[Erlich Bachman's Incubator, Palo Alto — March 2014, Afternoon]
Monica's call had been brief and specific. Partner meeting scheduled for Thursday. Ethan had two days to prepare. In the meantime, she'd positioned the pitch as competitive intelligence — "a technology we should be tracking" — which gave her cover to bring an outsider into the room without needing prior partnership approval.
Two days. Forty-eight hours with nothing to do except refine a demo he'd already refined into its best possible state, and try not to calculate how many meals he could afford with the $217 remaining across two bank accounts that were both approaching terminal velocity.
Sarah had suggested he get out of the apartment. "You've been staring at the same four walls for a week. You're going to optimize the demo into dust. Go somewhere. Talk to a human being who isn't me."
So he'd driven to Palo Alto. Not to the Raviga offices — that would be desperate, and Monica had specifically said "don't come here before Thursday." Instead, he'd navigated to 5230 Newell Road. Erlich Bachman's incubator. The Hacker Hostel. The place where Pied Piper lived.
The house looked exactly like what it was: a residential property that someone with more confidence than taste had converted into a startup workspace. A wraparound porch held mismatched chairs and a table covered in empty energy drink cans. Through the front window, Ethan could see the blue glow of multiple monitors. A whiteboard was visible against the far wall, covered in diagrams he couldn't make out from the driveway.
He'd found the address through a simple search — Erlich's incubator was listed on AngelList as a "pre-seed innovation accelerator," which was a generous description of a house where people coded in exchange for giving away ten percent of their companies. The listing included open office hours on Tuesdays and Thursdays, where aspiring founders could tour the space and "explore collaboration opportunities with the Aviato founder."
It was Tuesday. Ethan walked up the porch steps and knocked.
The door opened to a wall of warm air, stale pizza, and the particular musk of multiple engineers who'd been coding in close proximity without adequate ventilation. A thin man with glasses and a nervous posture stood in the doorway — not Erlich. Too young, too anxious, and radiating the specific kind of discomfort that came from being assigned door duty at a social event you'd rather be coding through.
"Hi. I'm here for the open hours?"
"Oh. Yeah. Come in. Erlich's in the kitchen. I think." The man — Ethan's Talent Resonance pinged a five, solid but unremarkable — stepped aside.
The interior was organized chaos. Four desks pushed together in the living room, each buried under monitors and cables. A couch that had seen better decades held three people typing on laptops. The kitchen was visible through an archway, where a large man was holding forth to an audience of two with the theatrical energy of a stage actor who'd been denied a proper venue.
Erlich Bachman. Unmistakable. The blazer from TechCrunch Disrupt was gone, replaced by a graphic tee and an open flannel, but the voice carried the same grandiose certainty that Ethan remembered from the show. He was describing something — a pivot strategy, a market opportunity, the correct way to negotiate with Sand Hill Road — and his audience was nodding with the glazed expressions of people who'd heard this speech before.
Talent Resonance: three. Same as Disrupt. The number hadn't changed because the man hadn't changed. Erlich Bachman's gifts were charisma, confidence, and the unshakeable belief that proximity to success was the same as creating it. None of those things registered on a scale that measured technical capability.
Ethan nodded politely at Erlich, declined the tour, and drifted toward the living room. He was here for one person.
Richard Hendricks sat at the far desk, the one pushed against the wall, partially hidden behind a monitor that was too large for the space. He wore a hoodie — not the same one from Disrupt, but close enough to be a uniform. His posture was the concentrated hunch of someone who'd forgotten other people existed. Code filled his screen in dense, neatly formatted blocks.
Ethan approached. "Richard Hendricks?"
Richard startled. His knee hit the underside of the desk, rattling the monitor. Coffee sloshed in a mug that sat too close to the keyboard.
"Uh. Yes. Hi. Do I—"
"Ethan Gardner. I was at Disrupt. Your compression demo was the best thing in the room."
Richard's expression cycled through confusion, suspicion, and a reluctant flicker of pride. "Oh. Thanks. That's... thanks. Are you a founder, or—"
"Founder. AI. Different space from yours, but I wanted to ask you something technical if you've got a minute."
Richard glanced at his screen. The code was calling to him — Ethan recognized the gravitational pull, the resistance to conversation that every deep-focus engineer experienced when interrupted mid-thought. But curiosity won. Richard minimized his editor and turned his chair.
"What kind of AI?"
"Language generation. Neural networks that produce text."
"Like... chat bots?"
"More advanced. The architecture processes entire documents simultaneously and generates original content. But the models are large. Hundreds of millions of parameters. Training takes thousands of GPU-hours. And the resulting weights are—"
"Heavy." Richard's eyes changed. The social awkwardness receded, replaced by the sharp, focused attention of a mind engaging with a problem it found interesting. "How heavy?"
"Three hundred megabytes for the base model. Bigger models will be gigabytes. Moving those weights around, storing them, deploying them — the data overhead is enormous."
"And you're thinking about compression."
"I'm thinking about compression."
Richard leaned forward. The hoodie bunched at his shoulders. "Nobody's asked about that application before. Compression gets pitched as storage, bandwidth, media files. But applying it to model weights — that's a different optimization problem. The weights aren't arbitrary data. They have structure. Statistical distributions. You could design a compression scheme that exploits the distribution patterns instead of treating them as generic bytes."
Ethan's Talent Resonance held steady at nine. The same reading as Disrupt, but richer now — informed by proximity, by watching the way Richard's mind moved from the specific question to its general implications in three sentences. Richard wasn't just a compression genius. He was a systems thinker who saw connections between domains that other people treated as separate.
"Exactly," Ethan said. "Middle-out wouldn't just reduce file size. It would reduce inference latency. If the weights load faster, the model generates faster. The user experience goes from waiting ten seconds for output to waiting two."
"What's your— what architecture are you using? For the language model?"
The question required the same careful navigation he'd practiced with every conversation about his work. "It's novel. Proprietary. Uses an attention mechanism that lets the model weigh the relevance of every input token simultaneously."
Richard's brow furrowed. "Simultaneously? Not sequentially?"
"Parallel processing of the entire input sequence. Every position attends to every other position."
"That's... computationally expensive."
"Quadratic with sequence length."
"So you need compression and efficient inference." Richard pulled a pen from behind his ear — the same place Sarah kept hers — and started sketching on a napkin. "If you quantized the attention weights post-training, you could reduce the memory footprint by half without significant quality loss. And if you applied middle-out to the quantized weights..."
He trailed off, lost in the math, the pen moving faster than the napkin could absorb. Ethan watched. The two most technically capable people he'd encountered in this timeline — Richard Hendricks and Sarah Chen — shared the same trait: they forgot about the conversation when the problem got interesting enough.
"Here." Richard tore off the napkin and handed it to Ethan. "It's rough, but the principle is sound. Compression ratios on structured weight matrices should be significantly higher than on arbitrary data. It's— actually, this is really interesting. Can I think about this?"
"Take your time."
Richard pulled a business card from a desk drawer — PIED PIPER, RICHARD HENDRICKS, CTO — and handed it over. Ethan reciprocated with one of the Gardner Analytics cards from the dead man's box. The cards were outdated — "Data-Driven Solutions for the Modern Enterprise" — but the contact info was still valid.
"What architecture did you say?" Richard asked, already turning back to his monitor. "The attention thing?"
"I'll send you a paper when it's published."
"Cool. Yeah. Send it."
Ethan left him to his code. On the way out, he passed through the kitchen, where Erlich had shifted from market strategy to a monologue about the time he'd pitched Aviato to Peter Thiel. The audience had grown by one — a woman with headphones around her neck who appeared to be there for the free pizza rather than the wisdom.
Erlich spotted Ethan. "Hey. You. New face. What are you building?"
"AI that generates text."
Erlich's expression was the same mix of condescension and curiosity that Silicon Valley's self-appointed gatekeepers wore as a default. "Like Siri?"
Ethan almost laughed. The bartender from his first rejection. Alan Rao. And now Erlich Bachman. The question was a running tally at this point — a scoreboard for how poorly the world understood what he was building.
"Something like that," he said, and walked out.
The drive back to San Francisco took forty minutes. The Honda Civic's gas gauge hovered near empty. The Red Bull Richard had pressed into his hand before he left — the universal engineer currency, offered without ceremony from a mini-fridge beside the desk — sat in the cupholder, warming in the afternoon sun. Ethan cracked it open at a stoplight and drank. The sweetness was overwhelming after weeks of black coffee and water.
His phone buzzed. Monica.
Partner meeting confirmed. Thursday, 2 PM. Raviga offices. Bring everything you've got.
Author's Note / Promotion: Your Reviews and Power Stones are the best way to show support. They help me know what you're enjoying and bring in new readers! You don't have to. Get instant access to more content by supporting me on Patreon. I have three options so you can pick how far ahead you want to be: 🪙 Silver Tier ($6): Read 10 chapters ahead of the public site. 👑 Gold Tier ($9): Get 15-20 chapters ahead of the public site. 💎 Platinum Tier ($15): The ultimate experience. Get new chapters the second I finish them . No waiting for weekly drops, just pure, instant access. Your support helps me write more . 👉 Find it all at patreon.com/fanficwriter1