HC PPMD Leadership Summit · 2025 · GenAI
Ambition
Animals.
12 posters for an Austin event. Then someone said "what if everyone gets one?" — and we had 60 days to figure it out.
12 posters for an Austin event. Then someone said "what if everyone gets one?" — and we had 60 days to figure it out.
It started as 12 weird posters for a conference wall. It became a GenAI production system that put a personalised animal in every leader's inbox.
Our intent was simple: create a series of super-sized ambition animal posters to adorn the ballroom walls of the HC Leadership Summit in Austin. We explored prompt structures, image gen models, and a slew of silly combinations — and had our 12 favourites.
Then the project lead said, in true Oprah style: "What if everyone can create their own ambition animal?" And suddenly we had 60 days to figure out how to scale from 12 cool posters to potentially 335 unique animals — one for every leader in attendance.
What if everyone can create their own ambition animal?
I found an opportunity to run a test: 40 leaders were meeting in New York. Could we create animals for each of them by the end of their meeting? A team of 5 spent the day receiving survey responses, researching animals with matching characteristics, and using Adobe Firefly to craft prompts that merged each animal with the participant's product and ambitions.
At 4:30 that day, we delivered all 40. And received some very valuable lessons in return.
What was hard
Text placement varied wildly by response length. Background color randomization could clash with certain animals. Every edge case required a designer to catch it.
Where design expertise was irreplaceable
Animal selection, prompt crafting, reviewing final quality — these needed a creative eye. Automation couldn't replace the judgment calls that determined whether an animal was right.
What could be automated
The Photoshop template build — clipping, compositing, file naming, batch saving as JPG. Pure repetition, which is exactly what automation exists for.
The 8-step human process (left) and a sampling from the NY experiment — the Photoshop file grid showing the variety of output from 40 leaders
I am not one for tall tales of perfect projects. Poorly merged bodies, animals with extra limbs, prompts that produced something completely off — we saw our fair share of failures. But every failure informed a process improvement. The retrospective after the NY experiment is where we actually designed the scaled system.
The failures were the curriculum. Each one sharpened our prompting instincts, revealed where automation needed guardrails, and showed us exactly where a human had to stay in the loop.
A sampling of the GenAI failures — from extra limbs to completely off-target merges. Each one taught us something about prompting, model limitations, and where human review was non-negotiable.
We ran a design team retrospective and mapped every step of the human process against a single question: can this be automated without sacrificing quality? The ideal workflow separated three distinct lanes — what the attendee does, what AI handles automatically, and where the designer stays essential.
In the final 2 weeks, 2 engineers used Claude to build a simple application that compressed the entire pipeline into a single tool running on the designer's machine.
Things we standardised
One unified background for all event posters — removing the color decision and unlocking Photoshop batch processing that saved every file in seconds
The Claude application
Intakes survey responses → suggests matching animals → drafts a full prompt → generates the image via ChatGPT 4.0 in 10–20 seconds per animal
Human still in the loop
Designer selects the best image, edits the prompt if needed, reviews the final file and marks as FINAL — creative judgment stayed human, mechanical production became automated
The ideal automated workflow — three distinct lanes: attendee input, AI/automation pipeline, and human-in-the-loop creative review
Two engineers used Claude to rapidly build a local application that compressed the entire multi-step workflow into a single tool. Running on the designer's machine, it handled the parts that were slowing us down — without removing the creative judgment that made the animals worth receiving.
The Ambition Animals Accelerator — built with Claude. Animal selection → prompt editing → image generation → batch records, all in one local application
Each animal was unique to its leader — built from their survey answers, their ambition keywords, the product that represented their drive. The nautilus tape roll, the squid golf clubs, the peacock standing alone. Every one of them slightly surprising.
The shift from a team of 5 spending 5 hours on 40 animals to a streamlined human-in-the-loop process that delivered 335 unique animals on time represents more than a process win. It represents a team that fundamentally rewrote how their work gets done — and proved that AI adoption grows fastest when the stakes are real and the purpose is genuine.
Underneath these adorable animals is a team that built genuine AI capability — not through a training program, but through a project with real stakes and a fixed deadline. Experimentation with new AI technologies is critical for adoption. Data shows adoption grows when employees have psychologically safe spaces to try, fail, and learn.
I believe in creating with purpose. Even when the purpose is fun.
Stronger in text-to-image processes and models — we understand what's feasible, what's hard, what limitations exist
Learned to scale an AI process — from something to play with, to something that works for us
Rewrote our role in how the work gets done — human creativity directing AI production, not competing with it
Brought joy to others through design — 335 people received something unexpectedly personal
Poorly merged bodies, animals with extra limbs, prompts that produced something completely off — we saw our fair share. But every failure informed a process improvement. The retrospective after the NY experiment was where we actually designed the scaled system. We couldn't have built the automation without the manual mess first.
The single biggest quality improvement came from standardising the background. Removing one variable — color — eliminated a whole category of failures and unlocked Photoshop batch processing. Simplification enabled scale. The more decisions we took off the table, the faster and more consistent the output became.
The hardest part wasn't building the automation — it was being honest about what shouldn't be automated. Animal selection, prompt creativity, final quality review: these needed a human. File naming, compositing, export: these didn't. Mapping that boundary clearly is what made the system work at 335.
This project started as decoration for a conference wall and became a GenAI capability-building exercise disguised as a gift. The joy of receiving something personalised created psychological safety to experiment. Fun with purpose is still purpose — and sometimes it's the fastest path to genuine organisational learning.