could handle the prep work around those problems without them babysitting every step.

TLDR:

• At IMTS 2026, 91,000 attendees gathered as the event featured a dedicated industrial AI conference program for the first time

• Engineers brought day-to-day CAD and PLM bottlenecks: generating production drawings, verifying GD&T callouts against DFM standards, and reconciling CAD-to-PLM revision drift

• The agent rebuilds dead STEP geometry into parametric feature trees, runs GD&T and DFM checks, and prepares simulation setups, stopping for your review before anything runs

• Engineering files, specifications, and change orders stay inside your own infrastructure, with zero data retention and on-premises deployment available

• Cosmon's AI agent coordinates CAD, simulation, and PLM hand-offs as a single logged, reversible sequence inside your existing tools

The IMTS 2026 Scene

Chicago in September. McCormick Place packed with 91,000 engineers, manufacturing executives and technology enthusiasts moving between booths at a pace that told you this wasn't a browsing year. The floor energy was different from 2024: less wide-eyed, more purposeful. The shift was hard to miss. Where previous shows centered on deep automation, this year's emphasis had moved to autonomy, scaling systems to run as part of bigger, AI-driven ecosystems.

IMTS 2026 introduced the Industrial AI Arena. Its presence alone said something about where the industry thinks we are. Keynote remarks from major tech leaders reinforced the point: the speculative pilot phase is winding down, and teams are being asked to show results. The phrase "proof of concept" came up at multiple booths, almost always as something the speaker wanted to move past.

For mechanical engineering specifically, that shift changed the texture of conversations on the show floor. Nobody was asking "should we be using AI?" The questions had moved forward. The working assumption, at least among the engineers who stopped to talk, was that something useful was now possible. The real question was whether what they'd seen in demos would actually hold up against their files, their tools, and their workflows.

What Engineers Actually Asked at the Cosmon Booth

The questions we heard were grounded in daily design-office and analysis realities. Mechanical leads, structural analysts, and CAD managers stopped by the Cosmon booth not to ask high-level questions about the future of AI, but to test whether an agent could handle the friction points stalling their active product development efforts. They arrived with day-to-day workstation bottlenecks in mind, eager to see how automation works directly inside their existing software instead of in an isolated demo sandbox.

The Problems Engineers Brought to the Booth

A CAD lead from an automotive supplier asked whether an agent could rebuild a dead supplier STEP file into a fully parametric model with a complete feature tree, so his team could make downstream design edits without starting from scratch. An aerospace engineering team asked about catching revision drift between CAD models and PLM release records across multi-site programs. Another design engineer described spending entire afternoons checking GD&T callouts and running manual DFM reviews against drawing standards before releasing part families.

None of these were edge cases. Every engineer who stopped had a version of the same story: the actual engineering judgment took an hour, the software management and cross-tool workflows around it took the rest of the day.

"I know what the model needs. I just need something that can actually set it up without me babysitting every step."

That line came up more than once. Engineers weren't debating AI adoption in principle. They were asking whether it could handle the unglamorous, repetitive setup work that sits between a design decision and handoff to production.

The Moment the Conversation Changed

During a live booth demo with an imperfect supplier CAD model, Argus repaired the geometry, generated the mesh, and configured boundary conditions and surface contacts.

Then it stopped. Before running the solver, Argus assembled a complete model summary covering mesh statistics, material assignments, and boundary logic in plain language. Nothing ran without engineer approval.

Running directly inside tools like Abaqus and Ansys, every action stayed logged and fully reversible. Offloading the setup while keeping engineering judgment in place highlighted the difference between what a copilot suggests versus what an agent executes.

Presenting at the Google Cloud Booth

Cosmon partnered with Google Cloud at IMTS 2026 to show how agent workflows scale across enterprise engineering environments. Instead of isolating automation to single desktop sessions, teams can tie CAD routines, solver runs, and PLM hand-offs directly into their broader development pipelines. In multi-site programs where CAD geometry and PLM release records frequently drift apart, the agent reconciles revision states, performs bulk PDM data card updates as studies finish, and runs release checks in parallel across distributed teams without manual data entry.

At the same time, pairing Cosmon with Google Cloud infrastructure gives engineering managers and IT teams confidence around data security and model protection. Teams evaluating cloud adoption can find practical guidance on adopting AI without risking engineering IP. Proprietary geometry, simulation setups, and release records remain fully protected, meeting SOC 2, GDPR, and CCPA requirements under zero data retention terms. It resolves the classic split between engineering capability and enterprise security, letting teams automate heavy CAD and CAE workflows while retaining complete control over their files.

Cross-Tool Orchestration: the Glue Between Tools

The engineers who stopped by ran mature toolchains, and almost none of them wanted to replace what they had. SolidWorks, Ansys, Abaqus, COMSOL, NX, CATIA, a PLM system of record: all chosen for good reasons, all staying, with decades of trust built into using these tools. What they wanted was something that could work across those tools without migrating files or adopting one more ecosystem. For many CAD leads, that meant examining dedicated AI tools for SolidWorks and CAE solvers that run directly in the native environment.

That framing matters, because the hardest problem they described wasn't inside any single tool. It was in the hand-offs between them.

One engineer walked us through a typical cascade: a geometry change in CAD forces a remesh, which forces a boundary-condition recheck, which forces a re-run of the study, which forces a PLM record update. Five steps, four tools, a manual hand-off at every seam. His point wasn't that any one step was hard; it was that he was the glue holding the sequence together, carrying context from one session to the next by hand. Miss a seam and you ship a stale result.

That's the problem Argus is built for. It runs inside the tools the team already owns and treats the hand-offs as part of the task, not the engineer's problem to bridge: a design change, the simulation setup it triggers, and the release record it touches connect as one continuous, logged, reversible chain, with outputs landing back in the same tools and nothing migrated.

Tool by tool, that's the setup work engineers came to the booth hoping to offload:

The reaction on the floor kept coming back to the same relief: not "it can do my CAD," but "it can carry the work across my CAD, my solver, and my PLM without me babysitting the seams."

FAQ

What was the main takeaway regarding industrial AI at IMTS 2026?

IMTS 2026 marked a clear transition from speculative pilots to practical workstation execution. With the debut of the Industrial AI Arena and packed conference sessions, manufacturing and engineering leaders signaled that text assistants and isolated demos are no longer sufficient. Attendees focused on whether automation could execute multi-step CAD, CAE, and PLM routines directly inside their existing software stacks while leaving final judgment calls with the engineer. For an analysis of how this execution model compares to conversational tools, see our breakdown of AI copilots versus AI agents.

How did teams at the booth evaluate data security and IP protection?

Security was the first question for most defense, aerospace, and automotive leads. Your CAD files and simulation inputs stay inside your own infrastructure. Argus runs natively in your tools, with no vendor ingestion of proprietary geometry and no training on your data. On-premises deployment removes any external data path entirely, and cloud processing runs under a zero-data-retention contract with the option to disable telemetry (SOC 2, GDPR, and CCPA).

Why did engineers contrast Argus with a general AI agent or chat tool?

General AI tools can describe the steps; they can't execute them inside SolidWorks, Abaqus, or your PLM system. Argus runs the full hand-off sequence (geometry cleanup, mesh setup, boundary conditions, and PLM version checks) as a single logged, reversible task inside the tools your team already owns.

How does the sign-off checkpoint work?

Before executing, Argus assembles a complete model summary (mesh statistics, material assignments, boundary-condition logic, and step sequencing) in plain language. Nothing runs until you review it and give the go-ahead. The agent proposes; the call is yours.

Which tool versions were attendees asking about?

Teams running mixed toolchains asked about compatibility across their stacks. On the CAD side, Argus supports SolidWorks 2021 and later, Siemens NX, and CATIA V5; for FEA, the Abaqus integration runs on Abaqus 2024 and later (Python 3.10+).

Final Thoughts: What IMTS 2026 Signals for the Road to 2028

Every IMTS is a snapshot of where manufacturing stands, setting the pace for the two-year cycle until the industry reconvenes at McCormick Place. The 2026 show made one thing clear: engineering teams have moved past speculative discussions about whether AI belongs in design offices. The Industrial AI Arena anchored the floor because teams wanted to see practical workstation execution on active programs, treating isolated pilots as something to leave behind.

Between now and IMTS 2028, the defining challenge will not be raw software capability, but trust and integration across existing engineering stacks. For years, automation has been framed as a trade between speed and control: accelerate throughput, but sacrifice oversight. What engineers responded to on the floor was automation that shoulders the tedious setup while keeping the engineer as the final decision-maker. Over the next two years, the standard will hinge on whether tools run natively where files live, log every step, and return the hard calls to the user.

When the industry gathers again in Chicago in 2028, the tools that remain in production will not be the ones that gave the flashiest stage demos. They will be the agents that quietly handled the CAD, CAE, and PLM bottlenecks on live production programs without leaking proprietary geometry or cutting the engineer out of the loop. The appetite for practical execution is here. Run Argus against your own geometry and see what accelerated throughput actually looks like in your tools.