How Transportation Engineers Decide Which AI Workflows to Deploy First

Your queue includes projects that involve repetitive tasks: searching agency manuals for the right specification, setting up microsimulation scenarios from scratch, reviewing pavement inspection images one by one. Somewhere between those tasks and a full AI transformation program lies the gap most transportation engineering firms haven't crossed yet.

AI in transportation engineering is genuinely useful today— but only in specific workflow categories. Transportation engineers and Department of Transportation professionals don't need more headlines about what AI might eventually do. They need a practical map of where it works right now— and where professional accountability still draws the line. Here's that map.

The Federal Highway Administration (FHWA) has made its position clear. The agency published a multi-year AI program roadmap (FHWA-HRT-26-063) covering fiscal years 2025 through 2036, aimed at accelerating the responsible adoption of AI to reduce traffic crashes and optimize infrastructure¹. This is a ten-year institutional investment, not a pilot program. And the Texas Department of Transportation (TxDOT) has already mapped 230 potential AI use cases— with 22 of those currently active as of its December 2024 AI Strategic Plan².

This article covers the five workflow categories where AI tools are proven and deployable right now, the categories where they aren't, what's holding most DOTs back, and how to sequence a realistic starting point for implementing AI across your operations.

Where AI Is Production-Ready in DOT Work Today

Five categories of transportation engineering work have AI tools that are proven and deployable right now: traffic signal optimization, documentation and manual research, infrastructure inspection, pavement and asset management, and traffic simulation setup.

Traffic Signal Optimization

AI-driven adaptive signal control in Maricopa County, Arizona reduced average vehicle intersection delay by 46 percent— from 29.5 seconds to 13.7 seconds per vehicle— and cut estimated operating costs by 75 percent compared with conventional signal operations³. This isn't a research result. It comes from the U.S. DOT ITS Knowledge Resources program, and the data is public.

The system uses real-time camera and detector data to adjust signal timing continuously across single intersections, coordinated corridors, and wider networks. Beijing's Tongzhou District deployed a similar architecture, reporting a 32.5 percent reduction in average travel time on major urban roads⁴. Lead with Maricopa County when talking to a U.S. agency— that's the relevant precedent— and use Beijing as confirmation that the pattern holds internationally. Deployment requires traffic data infrastructure already in place: cameras and detectors are the prerequisite.

Documentation and Manual Research

MassDOT's HEKA (Highway Engineer Knowledge Agent) is the clearest example of an AI tool already operating in a U.S. DOT environment. Developed through the Massachusetts Governor's AI for Impact program with Northeastern University, HEKA launched in November 2024 trained on MassDOT's SOPs, Engineering Directives, and permit documents⁵. Engineers using it completed manual research tasks 78 percent faster⁵.

That result matters— and it matters to read precisely. HEKA is a documentation lookup tool. It helps engineers find information in existing DOT manuals faster; it doesn't design roads or generate specifications. MetroPlan Orlando's June 2025 white paper confirmed parallel value: AI tools demonstrated practical results in drafting technical memos, interpreting signal timing SOPs, and day-to-day workflow tasks across transportation asset management contexts⁶. The pavement management and asset tracking workflows that parallel this documentation work present the same pattern— the AI amplifies what your team already knows. But it doesn't build that knowledge for you.

Infrastructure Inspection

Computer vision AI flags structural anomalies in bridge and pavement inspection images, surfacing items for engineer review rather than replacing that review. TxDOT has active use cases in this category within its December 2024 AI Strategic Plan². The pattern recognition that AI excels at— finding consistent visual signatures across thousands of images at scale— matches what field inspection data provides. An engineer still reviews and decides. The AI reduces the time to flag, not the time to judge. And the judgment is the job.

Pavement and Asset Management

AI analyzes sensor and inspection data to predict pavement condition deterioration and prioritize maintenance schedules. This works best when historical pavement condition data is clean, consistent, and complete— a prerequisite most large state DOTs can meet, but smaller agencies often can't. TxDOT has active use cases in both maintenance decision support and predictive asset management². The data quality prerequisite isn't a minor asterisk. It's the determining factor in whether the pilot produces useful output or just expensive noise.

Traffic Simulation Setup

Oak Ridge National Laboratory developed RealTwin, a tool that automates scenario generation for microsimulation in SUMO, VISSIM, and AIMSUN⁷. For engineers running dozens of network scenarios, the manual setup time reduction is real. The engineer still validates scenarios and interprets results— RealTwin automates the setup process, not the engineering judgment that follows it.

Where AI Falls Short in Transportation Engineering

AI is not production-ready for final design decisions, PE-stamped deliverables, or any output where a professional's license is directly on the line for public safety determinations.

ASCE's AI RACE (AI Roadmap for Application in Civil Engineering) is explicit about this boundary: "Artificial intelligence can enhance engineering practice, but responsibility for public safety, professional judgment, and responsible decision-making must remain with people."⁸ This is professional consensus, not a regulatory statute— but it reflects both the actual state of AI capability and the correct division of labor.

Where AI isn't ready in current transportation engineering practice:

• Final design and PE-stamped drawings: No AI tool currently produces sealed drawings. Engineer of record accountability for any public safety determination remains a human requirement.
• Site-specific contextual judgment: AI finds patterns in data; it does not apply topographic constraint analysis, right-of-way context, or the accumulated judgment of an engineer who knows a specific corridor. Kittelson & Associates put it directly: "AI is not a replacement for engineering judgment or physics-based modeling."⁹
• NEPA environmental review: Platforms like PermitAI claim up to 50 percent efficiency gains for federal subject-matter experts¹⁰— but those are vendor projections, not independently verified results. AI for NEPA review is still at a nascent stage in U.S. practice.
• Highway design validation: AI can surface options and flag anomalies. It cannot substitute for physics-based simulation validation or for the engineer who has to stand behind the output.

You already know where the line is. Stating it clearly here is still the right call.

What's Holding DOTs Back

Data quality is the primary barrier to AI adoption at most DOTs— 76 percent of state DOTs cite data quality, fragmentation, and availability as their top obstacle, according to a 2024 National Academies survey of 43 state agencies¹¹.

Most transportation firms researching AI spend their time asking what AI can do. The more valuable question is which three workflows in their practice have data that is clean enough to actually use today. That reframe is what separates agencies currently deploying from agencies still stuck in the research phase.

The three real barriers:

• Data quality and fragmentation: Legacy DOT data is siloed, inconsistent, or paper-based. AI systems need clean, structured, consistent inputs. Getting data right is the pre-work; it's not something AI solves for you. Full stop.
• Workforce readiness: The gap here is a training, validation-framework, and trust gap— not a raw capability one. Engineers need to know when to trust AI outputs and when to override them. Miovision noted that AI adoption in traffic engineering "is gated by accountability, not capability."¹²
• Institutional accountability: Agencies can't publicly deploy a system they can't explain— to their board, their funding sources, the public. Building a governance framework isn't optional overhead. It's a deployment prerequisite.

For a broader AI governance and accountability framework, the pattern holds across professional services: data readiness first, governance second, deployment third.

The National Academies found that while AI is advancing rapidly in the broader transportation sector, public agency adoption remains in relatively early stages¹³. Small and rural DOTs face steeper versions of all three barriers— most documented success comes from large, well-resourced agencies like TxDOT and MassDOT, which have data infrastructure and cross-functional teams that smaller organizations don't.

How Transportation Agencies Are Starting: The Phased Approach

The DOTs seeing early results started with documentation and research assistance— low risk, high ROI, and executable without new data infrastructure— then moved progressively toward applications that require clean operational data.

The ITE 2024 Developing Trends report established that AI readiness hinges on training, policy, and validation frameworks— not on AI capability alone¹⁴. TxDOT built ethics and governance in from day one of its three-year roadmap, not as a follow-up². That sequencing choice is worth copying.

Phase 1: Documentation and research AI Train AI on existing SOPs, manuals, and directives. No new data infrastructure required. MassDOT's HEKA is the model: build on what exists, target the highest-friction repetitive task, measure the result.

Phase 2: Operational data analytics Traffic signal data and pavement inspection records are typically the cleanest datasets a DOT owns. Start AI analysis there— the data quality problem is more manageable, and results are measurable in concrete operational terms.

Phase 3: Inspection and monitoring automation Apply computer vision to existing camera infrastructure. Run predictive maintenance models on cleaned asset management records. TxDOT's active use cases in both categories confirm this phase delivers— but only when the governance framework is already in place, not still being designed.

FHWA's AI-Enabled ITS Capability Maturity Model (FHWA-JPO-24-136) provides a formal self-assessment framework for agencies to gauge their own readiness before committing to deployment¹⁵. For building an AI implementation strategy for your firm, the phased sequence matters as much as the tools you choose.

TxDOT's development process offers one useful signal: the agency involved engineers, IT experts, planners, and employees from across the state in creating its use cases². AI strategy at a leading DOT is a cross-functional exercise. Not an IT project.

FAQ: Common Questions About AI in Transportation Engineering

Before committing to a starting point, most transportation engineers and DOT staff ask the same set of questions. Here are the ones that come up most— answered without hedge.

What is AI used for in transportation engineering?

Traffic signal optimization, infrastructure inspection, pavement condition assessment, documentation assistance, traffic simulation setup, and asset management analytics are the current production-level and pilot applications. The Institute of Transportation Engineers organized these into six overarching categories in its 2024 Developing Trends report¹⁴. FHWA's program roadmap establishes the federal framework for expanding this across the full transportation infrastructure lifecycle over the next decade¹.

Is AI replacing transportation engineers?

No. ASCE's AI RACE roadmap is explicit: AI enhances engineering practice, but professional responsibility for public safety and design decisions stays with licensed engineers⁸. The question of whether AI violates PE professional standards has a settled professional consensus answer— it does not, when used to support human judgment rather than replace it. The PE stamp stays with the person.

What AI tools do state DOTs actually use?

MassDOT uses HEKA, a generative AI assistant trained on DOT manuals and SOPs that launched in November 2024⁵. TxDOT has 22 active AI use cases as of its December 2024 strategic plan, covering traffic monitoring, incident detection, and maintenance decisions². Oak Ridge National Laboratory developed RealTwin for automated traffic simulation scenario generation for SUMO, VISSIM, and AIMSUN⁷.

Why aren't more DOTs using AI?

Seventy-six percent of state DOTs cite data quality, fragmentation, and availability as their primary barrier, per a 2024 National Academies survey of 43 state agencies¹¹. Workforce readiness and institutional accountability processes are secondary barriers. The data infrastructure problem is solvable— it just comes before the AI deployment, not after.

What does FHWA say about AI in transportation?

FHWA published a multi-year AI program roadmap (FHWA-HRT-26-063) covering FY25–FY36 and an AI-Enabled ITS Capability Maturity Model (FHWA-JPO-24-136) to help agencies assess and sequence their readiness¹¹⁵. Both documents represent formal federal investment in AI for transportation infrastructure— not experimental interest, but institutional commitment with an eleven-year planning horizon.

The Map You Actually Needed

AI in transportation engineering is not a question of whether it works— it does, in specific contexts, with documented results from U.S. DOT programs. The question for engineering firms and DOTs is which of their existing workflows have data that is clean enough to use today.

Those five production-ready categories— signal optimization, documentation lookup, infrastructure inspection, asset management, and simulation setup— share a common characteristic. They all operate on data transportation agencies already collect. You don't need new data infrastructure to start. You need to find where your existing data is already clean enough to act on.

The engineering knowledge is yours. AI amplifies what you already know— it doesn't replace the judgment behind the PE stamp. TxDOT confirmed it: involve your engineers in identifying the use cases, not just your IT department. That's where the map gets useful.

If your firm is evaluating where AI fits in your DOT project work, Dan Cumberland Labs helps engineering and AEC firms map the right starting point to their specific workflows. The data quality conversation usually comes first— and it's almost always shorter than firms expect.

References

1. Federal Highway Administration, "AI for Safety and Efficiency Program Roadmap (FHWA-HRT-26-063)" (2026)— https://highways.dot.gov/sites/fhwa.dot.gov/files/FHWA-HRT-26-063.pdf
2. Texas Department of Transportation, "TxDOT Launches AI Strategic Plan" (2024)— https://www.txdot.gov/about/newsroom/statewide/txdot-launches-ai-strategic-plan.html
3. U.S. DOT ITS Knowledge Resources, "AI-Driven Adaptive Signal Control Technology, Maricopa County AZ" (2025)— https://www.itskrs.its.dot.gov/2025-b02021
4. International Telecommunication Union (ITU), "Beijing's Autonomous Intelligent Signal Light" (2026)— https://www.itu.int/hub/?p=74491
5. Northeastern University, "Northeastern Graduate Builds AI Assistant for MassDOT" (2025)— https://news.northeastern.edu/2025/07/14/ai-assistant-massdot-engineer/
6. MetroPlan Orlando, "Leveraging AI in Transportation Engineering White Paper" (2025)— https://metroplanorlando.gov/wp-content/uploads/Leveraging-AI-in-Transportation-Engineering-White-Paper-2025-06-FINAL.pdf
7. Oak Ridge National Laboratory, "Developing an Automated Microscopic Traffic Simulation Scenario Generation Tool" (2025)— https://www.ornl.gov/publications/developing-an-automated-microscopic-traffic-simulation-scenario-generation-tool/
8. American Society of Civil Engineers, "AI RACE: AI Roadmap for Application in Civil Engineering" (2026)— https://www.asce.org/initiatives/ai-race
9. Kittelson & Associates, "How AI Thinks About Transportation Planning" (2024)— https://www.kittelson.com/ideas/how-ai-thinks-about-transportation-planning/
10. Shipley Group / PermitAI, "Deploying AI and Emerging Technology for More Efficient Environmental Reviews" (2025)— https://www.shipleygroup.com/deploying-ai-and-emerging-technology
11. National Academies of Sciences, Engineering, and Medicine, "Artificial Intelligence Opportunities for State and Local DOTs: A Research Roadmap (NCHRP Project 23-12)" (2024)— https://www.nationalacademies.org/read/27865/chapter/12
12. Miovision, "Barriers to AI Adoption in Traffic Engineering" (2024)— https://miovision.com/blog/barriers-to-ai-adoption/
13. National Academies of Sciences, Engineering, and Medicine, "Artificial Intelligence Opportunities for State and Local DOTs: A Research Roadmap (NCHRP Project 23-12)" (2024)— https://www.nationalacademies.org/read/27865/chapter/10
14. Institute of Transportation Engineers, "Developing Trends 2024: AI in Transportation" (2024)— https://www.ite.org/ITEORG/assets/File/public/Developing%20Trends%202024_final.pdf
15. Federal Highway Administration, "AI-Enabled ITS Capability Maturity Model and Readiness Checklists (FHWA-JPO-24-136)" (2024)— https://rosap.ntl.bts.gov/view/dot/93374