A Team of AI Agents Reviews Your Construction Drawings - Here's What They Catch

A senior engineer opens a 300-sheet drawing set at 4 PM. Submission is tomorrow. She needs to verify grid alignment, confirm every detail callout leads somewhere, check seismic parameters in the structural notes, and make sure the architectural door schedule matches the floor plans.

By 11 PM she's found a mislabeled grid line on S-42, a detail bubble on A-15 pointing to a detail deleted two revisions ago, and a concrete spec that still says ACI 318-14. She hasn't started cross-discipline coordination.

Not the engineering - the checking. Drawing review is exhaustive, repetitive, and critical. Missed errors become RFIs, change orders, rework. Yet review quality depends on how much attention a human has left at the end of a long day.

What If a Team of Specialists Reviewed Every Sheet Simultaneously?

That's what Stru AI's Review Agent does. It's an autonomous multi-agent system - a team of specialized AI agents auditing your drawing set in parallel. Structural detailing agents. Code compliance agents. Cross-reference checkers. MEP coordination agents. Architectural reviewers. Geometry validators. Each one focused on a specific domain, running simultaneously across every sheet.

Upload your PDFs, get a structured report in under 30 minutes.

They don't get tired. They don't skip sheets. They don't lose focus on sheet 247 because they've been staring at drawings since lunch. Every reference, every label, every dimension - checked across every page, every time.

What the System Actually Catches

This isn't a simple overlay comparison or a PDF diff tool. The Review Agent reads your drawings the way an engineer would - understanding references between sheets, interpreting structural details, checking code compliance, and flagging coordination conflicts across disciplines.

Here's what that looks like across each domain.

Structural Detailing

The structural agents check your details against the codes that govern them.

Steel - Weld symbols get verified for completeness: type, size, length, field/shop designation. Bolt specifications checked against connection demands. Edge distances measured against AISC 360 minimums. Connection details compared against the member forces they need to resist.

Concrete - Rebar cover checked against ACI 318 requirements for the exposure condition. Lap splice lengths verified. Development lengths at hooks and terminations. Seismic detailing requirements for special moment frames, shear walls, and coupling beams - confinement zones, hoop spacing, crossties.

Masonry - Grouting patterns checked against TMS 402. Bond beam locations at required intervals. Control joint spacing. Lintel bearing lengths. Wall anchorage details at diaphragm connections.

Wood - Hanger selections verified against loads. Sheathing nailing patterns checked - edge nailing, field nailing, blocked vs unblocked diaphragms. Hold-down anchorage. Bearing lengths at supports. Shear wall aspect ratios against NDS limits.

These aren't surface-level checks. The system understands that a W21x44 beam framing into a W14x90 column needs a connection detail that works for both members, and flags when something doesn't add up.

Code Compliance & Design Criteria

A separate set of agents focuses on whether your drawings reflect current codes and correct parameters.

Code editions - Is your general notes sheet referencing IBC 2021 or still showing IBC 2018? Does the seismic design reference ASCE 7-22 or an older edition? The system checks every code citation across every sheet, not just the cover page.

Seismic parameters - SDS, SD1, seismic design category, response modification coefficient, importance factor. These values need to be consistent between your general notes, your analysis assumptions, and your detail requirements. The agents cross-check all of them.

Wind parameters - Basic wind speed, exposure category, internal pressure coefficient, components and cladding pressures. Same cross-check logic - values stated on the drawings should match what the design requires.

Load requirements - Live loads on floor plans matching ASCE 7 Table 4.3-1 for the stated occupancy. Roof live loads, snow loads, rain loads. Special loads at mechanical equipment, storage areas, assembly spaces.

When a code parameter is wrong, the downstream effects cascade through the entire design. Catching it on paper costs nothing. Catching it during construction costs everything.

Cross-Reference & Dependency Integrity

This is where the system does something humans genuinely struggle with at scale. It builds a complete dependency map of your drawing set.

Every detail bubble on a plan references a detail on another sheet. Every section marker points to a section cut. Every schedule references tags on the plans. Every sheet in the index should exist in the set.

The agents trace every one of these connections and flag:

• Broken references - A detail callout on S-12 points to detail 3/S-45, but S-45 doesn't have a detail 3. Maybe it was renumbered. Maybe S-45 was deleted in the last revision. Either way, you have a callout leading nowhere.

• Orphan details - Detail 7/S-23 exists, but nothing in the drawing set references it. It's either leftover from a previous design or it's a detail that should be called out somewhere but isn't.

• Missing sheets - The sheet index lists S-501 through S-510, but S-507 is missing from the package. Or a general note references "see sheet M-201 for equipment layout" but M-201 isn't in the set.

• Schedule mismatches - A door tag on the floor plan reads D-105, but the door schedule doesn't have a D-105 entry. Or the schedule says D-105 is a 3'-0" x 7'-0" hollow metal door, but the plan shows it in a 4'-0" opening.

On a 100-sheet set, a thorough cross-reference check might take an engineer a full day. On a 300-sheet set, it's practically impossible to do comprehensively by hand. The system does it in minutes.

MEP Coordination

Cross-discipline conflicts are the most expensive errors to fix in the field. The MEP coordination agents compare mechanical, electrical, and plumbing drawings against the structural and architectural sets.

Equipment coordination - Equipment tags on mechanical plans match the equipment schedule. Panel schedules account for all connected loads. CFM totals at air handling units match the sum of downstream diffusers.

Spatial conflicts - Ductwork routing through structural members. Pipe penetrations at locations that compromise structural capacity. Equipment placed where structural framing doesn't support the loads.

Fire protection - Fire dampers at rated wall and floor penetrations. Fire-rated assemblies matching the ratings shown on architectural plans. Firestopping details where services penetrate rated assemblies.

Code references - Electrical code references current NEC edition. Plumbing code references correct jurisdiction. Mechanical code citations consistent across the package.

Architectural Review

The architectural agents check coordination with structure and internal consistency within the architectural package.

Schedule integrity - Door schedules match plan tags. Window schedules match elevations. Finish schedules match room tags. Hardware groups assigned and complete.

Life safety - Egress path continuity. Travel distances within limits. Exit widths matching occupant loads. Exit signage shown. Area of rescue assistance where required.

Accessibility - ADA-compliant door clearances. Accessible route continuity. Restroom fixture counts and clearances. Ramp slopes and landing dimensions.

Fire ratings - Rated walls continuous from floor to deck above. Rated openings with correct frames and hardware. Rating labels consistent between plans, sections, and details.

Grid coordination - Architectural plans aligned with structural grid. Column locations matching between disciplines. Floor-to-floor heights consistent.

Geometry, Dimensions & Drawing Quality

The system checks the internal consistency of the drawings themselves.

Dimension verification - Do the individual dimensions add up to the overall? Does the column grid spacing match between plans and sections? Are level datums consistent across all elevation views?

Annotation quality - Leaders that don't terminate at what they're labeling. Text overlapping other text or linework. Callout bubbles without content. View titles that don't match what's shown.

Scale consistency - Details at different scales that show the same element at different sizes. Scale bars that don't match the stated scale.

Consistency across views - A beam shown as W21x44 on the framing plan but W21x50 in the section. A slab thickness of 8" on the plan but 7" in the detail. Rebar shown as #5 in one view and #4 in another.

These are the errors that make a drawing set look unfinished - and the ones that generate clarification RFIs during construction.

Three Levels of Issues

The Review Agent categorizes findings into three tiers, from documentation cleanup to engineering-critical flags.

Level 1 - Drafting & Documentation Errors. Mislabels, broken leaders, dimension string errors, mismatched callouts, missing references, placeholder text left in, revision inconsistencies. High volume, low individual severity, but collectively they erode confidence in the package and generate dozens of small RFIs.

Level 2 - Design & Requirement Gaps. Missing material specifications, clearance violations, minimum area conflicts, code requirements not reflected in the drawings, schedule entries that don't match what's drawn. These are the issues that stall permitting or generate change orders during construction.

Level 3 - Engineering Validation Flags. The system identifies inconsistencies that suggest a deeper engineering issue - a load path discontinuity, a connection detail that doesn't match the member demand, a seismic detailing requirement that isn't met. On the Enterprise Deep Check, it goes further: autonomous calculations, verification against uploaded calc sheets, and value engineering suggestions.

How It Works in Practice

Step 1 - Upload your drawing set as PDFs. The system ingests every sheet and builds a dependency graph - mapping detail bubbles, section markers, view references, schedules, and sheet index entries across the entire package.

Step 2 - Agents deploy in parallel. Each agent audits its domain across the full set. Cross-discipline agents compare Architecture against C&S against M&E simultaneously. The system evaluates the full submitted package every run - not just changed sheets - to catch cascading reference impacts.

Step 3 - Structured report delivered. Issues organized by discipline, type, severity, and sheet location - each with evidence and context. Output as PowerPoint, Excel, Word, or PDF. Provide a sample template and the report matches your firm's format.

Step 4 - Bring your own checklists. Have authority requirements (UBBL, DBKL, MS1184)? Client-specific QC rubrics? Internal standards? Upload them. The agents audit against your rules, not just generic best practices. Multiple guidelines can run in a single prompt - the system explores them in parallel.

Plans Built for Real Project Workflows

Credit-based usage - no per-project charges, no seat limits. Run checks on multiple projects concurrently within your credits.

For large projects (1,000+ sheets), split into logical packages on Pro/Max - Tower A Architecture as one run, Tower B Structure as another. On Enterprise, upload everything in a single click for a comprehensive Deep Check.

Why This Matters

Every structural or architectural firm has experienced this: an RFI lands on your desk three months into construction. The contractor found a conflict between the structural framing plan and the mechanical ductwork routing. Or a detail callout that leads nowhere. Or a beam size that's different on two sheets.

The fix takes an hour. The cost - remobilization, schedule delay, change order markup - takes weeks and thousands of dollars.

Manual review catches what the reviewer has time and energy to find. An automated review catches everything it's designed to find - every time, on every sheet, at 2 AM or 2 PM.

The Review Agent doesn't replace engineering judgment. It handles the systematic, exhaustive checking that humans are bad at sustaining - cross-reference tracing across 300 sheets, dimension verification across every view, code parameter consistency across every note. So your engineers spend their time on the issues that actually need their expertise.

Fewer RFIs. Fewer change orders. Cleaner submissions. Faster approvals.

Ready to see what the agents find in your drawings? Start a review at stru.ai