Trunk Tools Ditched General-Purpose AI and Cut Document Review from 60 Days to 10

Quick Facts

• Trunk Tools reduced construction submittal cycles from 50-60 days to 10 days using a purpose-built three-layer AI architecture.
• The median TrunkSubmittal user saw a 74% reduction in cycle time over 12 months, with “stuck submittals” dropping from 42% to 2%.
• Frontier general-purpose models top out around 26% accuracy on basic construction tasks like detecting a door; Trunk Tools only ships agents that achieve 95% accuracy.

Construction AI startup Trunk Tools says it cut document review times from 60 days to 10 by abandoning general-purpose AI models and building a specialized three-layer architecture from scratch. The company’s results challenge the assumption that frontier models can handle industry-specific workflows out of the box.

CEO Dr. Sarah Buchner founded the company after a career that started as a carpenter in Austria at age 12 and led through roles as superintendent, project manager, and group leader before she earned a PhD in civil engineering and data science. Her diagnosis of the industry’s core problem is direct: general-purpose AI falls short because construction runs on proprietary data, implicit workflows, and symbolic documentation that trained human eyes learn to read over years of practice.

“Any PE who has fed a drawing set into a general-purpose AI model knows how short it falls,” Buchner said. Research she cited shows frontier models achieve roughly 26% accuracy on something as basic as detecting a door in construction drawings.

The Architecture

Trunk Tools built its platform around three layers: perception, semantics, and agents. The perception layer teaches the system to read construction-specific visual language. A door in a construction drawing is often just an arc on a wall, not a labeled object. The semantic layer connects that recognized element to the drawing that details it, the specification that governs it, and the trade that installs it.

The agent layer sits on top of that structured knowledge graph. Trunk currently runs seven purpose-built AI agents that handle tasks like RFI analysis, bid reviews, and submittal processing. The agents now communicate directly with each other. An architectural drawing review agent can flag problems and autonomously hand off to an RFI agent, which then reaches out for clarification.

“We really set out to take the data from dispersed systems, pre-process it, structure it, go through our ontology into a knowledge graph, and then train AI models,” Buchner said.

The Scale of the Problem

The numbers behind construction document volume make the case for purpose-built tooling. Buchner estimates the average high-rise generates about 3.6 million pages of documentation. One early customer building a Manhattan high-rise had 120,000 files averaging 30 pages each synced into the platform, totaling 3.5 million pages for a single building.

The $13 trillion construction industry accounts for roughly 5% of U.S. GDP and remains one of the least digitized sectors in the economy. Buchner describes most construction data environments as “ugly documents, proprietary schemas, implicit workflows, and long-running tasks” that general-purpose models are not built to handle.

Documented Customer Results

Cleveland Construction used Trunk Tools alongside Autodesk Forma to reduce submittal review times from days to hours, saving more than 790 hours and $60,000 across four projects. Torcon, a construction manager, completed 64 submittal reviews over four months using the platform and saved nearly 47 hours.

Customers have reported a 40x return on investment when factoring in labor savings and rework prevention. One customer noted the platform reduced internal labor hours by 50% on projects, equivalent to 0.5% of total project cost.

The TrunkReview product cuts bulletin review time from nearly six hours to under five minutes. Field teams report saving 20 to 40 minutes per question answered through the platform.

What It Means for Software Buyers

Trunk Tools’ results make a concrete argument against deploying horizontal AI tools in specialized, document-heavy industries. The 74% cycle time reduction and the drop in stuck submittals from 42% to 2% are the kind of metrics that justify vertical AI investment over general-purpose alternatives.

On June 17, 2026, the company launched Cortex, a new AI intelligence layer it describes as the purpose-built brain of construction, developed over four years of domain-specific training. The release signals that Trunk Tools is building infrastructure, not just applications, for the sector.

Read more: Trunk Tools’ stack cut document review from 60 days to 10 by ditching general-purpose models