[Gardner Analytics Office — May 2014, 10:00 AM, Launch Day]
The livestream loaded at 10:02, two minutes late, which Sarah attributed to "server load from people who want to watch a car crash" and Marcus attributed to "Hooli's infrastructure team probably being as overpromised as their AI."
They'd arranged the office for viewing: Marcus's monitor turned to face the center of the room, three chairs pulled into a semicircle, coffee from Blue Bottle in paper cups balanced on desk edges. The stream showed the Hooli developer conference main stage — a cavernous auditorium with the production values of a stadium concert. LED screens. Atmospheric lighting. Two thousand developers in matching Hooli t-shirts that had been distributed at registration.
Gavin Belson took the stage at 10:07. The audience applauded. The LEDs pulsed.
"We are at a precipice," Gavin began. His voice was the same instrument Ethan had heard at TechCrunch Disrupt four months ago — modulated, rehearsed, every pause calibrated for maximum gravitas. "The precipice of intelligence itself. Today, Hooli doesn't just announce a product. Today, Hooli announces a partner. An intelligence that works alongside you. That understands you. That thinks with you."
"Oh no," Sarah said.
"HooliBot."
The name appeared on the LED screens in a font that must have cost a design team three weeks: rounded, friendly, the letters spaced to suggest approachability. Below it, the tagline: Your Intelligent Assistant.
Gavin ran the demo. A staged conversation between a Hooli engineer and HooliBot, projected on the main screen. The engineer asked HooliBot about the weather. HooliBot responded with the current temperature in Mountain View and a suggestion to "bring a light jacket." The audience clapped.
The engineer asked HooliBot to summarize a document. HooliBot produced three bullet points that were clearly pre-loaded — the response appeared instantly, with none of the token-by-token generation lag that real inference would produce. The audience clapped again, slightly less enthusiastically.
Then the engineer went off-script. "HooliBot, what's the meaning of life?"
HooliBot responded: "The meaning of life is to make the world a better place through synergy and innovation."
The audience laughed. Gavin laughed. The laughter was supposed to be warm — a shared joke about the product's earnest literalness. But underneath it, Ethan heard the particular frequency of an audience beginning to suspect they were watching a magic trick instead of magic.
"This isn't neural," Sarah said. She was leaning forward, studying the screen with the diagnostic intensity she applied to code review. "The response latency is wrong. Real inference at that quality level takes seconds. These responses are appearing in under half a second. It's retrieval-based. Template matching."
"Pattern matching from the 1980s," Ethan agreed. "With a conversational wrapper."
"They spent how much on this?"
"Hooli's AI division budget is probably in the hundreds of millions."
"And they built a chatbot."
Ethan picked up his phone. Opened Twitter. His account — @ethangardnerai, created three weeks ago at Monica's suggestion for "founder visibility" — had 340 followers, mostly VCs and tech journalists who'd followed him after the TechCrunch article.
He started typing.
Watching the HooliBot demo. Impressed by the production values. Less impressed by the technology. This appears to be retrieval-based pattern matching, not neural language understanding. The response latency is inconsistent with real-time inference. #HooliBot
Send.
A second tweet, thirty seconds later.
For comparison: attention-based neural architectures process input through learned representations, not pre-scripted templates. HooliBot's responses show no evidence of generative capability. This is 1990s NLP with 2014 marketing. #HooliBot
Send.
A third.
"Making the world a better place through synergy and innovation" is not a machine learning output. It's a string literal in a response database. Real AI doesn't speak in corporate buzzwords. It generates novel text based on learned patterns. #HooliBot
Sarah watched him type. Her expression cycled through phases: surprise, assessment, and the particular half-smile she reserved for decisions that were either brilliant or catastrophic.
"This is either brilliant or stupid," she said.
"Possibly both." He kept typing.
The thread grew. Four tweets. Five. Six. Each one a technical correction, phrased in accessible language, systematically dismantling HooliBot's claims by comparing them to what a real neural architecture could do. He didn't name Gardner Analytics. Didn't link to the company. Didn't self-promote. Just delivered a clinic in what AI was and what it wasn't, using Hooli's own demo as the anatomy lesson.
The thread went viral within the hour. Not Marcus-Webb-viral — a different kind of spread, driven by engineers and researchers who recognized the technical accuracy of Ethan's observations and amplified them with their own commentary. Hacker News picked it up. A Google Brain researcher retweeted with the comment "This thread is the most accurate assessment of HooliBot I've seen." A Stanford professor quote-tweeted: "When your live demo has lower latency than inference, you're not running inference."
By noon, the thread had been viewed four hundred thousand times.
Marcus pulled up Hooli's stock chart on his monitor. Down 0.3% since the morning — a modest dip that, for a company Hooli's size, represented hundreds of millions in market cap. The dip wasn't caused by Ethan's thread alone — the broader market reaction to HooliBot was tepid, and several tech analysts had published lukewarm assessments. But the thread was being cited in those assessments. Ethan's name was appearing in tech journalism as a source of credible AI criticism.
Which meant Gavin Belson's team was reading it.
---
[Hooli Headquarters — Same Afternoon]
[GAVIN BELSON]
The smoothie had gone warm. Gavin set it aside and stared at the tablet, where a social media dashboard showed the real-time response to HooliBot's launch. The metrics were not good. Positive sentiment: 31%. Negative: 42%. The rest was neutral commentary that read, to Gavin, as people too polite to say what they thought.
One thread dominated the negative category. Six tweets. Technical. Specific. Ruthlessly accurate.
@ethangardnerai.
"Who is this person?" Gavin asked. His voice was quiet — the particular quiet that his staff had learned to associate with decisions that were about to become expensive for someone.
Vincent Chen stood across the desk. "Ethan Gardner. The AI startup I flagged last month. Raviga-funded. Three-person team."
"Three people."
"Three."
"Three people just humiliated a company with four thousand engineers in front of the entire technology industry."
"His criticism is technically accurate—"
"I don't care if it's accurate. I care that it's visible." Gavin picked up a red pen from the desk caddy — the same Montblanc he used for signing acquisition offers and marking names on org charts. He circled "Ethan Gardner" on the printout of the Twitter thread. Pressed hard enough that the pen's tip scored the paper.
"I want to know everything about Gardner Analytics. Revenue, clients, cloud infrastructure, team, investors, everything. And I want options."
"What kind of options?"
"The kind that make problems go away."
Vincent left. Gavin set the pen down. Through his office window, the Hooli campus spread green and ordered beneath the afternoon sun — the volleyball courts, the commissary, the meditation garden where he occasionally performed mindfulness exercises that his therapist had suggested and his ego had repurposed as photo opportunities.
Three people. A sandwich shop office. A Raviga seed check.
The pen had left a red circle around the name. Gavin picked up the printout and placed it in the top drawer of his desk, where it joined a short list of other circled names — founders, competitors, journalists — who had made the mistake of embarrassing him in public.
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