Google Wasn't Caught Sleeping. It Was Wide Awake and Chose Not to Ship.

On February 8, 2023, Bard answered a question about the James Webb Space Telescope and got it wrong - it claimed the telescope had taken the very first pictures of an exoplanet, a feat actually achieved by a ground-based observatory back in 2004.³ The error sat in a promotional demo for all the world to read. By the close, Alphabet had shed about 7.7% of its value - roughly $100 billion - on a single bad sentence from a chatbot.³ It looked like the moment a sleeping giant was caught mid-yawn by a faster rival.

The official story goes like this: ChatGPT appeared, Google panicked, declared a 'code red,' and rushed a half-baked competitor out the door. Almost every beat of that is misleading. Google didn't lack the technology - it had conversational AI in production for years. What it lacked was the willingness to ship it. The scramble wasn't an engineering emergency. It was a strategy reversal.

The 'code red' was one sentence in one newspaper

Start with the phrase everyone repeats. The 'code red' - Sundar Pichai supposedly reassigning teams across research and Trust and Safety to crank out AI prototypes - traces to a single New York Times article in December 2022, sourced entirely to unnamed people inside the company.¹ Google has never confirmed it. That doesn't make it false; internal alarm was surely real. But it means the most dramatized fact in the whole narrative is an attributed-to-source claim, not a documented edict - and the difference matters, because the 'code red' framing does heavy lifting. It paints Google as reactive, surprised, behind. The story needs Google to have been asleep, because a giant that chose to wait is a far less satisfying villain than a giant that got beaten.

Here is the harder fact the ambush story has to ignore. Days before Bard's public debut, Google's own AI chief told employees the company carried 'much more reputational risk' in serving wrong answers, and was therefore moving 'more conservatively than a small startup.'⁴ Read that again. That is not the voice of a company that lacks the engine. It is the voice of a company that built the engine, looked at it, and decided the cost of a hallucination at Google's scale was not worth the launch. The capability was sitting there. The brake was the strategy.

The $100 billion sentence - and what it really proved

The exoplanet blunder is told as a clean morality tale: chatbot lies, market punishes, lesson learned. The truth is messier. The 7.7% drop came the day after Microsoft staged its own Bing-and-ChatGPT event, and the sell-off carried a broad anxiety about Google's whole position, not just one bad fact.³ Pinning the entire $100 billion on a single sentence is the kind of tidy attribution the market never actually makes. But the error did prove something - just not the thing people think. It didn't prove Google's AI was bad. It proved Jeff Dean had been right. The exact failure mode he warned about - Google saying something confidently false, at scale, on the record - happened on day one, in the company's own marketing. The brake existed for a reason. Releasing under pressure removed it, and the warning came true within hours.

Bard, PaLM, Gemini: the rename that fooled the timeline

Once you stop reading this as a panic and start reading it as a forced sequence of compromises, the messy product chronology snaps into focus. Bard launched in February 2023 on a 'lightweight model version of LaMDA' - not the more powerful PaLM 2 it's often credited with, which only arrived in May.²⁵ Then came the part almost everyone collapses into a single moment. Gemini the model launched December 6, 2023, with Gemini Pro slipped into Bard that same day.⁶ But the chatbot wasn't renamed 'Gemini' until February 8, 2024 - two months later, alongside the top-tier Ultra model.⁶⁷ The model launch and the brand launch were separate events, and conflating them makes the recovery look smoother and faster than it was.

But didn't Google still lose - rushing and stumbling?

The fair objection is that none of this gets Google off the hook. A company that has the technology and won't ship it has still failed - arguably failed worse, because it owned the future and handed it to a rival anyway. And the rush left fingerprints: the splashy Gemini launch video was criticized for being edited to look like a real-time, fluid interaction when it wasn't, and internal messages leaked to Bloomberg reportedly called Bard 'worse than useless' and a 'pathological liar.'⁸ That's a damning picture, and it's true. But it sharpens the thesis rather than refuting it. The failure was never an absence of capability - it was a failure of nerve, then a failure of judgment when the nerve broke. Google over-corrected. Having held the brake too long, it slammed the accelerator and shipped demos it knew weren't ready. Both mistakes share one root: Google never decided, calmly and in advance, how it wanted to weigh its own trust against its own speed.

The 'scramble' is a comforting story because it lets everyone off easy: Google was just surprised, ChatGPT was just faster, the giant just needed a jolt. The duller, truer story is that Google was awake the whole time, holding a working product behind a self-imposed line about reputational risk - and then, when a rival made waiting look like weakness, it crossed that line and proved, on day one, exactly why the line was there. The exoplanet error wasn't the moment Google fell behind. It was the moment Google's caution was vindicated and abandoned in the same breath. The hardest decisions aren't about what you can build. They're about what you'll allow yourself to ship - and whether you'll make that call before someone else makes it for you.