Argues Google is no longer a software company in any meaningful sense but a holding company for its search ad monopoly, with sprawling middle management that pretends to ship products. Points to GRAD/Perf incentives rewarding visible launches over durable products, 290+ entries on Killedbygoogle.com, and parallel teams (Bard/Duet/Gemini/Assistant; Hangouts/Allo/Duo/Meet/Chat) competing for the same surface as evidence of organizational rot.
Frames Google's decline as the same pattern IBM exhibited from 1990-1993: a cash-cow monopoly generates enough margin to mask rot in surrounding businesses until collapse forces restructuring. Argues every company that builds a true monopoly eventually optimizes for protecting the monopoly rather than for the innovation that created it, and that the optimization is invisible from inside.
A blog post titled *The IBM-ification of Google* hit Hacker News this week and pulled 137 points and a comment thread that reads like a 20-year retrospective. The author's thesis is blunt: Google is no longer a software company in any meaningful sense — it is a holding company for a search ad monopoly, with a sprawling middle-management layer that pretends to ship products.
The specific charges are familiar to anyone who has watched the company for a decade. Performance review cycles (Perf, then GRAD) reward visible launches over durable products, which is why Killedbygoogle.com now lists more than 290 shuttered services. Parallel teams compete for the same surface area — Bard, Duet, Gemini, and Assistant all existed simultaneously as AI front-ends before being half-merged. Hangouts, Allo, Duo, Meet, and Chat fought for the messaging slot for nearly a decade. The Gemini app and the Gemini API ship features on different timelines. Engineers describe quarters spent on launch readiness reviews, privacy reviews, and design reviews for products that get cancelled before GA.
The IBM analogy is doing real work here. IBM in 1990 was the most profitable company in the world; by 1993 it had posted the largest corporate loss in US history at the time ($8 billion) and was forced to break itself into business units to survive. The pattern was the same: a cash-cow monopoly (mainframes / search ads) generated enough margin that nobody had to confront the slow rot in the surrounding businesses, until they did.
The easy read is schadenfreude — a hyperscaler gets a little smaller, a few thousand PMs eat humble pie. The more interesting read is what this tells us about how software organizations age, because Google is not unique. Every company that builds a true monopoly eventually optimizes for protecting the monopoly rather than for the thing that created it, and the optimization is invisible from inside.
Look at what Google actually shipped versus what it invented. The transformer architecture came out of Google Brain in 2017. The original GPT was OpenAI's reimplementation of Google's own paper. By the time ChatGPT launched in November 2022, Google had LaMDA running internally for at least 18 months — and the company that invented the entire field of modern LLMs spent 2023 visibly catching up to a five-year-old startup. Gemini 1.0's launch demo was later revealed to have been heavily edited, which is the kind of mistake a company makes when the PR layer is no longer in sync with what engineering can actually deliver.
The HN thread surfaced a more damning data point: ex-Googlers describe a culture where shipping is risky and not shipping is safe. A launched product can fail, get you a bad perf rating, and stall your promo packet. A product that never launches has no failure mode — and the engineer who spent two years on the internal infrastructure underneath it still gets credit for "foundational work." This is exactly the incentive gradient that turned IBM's research division — which won five Nobel Prizes and invented the relational database, the hard drive, and DRAM — into a place where nothing reached customers for fifteen years.
The community reaction split along predictable lines. Current and recent Googlers defended the company's engineering depth, pointed at TPU v5, Borg, Spanner, and the fact that Search still works at a scale nobody else has approached. Outside observers pointed at the gap between what Google could ship and what it does ship. Both sides are right, which is the whole problem: a company can have the best engineers on earth and still be structurally incapable of getting their work to a user.
The Sundar Pichai era — now ten years long — has presided over the conversion. Pichai inherited a company that was still recognizably the one Larry and Sergey built. He is handing off a company that looks like Microsoft circa 2010: huge, profitable, organizationally exhausted, and slowly losing the next platform shift. Microsoft, of course, eventually clawed its way back under Nadella by killing internal politics and betting on cloud. Google's equivalent move would be killing the stack-ranking review system and consolidating the AI org under a single technical leader. Neither has happened.
If you build anything that depends on Google as a platform, the IBM-ification thesis has direct operational consequences.
Treat any non-core Google API as having a 5-year half-life. Core means Search, Ads, YouTube, Gmail, Maps, and the GCP services that have direct revenue tied to them (Compute, Storage, BigQuery, Spanner). Everything else is at risk. The Killedbygoogle list contains products that had millions of paying users — Domains, Stadia, Inbox, Reader, Hangouts, the original Cloud IoT Core. The mean time from "strategic priority" to "sunset announcement" is shorter than your customer contract lifetimes.
Budget migration cost into your architecture decisions. If you pick Firebase Auth today, you should know what it would cost to move to Auth0 or Supabase in three years. If you pick Vertex AI today, you should have the same Bedrock or direct-Anthropic fallback that you'd have if you picked any other vendor. The right framing is not "will Google kill this" — it's "what is my exit cost if they do."
Take Gemini's API contracts seriously, but assume Anthropic and OpenAI are both safer long-term bets for production LLM workloads. Not because Gemini's models are worse — Gemini 2.0 Pro is genuinely competitive on benchmarks — but because the organization shipping them has a worse track record on API stability, sunset notices, and developer relations follow-through. Cloud Vision API has changed pricing models twice in the last three years. Workspace API quotas have been silently re-tiered. The Anthropic and OpenAI orgs are smaller, more focused, and have explicit deprecation policies; Google does not.
Google is not going to fail in any timeframe a developer needs to plan for. The Search monopoly throws off enough cash to fund a decade of strategic confusion, and the TPU and infrastructure assets are genuinely world-class. But the relevant comparison is not whether Google survives — IBM survived too, and is now a $200B enterprise services company that no developer thinks about — it's whether Google remains a place where the next platform shift gets invented. On current evidence, it isn't, and the next platform shift will be built by someone else on infrastructure that someone else owns. The developers who notice this early will spend the next five years building portable architectures and treating every Google dependency as a managed risk. The ones who don't will be writing migration RFCs in 2029.
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.