AI Tools for Infrastructure Project Managers
A working map of AI tools across the infrastructure industry, organized by function and updated quarterly.
The AI Tools Map
for Project Delivery
Infrastructure Catalyst tracks the AI tools that touch civil and infrastructure project delivery right now. This page is a working map of about forty products, grouped by what they do, not by what their vendors call them. Click any card to see pricing, data residency, vendor lock-in, and a short note on what the tool is good at.
How this map is made
The map is built around four functional types: See (computer vision on plans, sites, and drawings), Predict (machine learning for risk, schedule, and cost), Write (generative AI for specs, RFIs, contracts, code, and visuals), and Act (agents that take action on systems). Every tool sits in one type. Sub-categories inside each type group tools by the job they actually do.
Included: products that ship today, that publish enough information to verify capability, and that are plausibly procurable by a US infrastructure firm or public agency. Excluded: pre-launch products, tools without a recent release, and anything that conflicts with US public-agency data residency by default.
Permissions vary by state and contract. Read the Transportation AI Policy Tracker before adopting any tool with non-public agency data. Refreshed quarterly. Next refresh: August 2026.
TYPE.01
See
Computer vision applied to plans, photos, drone imagery, and field walkthroughs. Tells a project manager what is on site, what is on the drawings, and where the two diverge.
Aerial progress capture and safety scans.
360-degree site walkthroughs mapped to floor plans.
Element-level installation tracking from helmet-mounted vision.
Laser scanning and 360-degree progress on one platform.
Capture-agnostic reality intelligence across cameras, drones, LiDAR.
Site video and LiDAR compared to BIM and schedule.
Computer-vision takeoff from architectural and MEP plans.
Service-augmented automated takeoff across trades.
Cloud-native takeoff with trade-specific AI.
Auto takeoff, count, and scale on PDFs.
Vision takeoff on Google Cloud.
Claude-powered plan and spec review.
Jobsite hazard detection from photo and video.
Geospatial foundation models and AI Assistants.
TYPE.02
Predict
Machine learning trained on historical project data to anticipate schedule slips, cost overruns, RFI risk, and productivity bottlenecks.
Schedule risk forecasting from historical schedules.
Generative scheduling across time, cost, and resources.
Design, RFI, and quality risk prediction.
Workforce analytics and productivity prediction.
TYPE.03
Write
Generative AI for specs, RFIs, contracts, reports, emails, code, presentations, and visuals. The right tool depends on the document, not the brand.
Long contracts, multi-clause spec extraction.
Cheapest 1M-context option from a Tier 1 lab.
2M-token context, multimodal plan and photo review.
Agentic workflows, broad task coverage.
High-accuracy reasoning when wrong answers are expensive.
Multi-agent reasoning with live web search.
Lowest-cost reasoning. Procurement caveat for US public work.
Cited live web research. Defensible for PM memos.
AI browser with agent mode for web automation.
In-tenant chat across Outlook, Word, Excel, Teams, SharePoint.
Workspace Q&A and meeting notes inside Notion.
Field-side phone AI. Siri backed by Gemini.
Contract risk extraction (acquired by Trimble Q2 2026).
RFI workflow automation and submittal compliance.
Subcontractor scope extraction across project documents.
RFI drafting from document repositories.
Code compliance using licensed ICC and municipal codes.
Contract review inside Microsoft Word.
Source-grounded Q&A across a project corpus.
Enterprise search across SharePoint, Drive, Slack, Jira.
Lightweight retrieval over uploaded files in chat.
Live transcription and action-item extraction.
Semantic search across past meetings.
Local-first augmented notes. No meeting bot.
On-device transcription. Trailer-grade noise cancellation.
In-tenant Teams recap and Copilot summaries.
Zoom-native AI summary, bundled in Workplace.
Multi-file AI code editor.
Terminal-native agentic coding.
IDE completion and chat. Easy IT sell.
IP-indemnified commercial visuals.
Photoreal conceptual renderings.
Image editing and composition. In-image text rendering.
Photoreal API per dollar.
Vector outputs for logos and infographics.
TYPE.04
Act
Agents that operate browsers, drive desktops, route documents, and execute multi-step workflows. The category is the youngest. Most products are 12 to 18 months from boring-PM-reliable.
Web research and form-filling agent.
Packaged workflows for Office apps. Beta.
Self-hostable browser agent.
Autonomous software engineering tasks.
RFIs, submittals, daily logs against Procore data.
RFI and submittal-log drafting inside ACC / Forma.
Project search and smart email filing. Launched May 2026.
M365-native workflow automation. IT-approved.
7,000+ app integrations. Lowest learning curve.
Visual scenario builder. 3x cheaper than Zapier.
Self-hosted automation. Native LLM nodes.
NOTES.05
Tips
Working principles for matching the tool to the job.
- Match the tool to the document, not the brand. A 300-page Caltrans manual needs a long-context model. A two-page RFI needs a cheap one.
- Long context is real but not magical. Independent research (Chroma, "Context Rot") shows 10 to 25 percent accuracy degradation for facts buried inside a 200K-plus context. Chunk the document for mission-critical extractions.
- Build a two-tool habit. One general-purpose chat (Claude or ChatGPT) plus one citation-grounded research tool (Perplexity or NotebookLM) covers most working PMs.
- Default BUY for data-moat categories. Reality capture, takeoff, safety detection, and licensed code compliance depend on corpora that thin LLM wrappers cannot replicate.
- Consider BUILD for workflow-bound categories. RFI drafting against a firm's own historical record, internal spec Q&A, and meeting-to-record automation increasingly favor a thin wrapper over a frontier LLM with retrieval over firm-owned content.
- Check the underlying frontier model on vertical tools. Most AEC AI products sit on top of Claude, GPT, or Gemini. State DOT data-handling policies often require named-model approval, not just vendor approval.
- Read the data processing agreement before pasting agency drawings into a free trial. Several Tier-2 AEC tools retain inputs for model improvement unless that retention is negotiated out.
- Avoid the one-tool-for-everything trap. Each task has a better fit. Silent wrong answers from the wrong tool are expensive.
- Never paste agency-private data into a free-tier consumer chat. Free tiers often train on inputs. Use the paid no-training tier, the API with retention controls, or a self-hosted alternative.
- Confirm acquisition status before signing. Document Crunch is closing into Trimble Q2 2026. StructionSite is a DroneDeploy product line. Newmetrix is now Oracle Advisor for Safety.
RADAR.06
Watch but not yet recommending
Tools on the watchlist. Capability is real but procurement, ownership, or category fit is unresolved.
- Manus AI. Strong on long-horizon tasks but invite-only with a 500K waitlist. Ownership in doubt after the Meta acquisition was blocked by China antitrust in April 2026.
- Mem.ai. The category is being absorbed by Notion AI and Granola. Investor concerns documented in mid-2026 press.
- FDOT spec-interpreter procurement (DOT-RFP-25-9078-SJ). First-of-its-kind state-DOT AI tool. The outcome will shape what other DOTs procure next.
- Trimble SketchUp Connector for Claude (MCP). First agentic-modeling MCP shipped by a major AEC vendor (November 2025). Sets a new bar but at credit-based pricing that gets expensive at production scale.