> ## Content Index
> Fetch the complete content index at: https://www.zhaojinhua.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Making AI Work In DOTs: 3 Levels of Action
- URL: https://www.zhaojinhua.com/3levelsaction/
- Published: 2026-10-03T17:41:47.000Z
- Updated: 2026-10-04T18:17:59.000Z
- Description: From Individual Productivity to Project Teams to Organizational Change
- Author: Jinhua Zhao
- Tags: AI for Transportation, Memos

In AASHTO's 2025 survey of state DOTs, 79% of respondents said training would help them adopt AI, ahead of funding (67%) and pilots (61%).

But training in what?

AI tools change every quarter. What lasts is what a transportation professional can do but AI cannot. The question has moved from "how good is the AI?" to four others:

- What tasks do we delegate to AI?
- Who verifies and signs for it?
- How do we build trust?
- What do we benchmark?

Transportation professionals cannot wait for a perfect AI strategy or the next generation of tools. Start building the professional judgment and capability to use AI now. Professional judgment does not disappear. It matters more.

I suggest 3 levels of action.

Level 1 starts with individual staff. Level 2 is a project team. Level 3 needs the DOT C-suite to change how it decides, trains, and measures.

![](https://storage.ghost.io/c/4a/3d/4a3d3e5d-18ea-4656-a425-4dc671578d23/content/images/2026/10/AI-in-DOT-3-levels-2.png)

### Level 1: individual productivity boost

**1\. AI literacy: learn by doing.**

The reward is immediate: two hours of learning AI can save you time every day.

Start small. Build a tool for your own work, using the AI your agency has approved and material that is already public record.

**2\. Identify which task to delegate to AI**

Start with tasks that are routine, well defined, and easy to check: a first draft of a staff report, a summary of public comments, cleaning a public crash dataset, a simple analysis script.

Keep to yourself the tasks that need subtle context or your signature: setting the goal, reading the room, choosing among options, and signing the results.

If AI could do 50% of the tasks in a job, the rest, which only you can do, becomes more valuable.

**3\. Change your role from producer to director**

Any task has three parts: define the objective (the director), do the work (the producer), evaluate the result (the critic).

- Before AI: you define, you do, you evaluate.
- After AI: you define, AI does, you evaluate.

The first shift: from producer to director and critic. AI may do the work. Defining the goal and judging the result stay with you. The transportation domain defines what success is.

The second shift: to communicator and negotiator. Once you have directed and judged the work, you explain it to others and get the stakeholders to agree.

**4\. Build judgment.**

Ask AI for ten ideas: one is excellent, three mediocre, six garbage.

Your expertise is knowing which is which.

The reason humans keep our job is our long "context window": years of experience, the unspoken cue in a room, the local culture of this agency.

AI knows only what it is given. Domain expertise and institutional knowledge become more important, not less.

**Where does judgment come from?** It comes from practice, and AI cannot practice for you.

**AI techniques**

AI is moving from doing one piece of work to running a whole process: first a task, then a workflow, then an organization of work. The way we instruct it has changed with it.

- Prompt engineering (2022\~): the instruction. Tell AI what you want, for whom, and in what form.
- Context engineering (2025\~): the briefing. Give AI what it needs to know before it starts: your documents, standards, and local constraints. This is how your experience gets into the AI.
- Harness engineering (2026\~): the scaffolding around the model. Set up the work around the AI: its goals, tools, memory, access, guardrails, a definition of success, and a feedback loop that checks the result.
- Graph engineering (2026\~): the org chart of AI agents. Several agents work together, each with a role: who does what, who hands off to whom, what is shared.

**Three Pitfalls to Avoid**

Pitfall #1: AI as commonly used is terrible for learning.

Students who used AI scored higher on practice problems but learned less, because the AI did the thinking. A tutor version that gave hints instead of answers largely avoided the harm.

The same holds at work. Learning takes struggle. Remove all the friction and you remove the learning too.

Pitfall #2: AI can be confidently wrong.

It writes a wrong answer in the same fluent, confident tone as a right one: a reference that does not exist or a plausible number that is not true. The frontier models are getting much better but hallucination remains.

So check the key evidence: the numbers, the sources, the references to codes and standards.

A practical trick: ask a second, independent AI agent to act as an adversarial reviewer. Ask it to find what is wrong, missing, or unsupported. It often catches errors, or at least it highlights the contradicting points.

But even with that adversarial check, accountability stays with the person who signs. If your name is on the work, you answer for it.

Pitfall #3: Your prompts and AI outputs may become public records.

In a public agency, what you type into AI, and what it gives back, may be subject to public records requests, like an email. It may also come up in a lawsuit.

Two practical rules. Write every prompt as if it could be read aloud at a public meeting. And do not put into an AI tool anything you would not put in a public document, such as personal data, bid information, or security details, unless your agency has explicitly approved it.

### Level 2: Project Team

AI has made some of the work much easier. It has not made deciding easier. The bottleneck has moved from producing to verifying.

How AI changes the work: 3 DOT cases of Caltrans, TxDOT and UDOT \[ Mobility Forum Episode\]

#### **Four things a team has to do to own the result**

The cases point to four things a team has to do to own the result when AI does part of the work. They work best in this order.

**1\. Define what success is, how you measure it.**

**2\. Question the process: remove, simplify, and automate.**

AI provides an excuse to examine the process design that should have been done anyway.

Remove useless steps first. Simplify the rest. Decide which steps to delegate to AI. Only then automate them. Otherwise AI will run the old waste and the old biases, only faster and harder to see.

Decide where AI works and where a person decides\*\*:\*\* For example: AI drafts, a person chooses, AI refines, a person verifies.

A person can still say no at each hand-off point. Put these points where a mistake would otherwise go unnoticed.

The O-ring rule: when AI does most of the work, one failed human check can bring down the entire project.

The closer the output is to the physical world, the stricter the check.

Get the data ready. Poor data is the top barrier in the AASHTO survey (76%). Texas DOT first built one data platform from 51 sources. The team owns the data AI works from. Incomplete or biased data gives incomplete or biased answers.

**3\. A named person signs for the result.**

When several individuals or groups share a process, name one owner.

The tool for this already exists: the engineer's stamp. It is the most common way state DOTs govern AI. The state licensing board sets what a stamp means. But the DOT can act now. Its QC manual and its consultant contracts can require a record of which AI tool helped and how the engineer checked the work.

A signature means something only if the signer could actually check the work. Design the workflow so they can. Write down how the work will be checked in the RFP.

**4\. Build or buy, select the vendor, and pilot.**

AI moves the line between what the agency builds and what it buys. Building is cheaper now, so more could stay in-house.

Four options: use a generic AI tool directly, build a function yourself with simple AI tools, ask the agency's internal AI lab, or hire a vendor.

Utah DOT built its search agent in-house. Caltrans and Texas DOT bought from vendors.

When you buy, write the test into the solicitation: a paid proof of concept on your own data, against your own measure of a good result.

Secure three things in the contract. The data stays yours: the vendor cannot keep or reuse it without your approval. The method is documented: you know what data and model the tool uses, and the vendor tells you when they change. Each result can be checked: the tool shows the sources or steps behind it.

Whoever builds it, your team still answers for it, so you must be able to check what it produces.

Then pilot, measure, and scale. Start with one district or one route. Measure the result. Scale up only if the result matches the goal you identified in #1.

### Level 3: Organization Executives

Here are 4 ways DOT leaders can create the conditions to make AI work: trust, team, training, and what gets counted.

**1\. An AI system cannot build trust in itself; a leader has to.**

Start from an actual problem, not from an AI tool: why this problem matters, who will act, who is accountable. Then explain what the AI system does, and ask users whether it helped.

The leader can set the culture. Carlos Braceras, executive director of UDOT, fosters a workplace culture where staff are encouraged to view mistakes and failures as valuable lessons rather than things to hide.

Another effort is to normalize AI. AI is weird. It needs organizational experimentation, and leaders willing to ride the J-curve: a temporary dip before the gain.

Whether AI gets adopted is more about whether people trust it than about the technology.

**2\. Leader, Crowd, Gate, and Lab.**

The leader gives permission and mandate.

The crowd is most staff. They use AI for individual productivity, build small AI tools themselves, and, more importantly, they know which processes are critical and need to improve.

The lab is the technical team that builds advanced AI functions.

The gate is between the crowd and the lab. At Texas DOT, a readiness scorecard screens more than 200 candidate uses on data, a sponsor, resources, and risk. Staff have plenty of ideas. Data decides which go ahead. TxDOT staff gave more than 500 ideas. They became more than 200 candidate uses.

**3\. Protect how people learn.**

Apprenticeship used to play two roles: the junior did the work for the senior, and the senior trained the junior. Both benefited.

There is a risk of an apprenticeship breakdown: delegating the grunt work to AI agents removes the path by which juniors become the senior.

Even if AI can do the working half, we need to preserve the training half. Otherwise, in ten years, no one is qualified to sign.

Massachusetts DOT built an AI assistant, HEKA, to help a wave of new highway engineers find their way through thousands of rules, specifications, and procedures. Utah DOT’s search agent now answers what new staff used to ask a long-serving colleague. That saves time. It may also remove a conversation where juniors learned. None of the DOTs yet tracks whether juniors still learn when AI does their first drafts.

In a DOT this collides with the retirement wave and the years of supervised experience engineers need before a PE license.

The remedy follows from what made the junior competent: not the tedious work per se, but doing the work and having a senior check it. AI can take the volume. The junior still does a checked subset of it by hand. This subset needs to be carefully curated: it builds the critical learning path and skips the purely repetitive work.

Three remedies: intentional friction, AI-free zones, and formal assessment.

- **Intentional friction.** The junior tries first, then sees what AI produces. Ask for this learning mode in the next AI procurement.
- **AI-free zones.** A few core tasks the junior does by hand in the first years, such as checking a design calculation, before AI takes them over.
- **Formal assessment.** A senior checks, at set points, what the junior can do without AI. Give seniors a timesheet code for the hour they spend reviewing a junior’s work. Require the consultant's QC plan to say how its engineers-in-training had their work reviewed.

**4\. Benchmark augmentation, not just automation.**

Automation measures saved hours: 1,300 hours in Caltrans’s Copilot pilot, 22,000 staff hours a year on Texas DOT’s invoices. In Pennsylvania’s statewide pilot, 175 volunteers estimated they saved 95 minutes a day. But it should not be the only benchmark.

Decide in advance where saved hours go, and tell staff and the union. Otherwise staff may stop reporting savings.

Augmentation measures how much more capable the staff become, and that is much harder to measure. Count what you already can: alternatives considered per study, errors caught in review before they reached the road, engineers-in-training progressing to sign.

We still need significant progress on the measurement science of capability increase.

Develop more capable staff; then service quality follows.

### How the three levels fit together

Individuals learn by doing and build judgment. The team owns the result: it defines what success is, redesigns the process before automating it, names one person who signs, and pilots before it scales. Leadership builds trust, protects how people learn, and counts capability, not only hours.

Individuals and teams share a similar way of thinking: define the goal, decide what to delegate, and evaluate the result. An individual applies it to a task. A team applies it to a process.

Directing and judging stay with you. Keep the people who can sign, at every level.

**AI can generate options. It cannot decide what transportation future we should build. That remains our job.**

—Jinhua