Speaking

I have been speaking to developer audiences for most of two decades, and interviewing people who build software for about as long. These days I talk about AI systems: what they do well, where they fall over, and what it takes to get one working reliably for people who did not read the instructions.

I do not do hype talks and I do not do doom talks. What an audience gets is someone who has shipped these systems describing what actually happened, including the parts that did not work. If your audience has sat through three keynotes about transformation and wants somebody to just explain the thing, that is the talk I give.

Topics

For developer audiences

Building AI that cannot make things up. Grounding, retrieval, and hard citation requirements, drawn from building a legislative research tool where a confidently wrong answer is worse than no answer. Covers chunking, citation enforcement, and the feedback infrastructure that tells you when you got it wrong.

Nobody talks about what it costs to run. Inference economics as a design constraint rather than a surprise. Why features that work in a demo fall apart at real volume, and how to design the cost structure deliberately.

Choosing a model is not a spec sheet decision. What I learned running comparisons across a corpus of 4,600 podcast episodes. The model that writes the best summaries is not the one that does the best entity extraction, and neither is necessarily the one worth paying for.

Agents that hold a conversation for twelve weeks. Longitudinal state, handoff between specialised agents, and why most agent products quietly fall apart around week three.

Getting a model to disagree with you. Models are trained to be agreeable, which makes them structurally bad at coaching, review, and anything else where the useful answer is “no.” What it takes to work against that.

For civic and legislative audiences

How these systems actually work. Plain language, no math, no jargon. What a model is doing when it answers, why it is confident when it is wrong, and what that means for anyone writing rules about it.

What “artificial intelligence” means in a bill. Definitions carry the entire weight of a statute, and AI definitions are unusually hard to write. Where the common formulations catch too much, too little, or fail to survive the next model release.

What disclosure rules ask software to do. “Disclose when asked” is one sentence in a statute and a genuinely hard engineering problem. A worked example of the gap between what a law says and what a system has to be built to do.

Where the risk actually sits. Most public concern is aimed at model capability. Most real harm shows up at the deployment layer, in how a system is wired into a business and what happens when it fails. Why that gap matters for regulation.

Two theories of AI regulation. The EU regulates the system. Utah regulates disclosure. What each approach assumes about what is knowable and enforceable, and what the EU’s recent deadline deferrals suggest about which assumption is holding up.

Past talks

I have spoken at developer conferences for most of two decades, mostly about Ruby, JavaScript, and the business side of building software. Speaking about AI is newer for me, though it is what I have been building for the last few years.

[PLACEHOLDER — 3 to 6 selected earlier talks: title, event, year, video link where one exists]

For organizers

Everything here is free to use without asking.

Short bio

Charles Max Wood builds AI systems and explains how they work to the people who have to make decisions about them. He has been writing software for over 20 years and hosting podcasts for developers for 18, across 25 shows and more than 4,600 episodes.

Long bio

Charles Max Wood has been building software for over 20 years and hosting podcasts for developers for 18 of them, across 25 shows and more than 4,600 episodes. He now builds AI systems and advises on AI policy through Millwright, including a legislative research tool grounded in bill text and a production AI pipeline running across his own podcast network. He is vice chair of the Utah County Republican Party and lives in Utah.

Headshot — [PLACEHOLDER: link to a high-resolution headshot]

A/V — [PLACEHOLDER: laptop and HDMI, adapter needs, whether you need audio for demos]

Travel — [PLACEHOLDER: based in Utah, willing to travel where/how far, remote talks yes or no]

Booking lead time — [PLACEHOLDER: how far ahead you want to be asked]

Booking

Tell me the audience, the slot length, and roughly what you want them to walk out knowing. I will tell you honestly whether I am the right person for it.