The 3am Prompt: When Your Brain Won't Stop Building

By Ryan Vanshur

Also published on the Guild Letter: Read the original issue.

Subscribe now It’s 3am and I’m building a deal scoring system. Not at my desk. In my sleep. The architecture is elegant in the way things only are when your conscious mind isn’t interfering. Signal weights flowing into risk tiers. Competitive intel feeding a coaching layer that knows which rep needs what before which call. Every node in the knowledge graph is connected perfectly and the whole thing hums. I can see it. All of it. At once. And some part of my brain knows this is temporary. So I’m doing something absurd: I’m trying to create a markdown file. In my dream. Trying to write the solution down before the window closes, trying to commit it to something durable. As if my sleeping brain has internalized that workflow so deeply that “write it to a file” is now its instinct for preservation. Then I wake up. And it’s gone. I know I had something. I can feel the architecture the way you feel a word stuck on the tip of your tongue, except it’s an entire system. I reach for the specifics and get sand through my fingers. The signal weighting logic. The data flow between layers. Whatever the thing was that made it all click. Gone. Sometimes I try to go back in. Close my eyes. Hold the thread. Slip back into the dream like maybe I can re-enter the session and finish the build. It never works. The context window has collapsed and there’s no conversation history to resume from. Here’s the thing nobody tells you about building with AI every day: your brain doesn’t stop when you close the laptop. I’ve Never Had This Experience Before I want to be clear about something. I’m not an engineer. I’ve never written production code in my life. I didn’t study computer science. My background is sales, revenue operations, customer success. I’ve spent a decade in vertical SaaS building GTM motions for industries that most people couldn’t pick out of a lineup. Vocational education. Construction fintech. I’ve worked hard my entire career. Long hours. High stakes. I’ve laid awake thinking about deals, pipeline gaps, team problems, quarterly targets. Normal operator stuff. But I have never, in 15 years of go-to-market work, dreamed about building systems . Never woke up at 3am having architected something in my sleep. Never experienced that state where your brain is actively constructing a solution with the same flow and intensity you have during your best working hours. This started happening about eight months into building with Claude Code daily. And the dreams aren’t vague. They’re specific. I’m designing knowledge graph architectures. Structuring prompt chains. Building automation workflows that connect CRM signals to coaching outputs. Creating systems I’ve never been trained to create by any traditional definition. My brain has apparently decided that AI-assisted system building is important enough to rehearse in my sleep. And I don’t have a computer science degree to explain why. Developers Have Known About This for Decades It turns out this has deep roots in the engineering community. Programmers have been dreaming in code since the profession existed. Hacker News threads from 2017 and 2022 are full of developers describing exactly what I’m experiencing. Dreaming elegant solutions that evaporate on waking. Getting stuck in debugging loops all night. Waking up, implementing the dream logic, and finding actual bugs their sleeping brain caught. One developer captured the absurdity perfectly…“Sometimes I get up in the middle of the night and write them. Sometimes I get halfway through and realize they rely on a library called ‘BeeswaxAlienContention’ that only existed in my dream.” Harvard researcher Robert Stickgold named this the Tetris Effect in 2000. He proved that intensely repetitive cognitive activity replays during sleep onset. Not as a choice. As a biological default. Your brain takes the dominant pattern of your waking hours and runs it through consolidation circuits while you sleep. Even amnesic patients who couldn’t remember playing Tetris still saw the blocks falling. The science is settled on this. What’s new is who it’s happening to. When the Tetris Effect Crosses the Aisle Developers dream about code because they write code 8 hours a day. The repetitive loop (write, test, debug, iterate) gets replayed during sleep while the brain locks it into long-term memory. That tracks. But I’m not writing code. I’m not a developer. What I’m doing 6 to 8 hours a day is something different…I’m having conversations with an AI about go-to-market problems, and those conversations are producing real systems. Pipeline intelligence dashboards. Account research workflows. Signal detection architectures. Deal coaching frameworks. Knowledge bases with hundreds of nodes. The loop is pretty much…I spot a GTM problem, figure out how to ask it, prompt the AI, evaluate the output, refine, iterate, build. That’s my Tetris. That’s the pattern my brain decided is worth rehearsing at 3am. And here’s what makes AI different from any tool I’ve used before in GTM work. When I used Salesforce all day, I never dreamed about Salesforce. When I lived in HubSpot for three years at CourseKey, I never woke up designing HubSpot workflows in my sleep. I used those tools. They didn’t change how my brain operates. AI did something those tools never could. It turned me into a builder. Not a builder in the “learned to code” sense. A builder in the “my brain now thinks in systems because the gap between having an idea and creating the thing has collapsed to near zero” sense. When the implementation cost of an idea drops to zero (just describe it, the AI builds it), your brain starts generating implementations the way it used to generate slide deck ideas. Constantly. Including at 3am. What I’m Actually Dreaming About The dreams aren’t abstract. They’re working sessions without the screen. Last month, I woke up having designed a competitive intelligence layer that would detect when a prospect’s current vendor was experiencing service failures, then automatically surface that signal in the rep’s pre-call briefing. The whole architecture was there: the data source, the trigger logic, the delivery mechanism, the coaching prompt that would tell the rep exactly how to use the intel without sounding like they were ambulance-chasing. Was it a perfect system? I’ll never know. I couldn’t capture it fast enough. But pieces of it made it into what I built the next morning. The signal detection concept was sound even if the dream implementation relied on APIs that don’t exist. Another night, I spent what felt like hours building a knowledge graph that connected every customer call we’d recorded in the last 6 months to the specific deal outcomes. Not just “this call happened before this deal closed.” The connections were semantic. Why the deal moved. What phrase from the champion triggered the next stage. Which objections predicted stalls. The sleeping version of my brain had apparently decided that our call intelligence wasn’t being used deeply enough and started engineering the solution on its own time. I’ve dreamed about content systems too. Article architectures that branch into lead magnets that feed email sequences that trigger based on which section the reader spent the most time on. The whole flywheel, visible at once, all the connections alive and operational. The kind of systems view you can only hold for about 4 seconds when you’re awake before the complexity collapses back into sequential thinking. The Operator’s Tetris Effect Here’s what I think is actually happening, and why it matters for anyone building AI-native GTM: Your brain has reclassified you. I spent 15 years as an operator. Someone who runs systems that other people build. In the last 18 months of daily AI collaboration, my brain has apparently decided I’m now a builder. The identity shift happened below conscious awareness. The dreams are the evidence. You don’t dream about building systems unless your brain believes you’re the kind of person who builds systems. The “flow state leak” is real. There’s a state during AI building sessions where everything clicks. The problem is clear, the prompts are precise, the outputs are exactly right, and you’re iterating at a speed that feels like thought itself. That state is the most cognitively intense experience in my professional life. More intense than closing a six-figure deal. More intense than running a board presentation. My brain treats it as high-value and replays it during sleep. The research says this kind of replay is 10x more effective for skill building than consciously thinking about the problem while awake. The capture problem is solvable, partially. I keep my phone with my voice note taker close to my bed now. Not for full architectures (dream logic doesn’t survive compilation). But for connections. The associations my sleeping brain makes between ideas that seemed unrelated during the day. Last week I recorded a short note that “pulse signals should feed the content engine, not just the pipeline dashboard.” That connection was obvious once I said it out loud at 6am. But I’d never made it during working hours because I was too deep in each system to see across them. The sleeping brain doesn’t have that problem. It sees everything at the same altitude. The open loop problem is the real enemy. I’ve noticed the dreaming intensifies on nights when I stop a build mid-conversation. The Zeigarnik Effect is when unfinished tasks create cognitive tension that persists until discharged. AI conversations are natural open loops. Every response invites a follow-up. Research shows that writing down your next steps before stopping actually releases the tension. Your brain needs to believe the thread won’t be lost. I’ve started ending every session with a brief “where I left off” note. The 3am builds happen less when the day’s work has a clear bookmark. The Adoption Curve Nobody’s Measuring Every AI vendor on the planet is tracking the same metrics. DAUs. Sessions per week. Features activated. Prompts per user. They build dashboards around these numbers, run QBRs off them, pitch investors with them. And every single one of those metrics measures the wrong thing. They measure whether someone opened the tool. They don’t measure whether the tool changed how someone thinks. I know exactly when I crossed the line. It wasn’t the first time I used Claude Code. It wasn’t even the first time I built something useful with it. It was the first time I woke up at 3am having designed a system in my sleep that I’d never been trained to design while awake. That was the moment AI stopped being software I use and became a cognitive pattern my brain runs on its own. No dashboard captured that. No product analytics team will ever see it. But it’s the most meaningful adoption event in my professional life. Think about it from a GTM lens. We obsess over adoption metrics because they proxy for value. If someone logs in 5 times a week, they’re getting value, they’ll renew, they’ll expand. But what if the real signal for deep adoption isn’t usage frequency? What if it’s what happens when the user closes the laptop? A rep who uses your AI tool before every call is adopted. A rep who starts thinking differently about deals when they’re driving home, who starts seeing competitive signals in conversations they used to ignore, who wakes up with a pipeline strategy they didn’t consciously plan: that rep hasn’t just adopted your tool. They’ve internalized your system. That’s a fundamentally different level of lock-in, and nobody is measuring it because it doesn’t fit in a product analytics event. Here’s what I think the real adoption curve looks like for AI in GTM: Stage 1: Task Substitution. You use AI to do things you already did, just faster. Meeting prep. Email drafts. Account research. This is where 90% of teams are. The metric: time saved. Stage 2: Capability Expansion. You use AI to do things you couldn’t do before. Build dashboards. Create knowledge systems. Design coaching frameworks. You’re not just faster. You’re different. The metric: new outputs that didn’t exist before. Stage 3: Cognitive Integration. Your brain starts running the AI collaboration pattern without the tool present. You think in systems. You see connections across data sources in the shower. You architect solutions in your sleep. The metric: there isn’t one. No vendor can track this. But it’s where the compounding starts. Most adoption conversations stop at Stage 1. “We saved reps 45 minutes of prep time per week.” Fine. But the rep whose brain rewired itself to think in deal architectures at 3am isn’t saving 45 minutes. They’re operating on a different plane. And they’re never going back. I know what some of you are thinking. “Ryan, this just means you work too much and need better boundaries.” Maybe. But consider this…MIT proved that dreaming about a problem produces measurably more creative solutions. 43% more creative than undirected sleep. 78% more than staying awake and grinding on it. Northwestern doubled problem-solving rates by cueing dreams about specific puzzles. The 3am builds might be the most productive work I do. I just can’t capture it yet. Here’s the uncomfortable part for vendors…Stage 3 adoption has nothing to do with your product’s UX, your onboarding flow, or your feature releases. It happens when the underlying thinking pattern your tool enables becomes important enough for the user’s brain to practice on its own time. You can’t engineer that. You can only build something worth dreaming about. And here’s the uncomfortable part for operators…once your brain crosses into Stage 3, you can’t undo it. You don’t go back to thinking in slide decks after you’ve spent six months thinking in systems. The same way a chess player can’t stop seeing board positions or a musician can’t stop hearing chord progressions, you won’t stop architecting. The AI gave your brain permission to build, and your brain took the keys and started driving at night. Within the next decade, we’ll look back at this moment the same way early programmers looked at the first time they dreamed in code. It was the sign that a new mode of professional thinking had taken root. Not in the people trained for it. In the operators, the revenue leaders, the GTM professionals who never expected to think in systems and now can’t stop. The AI didn’t teach me to code. It taught my brain to architect. And now my brain does it on its own schedule, including the night shift. The Vertical GTM Guild is where operators who build with AI share what’s actually working. Not prompting tutorials. Not demo videos. Real systems running on real pipeline. If your brain is already building at 3am, the least you can do is give it better blueprints during the day. Subscribe now