Field Note Nº 20·Building with AI

Leadership Presentation Without a Deck

May 2, 2026·7 min read

The Setup

We won a global program. 24/7 coverage across three regions, multiple handoff windows, senior stakeholders spread across time zones and disciplines. Two weeks before our first leadership presentation, my EVP counterpart and I made a decision: no decks. We’d have a conversation with the room, not click through slides at them. That decision changed everything about how I prepared. And the thing that surprised me most wasn’t the final artifact. It was realizing how many separate components had to exist before the artifact could. This is a walkthrough of how I assembled those components, what AI did at each stage, and what I’d tell someone who wants to try this on their own.

My Approach + AI Role

The final output was an interactive HTML. A browser-based presentation that behaves like a deck but can be updated in minutes. But that was the last step. Here’s what came before it, in the order it actually happened.

Step 1: Research the room

I knew who was attending. Titles, roles, where they sat in the org. What I didn’t know was what they cared about… what “good” would look like through their specific lens. I used AI to research each leader’s professional background. Not just current roles, but career arcs, published perspectives, the disciplines they came from. One leader, for example, had deep creative strategy roots. Knowing that changed how I framed operational work to land as brand-building, not just process management. Yes fellow PMs, there are folks who don’t think process and operations lens first. The goal wasn’t to impress anyone with what I’d researched. It was to calibrate emphasis. Same story, different weight depending on who was absorbing it.

Step 2: Capture the game plan conversationally

My EVP and I got on a call and just… talked through it. What points did we want to hit? What should we avoid? Where were we confident? Where did we still have questions? That conversation became the raw material. I fed the transcript to AI and asked it to pull out a narrative outline based on what we’d naturally said and aligned on with each other. Not a polished structure, but a map of our thinking as it had actually unfolded. We reviewed that outline together, edited it, and aligned on the arc. The key here: the outline came from us, not at us. AI organized what we’d already said. It didn’t generate what we should say.

Step 3: Build the source file

This is the component that made everything else possible, and it’s the hidden gem. Over the weeks since we’d been bidding on this work, I’d been compiling a source file, a single structured document containing everything AI would need to understand this program. It included:

Proposal-stage information (how we described the work when we pitched it)
Internal notes about what leadership specifically cared about and would be watching for
Information architecture work (how the program was actually structured)
One critical strategic emphasis: smoothing out the handover windows between regions was the priority before anything else. If 24/7 coverage can’t demonstrate seamless transitions, the entire model is academic. Think of the source file as a knowledge container. Not a brief, not a script. The accumulated operational intelligence of the program in one place. Every insight, decision, and emphasis that had surfaced since the proposal stage, structured so AI could reference it holistically rather than piecemeal. This is what I call AI Kits: you maintain the knowledge once, and it can generate multiple outputs… a presentation today, an onboarding walkthrough tomorrow, a team orientation next month.

Step 4: Codify the brand

The client had specific brand guidelines. Colors, typography, spacing rules, visual treatment patterns. I turned these into a structured reference that AI could apply consistently across every iteration… a codified brand file. This took about 30 minutes upfront and removed much of that last mile polish detailing. Every time I regenerated or iterated, it always looked and felt like the brand.

Step 5: Assemble and generate

Here’s where the components converged. I fed AI three things simultaneously:

The audience research (who’s in the room and what they care about)
The synced narrative outline (what we want to say and in what order)
The source file (everything the program is and why it matters) Plus the codified brand guidelines as a constraint layer. What came out was the first version of the interactive HTML… slide-based navigation, expandable cards for detail without clutter, and a 24-hour timeline visualizing coverage windows across regions, all living in visual harmony within client brand style. It wasn’t a deck. It was a walkthrough that behaved like one but could be updated in minutes, not cycles.

Step 6: Iterate (this is where the real work happens)

The early iterations were mechanical and fast… fixing visual alignment, adjusting slide order, tightening small articulation issues. The substantial work started when my EVP began giving narrative feedback, and that’s where something clicked. Using AI to detect editorial intent. When she gave a note on one slide… say, tightening a phrase or reframing a capability description, I didn’t just fix that slide. I fed the edit to AI and asked: what’s the intent behind this change? What principle is this feedback expressing? Then I asked AI to carry that principle across every other slide where it might apply. One example: an early version led with operational scope distinctions on the title card. The feedback was that it was too granular for a first impression. The underlying intent? Don’t pull the room into “what exactly do we do” before they’ve oriented “where am I?” That principle didn’t just fix the title. It changed how I sequenced information across the entire presentation. This is wordsmithing at scale. One note, detected as a principle, applied everywhere it’s relevant. One discipline that’s easy to forget: the source file and the HTML artifact are a pair. When iteration produces fundamental changes in the narrative or thinking… not just visual tweaks, but actual shifts in how you’re framing the work. Carry those edits back into the source file. It feels like a chore in the moment. But the source file is your container of truth. The HTML is just one expression of it. If you only iterate on the artifact and neglect the source, you end up with a beautiful presentation and no record of the thinking that got you there. Six months later, someone asks you to adapt the story for a different audience, and you’re staring at an HTML file trying to reverse-engineer your own logic. Update the source. It pays dividends.

What Actually Happened

The presentation landed. One of the senior leaders closed the meeting with: “That’s a cool presentation, Ted. I haven’t seen that before”. But the build wasn’t without surprises. The responsiveness assumption. I assumed AI would inherently know how to build responsive HTML… that the presentation would resize gracefully if someone dragged the window or opened it on a phone. It didn’t. Layouts broke in ways that would have been embarrassing live. So I went on a side quest: I built a skills pack (a reusable set of best practices that AI references automatically in future builds) specifically for responsive interactive HTML. Viewport handling, font scaling, layout breakpoints. It took maybe 30 minutes. Every interactive HTML I’ve built since starts with that knowledge baked in. I also learned that emoji flags render differently across platforms. What looked perfect on my Mac showed up as two-letter country code boxes on Windows browsers. The fix was embedding SVG flags directly, which is completely outside AI’s awareness until you explain the constraint. The lesson in both cases: when you hit a repeated friction points during iteration, ask yourself whether there’s a skill waiting to be captured. The responsiveness issue wasn’t a distraction from the presentation work. It was the reveal to me for what to systematize.

The Real Insight

The valuable product isn’t the artifact. The components are. The interactive HTML impressed people, but it was the last thing I built. The audience research, the conversational game plan, the source file, the codified brand… those are the things that made the output good. Without them, AI would have generated something generic and hollow. With them, it generated something that sounded like us because it was built from what we actually think. What dawned on me: a deck lives and dies. It’s the last version someone exported. But this AI Kit approach? This is something we can keep updated. That’s the shift. We didn’t build a presentation. We built a communication layer that happens to present well. And because the source file is maintained independently, the next output (in whatever format) takes hours instead of weeks.

Try This If…

You’re building a high-stakes presentation and want to try this approach. Here’s the component checklist: Before you start building: Know your room. Research each audience member’s professional lens. What does “good” look like to them? This calibrates everything downstream. Before you outline: Talk it through with a partner. Record or transcribe the conversation. Let AI extract the narrative skeleton from how you naturally described it. Edit together. Before you generate anything visual: Build the source file. Compile everything you’ve accumulated about the project… proposals, strategy notes, architectural decisions, priority calls… into one structured document. This is the knowledge layer. The artifact is just its expression. When you generate: Feed AI the audience research, the narrative outline, and the source file simultaneously. Compound context produces compound quality. When you iterate: Use feedback as pattern detection. One note on one slide might contain a principle that applies across the whole artifact. Ask AI to detect the intent and carry it downstream. And when iteration shifts the actual narrative or thinking (not just visuals), carry those changes back into the source file. The artifact is disposable. The source file is the asset. When you hit repeated friction: Capture the skill. Responsive design, brand consistency, platform-specific rendering… if you’re fixing the same category of thing more than twice, spend 30 minutes packaging those best practices. Your future self will start where you’ve left off here every time.

Systems Lens

Every time you build something complex with AI, you’re generating two kinds of value: the deliverable itself, and the reusable knowledge about how to build that type of deliverable. The source file compounds. The skills pack compounds. The codified brand compounds. Each project teaches you what to systematize, and the next project starts further ahead because of it.