I’ve been thinking for a while about how to make Neuralift’s value proposition more tangible and fun to engage with. We’re a deep learning segmentation platform - not the easiest thing to make exciting in a cold outreach email. So I built a game instead.

DungeonLift is a pixel-art dungeon crawler where you play as a customer data professional descending through the “Data Dungeon”, fighting the enemies of insight in order to find hidden customer segments.

You pick a hero, a Data Analyst, Marketing Exec, Data Scientist, or Chief Experience Officer, and battle your way through six procedurally generated levels. The enemies are all themed around the real obstacles that people face when trying to understand their customers: Noise, Bias, Data Silos, rigid Rules Engines, death by Spreadsheet, and the final boss… the SQL Query.

Along the way you collect ML-themed items like GPU Shields, Denoisers, Neural Cloaks and Attention Heads. There are quiz gates where you answer data and marketing questions to unlock segment discoveries. And if you find Neuralift in a chest, you get powered up and clear the level.

The whole thing is pure HTML5 Canvas and vanilla JavaScript. No dependencies, no build step, runs entirely client-side. I built it over a weekend with a lot of help from Claude Code; the concept, theming and game design were mine but Opus 4.6 was genuinely incredible at turning ideas into working game logic at speed. It has a Wordle-style share card at the end too so people can post their scores.

Why This Matters More Than It Might Seem

Distribution as a startup in 2026 is genuinely brutal, and ironically AI is a big part of why.

First, the barrier to building has collapsed. Tools that took months to ship can now be prototyped in a weekend. That’s incredible for builders, but it also means every category is suddenly drowning in new entrants. When everyone can build, the bottleneck shifts entirely to distribution, and standing out in a market with 10x the noise is a fundamentally different challenge to the one startups faced even two years ago.

Second, and arguably worse, AI has industrialised outbound. Every founder and their SDR team now has access to tools that can generate thousands of personalised-looking emails at the push of a button. The result? Buyers have tuned out. The inbox is flooded with messages that read like they were written by someone who cares but obviously weren’t, and the response rates across the industry reflect it. The very channel most startups rely on for early traction is being poisoned by the volume of low-effort, LLM-generated outreach.

Third, the platforms themselves are shifting. When your potential buyer asks an AI assistant to recommend a tool in your category, you’d better hope your brand has enough signal to surface. SEO mattered before, but discoverability in an AI-mediated world is a different game entirely, and one where incumbents with large content footprints have a massive structural advantage over startups.

So the playbook that worked in 2022 - ship fast, run outbound, rank on Google - is breaking down. Not completely, but enough to force a rethink.

The flipside, though, is that all of this creates a real opening for anyone with the critical thinking to see these shifts, the creativity to try something different, and the bravery to actually ship it. When most of the market is sending the same AI-generated emails and running the same playbooks, doing something genuinely unexpected can cut through in a way it never could before. The bar for attention is higher, but the reward for clearing it is bigger too.

DungeonLift is my small experiment in that direction.

Is it a serious piece of enterprise software marketing? Absolutely not. Is it more memorable than a PDF? I think so.

Play DungeonLift - and if you can beat the SQL Query boss, you’re probably ready for a proper demo of Neuralift.

GitHub repo