A year of building with AI put me on three stages. I liked the message more than the microphone.

TL;DR: In July 2025 I started building with AI to see how far it could take me. It took me further than I planned: onto court conference stages in Albuquerque, Los Angeles, and Riverside, talking about frontline staff leading small AI pilots. This is the written version of that talk, the one thing I would change about it, and what three rooms taught me about doing something I don't enjoy.

I walked into the room in Albuquerque with a deck I had rehearsed for weeks and no idea how any of it would land.

It was February 2026, and I was there to speak at the NACM 2026 Midyear Conference, the National Association for Court Management's national gathering. I was also the only person there from my court. No coworker in the third row. Nobody to grab coffee with afterward and ask, "Was that okay?"

I had prepared. I had also never done this before. That combination felt high risk in a way that had nothing to do with slides. If it went badly, I would carry it home alone.

It was a familiar feeling. I had felt it once before.

Alone in the room, again

In July 2025 I applied to the California Community Colleges AI Summer Camp. Nobody suggested it. I got one of 100 seats, out of about 1,300 applicants, and I was the only court professional on the list.

I was so worried about keeping up with computer science students that I spent the weekend before camp building a jury chatbot called Your Honor, I Object, just to see if I could. At camp, at 4:59 PM on day one, my mentor asked me to build a PDF extractor the rest of our project depended on. I had ChatGPT open that night and my team beside me. Our team placed second out of fourteen.

I started building with AI to find out how far it could take me. Speaking turned out to be the same experiment on a different stage. When the chance to present at NACM came up, I said yes for the same reason I applied to camp. I wanted to find out.

The second time was harder

Five months later, in July 2026, I spoke at a court in Los Angeles. I expected it to be easier. I had done this once. I knew the talk.

I was so much more nervous.

Some of that was the room. The Los Angeles Superior Court is the largest trial court in the country, and they were excited to have me. Excitement from your hosts is lovely. It also raises the bar you set for yourself.

Some of it was me. After Albuquerque, I expected more of myself, not less.

And some of it was the name on my badge. When I speak, I'm not only speaking for myself. My court's reputation walks into the room with me. Santa Barbara is a smaller court, and standing in front of the biggest one, carrying that name, felt heavier than standing up alone.

I didn't expect the second time to be harder than the first. The first time, I had nothing to live up to.

Two months, and still editing

The third talk was yesterday, October 16, 2026, at the Southern Region Finance Officers Meeting, hosted by Riverside Superior Court. The theme was the Currency of Curiosity.

I had two months to prepare. I was still changing slides the day before.

One thing helped more than I expected. I asked Claude to turn my speaker notes into an MP3, and I listened to my own talk in the car. AI helped me build the pilots in the talk, and then it helped me rehearse the talk about them.

It went well. My jokes got laughs from a room full of finance officers, which I am counting as a win.

Then I walked off and waited for the feeling of accomplishment. What came instead was relief that it was over.

That was when I knew. I'm proud of this talk. I love its message. I don't love public speaking.

What the talk says

The odd thing about all those nerves: I trusted the work, even when I didn't trust myself to present it. I knew what I had built, what had failed, and why I stopped. That was the part I wanted people to hear.

The talk is called Building Bridges, and its argument is that frontline court staff can lead AI innovation under the right conditions. By lead, I mean five steps: initiate, ask for approval, build, produce something, and report back.

A bridge gets built from both banks. My side was curiosity. My court's side was boundaries, and a manager who reviewed my proposals and said yes to small tests. Every pilot started the same way:

• Approved in writing first
• Public, redacted, or fictional sample data only
• A person reviewing every result
• A stopping point written down before I started Both court pilots stopped, and that is my favorite part of the story.

Write of Access tested whether a script could read handwritten entries on records request forms. I set a 90% accuracy bar before I started. It came in around 80%, so I stopped and looked at the misses. They clustered around checkboxes, signatures, and dates, where handwriting overlapped the printed form. The useful answer wasn't better technology. It was a form with less handwriting on it.

Hung Jury tried to automate part of our annual jury data report for the Judicial Council. I started on a Friday afternoon and stopped Monday morning. Three approaches hit the same wall: messy data in the source system. The existing Excel workbook was still the best tool for the job. The fix was a data conversation, not a coding project.

I'd rather find a dead end in a weekend than in three months.

After the pilots, I kept building outside of work. Themis Lex helps court staff see where AI can help in their role and where it must stay out. Verity Lex checks a California court's public record against the rules of court on AI. Both started with questions from my own job.

What I'd change

Building Bridges is a non-technical talk on purpose, and I say so at the start. It's about people, permission, and small tests, not code. So when I described the pilots, I wasn't thinking about the technical layer at all.

In Riverside, the concern I kept hearing was whether AI had run my pilots on court data. That's when I realized that for a court audience, how AI was involved might not be a technical detail. It might be the first question. Here's the answer I'd add.

That table holds three different roles for AI, and I think they're worth keeping apart:

• Generative AI answering questions. Only Your Honor, I Object does this. It was me testing myself before camp, and it never went near a court system.
• AI helping me write code. ChatGPT did this for both pilots. It helped me write the Python scripts. It didn't run them.
• Machine learning reading sample documents. Only Write of Access did this. My code sent redacted sample forms to Amazon Textract, an AWS machine learning service that reads printed and handwritten text from scanned documents. Textract doesn't generate conversational answers. It extracts text and structured information, like form fields, from documents. I checked every result against the bar I set before I started. Sending those redacted samples to Textract was part of the written pilot approval. I think that distinction matters most to people who are nervous about AI in courts, and those are the people the talk is for. Next time, it goes on a slide.

The repos

Write_of_Access/
├── docs/
│   └── validation_log_template.md
├── input/
│   └── copyrequest (5).pdf
├── output/
│   └── sample_form_1_pg1_processed.txt
├── scripts/
│   ├── convert_and_preprocess.py
│   ├── textract_call.py
│   ├── textract_document_ocr.py
│   └── textract_test.py
└── README.md

I'd still say yes

If another court invited me tomorrow, I would say yes. I've been turning over what that says about me.

I think it says the message matters more to me than my comfort does. Enjoying something and finding it worth doing are two different things. Building with AI is both for me. Speaking, so far, is only the second.

In each of those rooms there was probably someone who asks why we do things the way we do them, and who tinkers with tools nobody assigned. That person is who the talk is for. I'd stand up again for them.

I'd still rather write it down. Next time, maybe with a coworker in the third row.

Your turn

What's the thing you'd do again even though you didn't enjoy it?

Drafted with Claude in October 2026 from my speaker notes for Building Bridges. The stories, the nerves, and the decisions are mine. Talk resources and templates: linktr.ee/ljcordero.

AI Assisted. Human Approved. Powered by NLP.