The blog post author argues Google has stopped being a product company and become an org chart with a search engine attached. Every major AI surface — Gemini, Vertex AI, AI Studio, NotebookLM, Antigravity, Project Astra — represents a different team's attempt to own the same underlying model, with parallel SKUs, quotas, SDKs, and auth systems that make it impossible to recommend a coherent path to developers.
The editorial frames the IBM analogy as overdue rather than novel — it's been applied to Microsoft, Oracle, and Cisco before. What makes it land harder on Google is the contrast with its original engineer-first culture (20% time, flat hierarchy), which has been gone for years; the difference now is that the deprecation cadence and fractured SKU surface make the decay measurable in the developer experience.
The synthesis points to the gemini-1.0-pro → 1.5 → 2.0 → 2.5 march happening in roughly 14 months as evidence that Google Cloud's one-year deprecation notice is honored only on paper. Model IDs change and behavior shifts faster than the published policy allows, eroding trust for anyone building production systems against the API.
A self-identified GCP employee summarized the thread by admitting they cannot confidently recommend which Google AI surface to use for any given workload. This is cited as the most damning evidence in the discussion: when the people building the platform can't disambiguate Gemini API vs. Vertex AI vs. AI Studio vs. Antigravity, external developers have no chance.
A blog post titled *Google is shattering under its own weight: the IBM-ification of Google* hit 169 points on Hacker News this week, and the comment section read like a group therapy session for ex-Googlers and frustrated API consumers. The author's thesis is blunt: Google has stopped being a product company and started being an org chart with a search engine attached. Every major Google AI surface — Gemini, Bard (RIP), Vertex AI, AI Studio, NotebookLM, Antigravity, Project Astra — is a different team's attempt to own the same underlying model.
The specific receipts are familiar to anyone who has shipped against Google's APIs in the last 24 months. Bard launched, got renamed to Gemini, which then forked into Gemini App, Gemini API, Gemini in Workspace, Gemini Code Assist, and Gemini in Search. Vertex AI exists as the "enterprise" version with a parallel SKU table, parallel quota system, and parallel SDK. AI Studio is the developer playground that uses different auth than Vertex. Antigravity, the new agentic IDE, sits adjacent to Code Assist with no clear positioning differentiator. The HN thread's top comment summarized it: "I work at GCP and I can't tell you with confidence which surface to recommend for any given workload."
The IBM analogy isn't new — it's been levied at Microsoft, Oracle, and Cisco at various points — but it lands harder on Google because Google built its brand on the opposite culture. The 20% time, the flat hierarchy, the engineer-first product calls. That Google is gone, and it's been gone for at least five years; what's new is that the consequences are now visible in the developer experience.
The technical decay shows up in three measurable places. First, deprecation cadence. Google Cloud's deprecation policy formally promises one year of notice for breaking changes, but the gemini-1.0-pro → gemini-1.5 → gemini-2.0 → gemini-2.5 march has happened in roughly 14 months, with model IDs changing, default parameters shifting, and rate limit tiers re-jiggered each time. Second, SDK proliferation. There is `google-generativeai`, `google-cloud-aiplatform`, `@google/generative-ai`, `@google-cloud/vertexai`, the new unified `@google/genai`, and a handful of language-specific community wrappers that Google occasionally blesses and occasionally forgets. Third, documentation entropy. Ask any developer who has Googled a Vertex error code in 2026: the top result is often a 2023 doc page that contradicts a 2025 doc page that contradicts the actual current SDK behavior.
The IBM playbook of the 1990s wasn't a failure of engineering — IBM had brilliant engineers — it was a failure of product consolidation. Mainframes, AS/400, RS/6000, OS/2, Lotus, Tivoli, Rational: each had a VP, each had a roadmap, each had a budget defended at the next reorg. Customers and developers had to become full-time amateur org-chart anthropologists to figure out which division actually owned the problem they were trying to solve. That is exactly what shipping against Google AI feels like in 2026.
The community reaction on HN was striking for its specificity. One commenter, claiming to be a current Googler, wrote: "The internal joke is that we have three different teams shipping the same agent framework and none of them know about the other two until launch week." Another pointed out that Antigravity's launch deck explicitly compared itself to Cursor and Windsurf — both products built by tiny teams that out-shipped Google Code Assist for two years running. The structural problem isn't that Google can't build good software; it's that Google can't decide what software it's building.
Compare this to Anthropic and OpenAI, which have roughly two product surfaces each — a chat app and an API — and treat them as the same product with different front ends. Even Microsoft, which has a sprawling org chart of its own, managed to consolidate its AI story into "Copilot" as a brand umbrella with surprisingly consistent semantics across surfaces. Google's failure to do the same isn't a technology gap. It's a political gap.
If you're building on Google's AI APIs in 2026, the practical guidance has three parts. One: pin everything. Pin model IDs, pin SDK versions, pin region endpoints. Do not write code that assumes `gemini-latest` will resolve to anything in particular six months from now. The auto-upgrade path has burned too many teams.
Two: pick one surface and commit. If you're a startup, use AI Studio and the `@google/genai` SDK — it's the closest thing to a stable API. If you're enterprise, eat the Vertex tax and use it for everything; don't mix surfaces. The cross-surface integration story is a graveyard of half-working IAM configurations and quota mismatches.
Three: hedge with a model abstraction layer. This was optional advice in 2023; it's table stakes in 2026. LiteLLM, Vercel AI SDK, or your own thin wrapper around the OpenAI-compatible chat completions interface. The probability that you'll want to swap Gemini for Claude or GPT for cost, latency, or capability reasons in the next 12 months is approximately 100%, and Google's API churn makes the wrapper cheaper than the alternative.
The more strategic implication: if your business model depends on a long-term partnership with a Google AI surface, get the partnership in writing with specific SKU commitments. The handshake-with-a-PM era is over. The PM you're talking to may not own that product in six months, and the product may not exist in twelve.
Google isn't IBM yet — IBM took roughly 15 years to fully ossify, and Google has the cash, the talent, and the data moat to course-correct if leadership wants to. The Sundar-era pattern of letting a thousand flowers bloom worked when Google was the default platform and developers had nowhere else to go. In 2026, with Anthropic shipping faster, OpenAI owning the chat surface, and open-weight models eating the cost-sensitive tier, the bloom-and-prune strategy is starting to look like just bloom. The next 18 months will tell us whether Google can do what IBM never quite did: consolidate before the org chart calcifies into the product.
Not the article's main point but I've never liked the "google killing products" complaints. People always talk about how big companies fail because they're unwilling to take risks and just recommit to their areas of strength, but this is what risk-taking looks like - you bla
Google is in an incredibly strong position. They're a top tier AI vendor, and in a world where content creation is largely commoditized and outsourced to AI, advertising companies will determine what gets seen, and what gets buried in the noise. They control both generation and visibility of wh
I am sure I am not alone in observing that starting around 2020... may be even a few years earlier... Google seemed to hire a lot of middle managers. I personally knew several managers who did not seem to contribute much beyond "organizing" meetings, and I saw them all join Google one by o
Recommending Hetzner as an alternative here is a mistake. It just exposes you to a different problem.There is a reason for the term "Hetznered" existing. Hetzner can suddenly and permanently terminates your account. They do this without warning or explanation. When it happens, you lose acc
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
> Look at the Railway GCP account ban situation. A literally billion dollar startup running on Google Cloud and Google just randomly snaps their fingers and deletes their account. Zero warning. No phone number to call. No account rep. Poof. Gone. It is actual insanity to me. A billion dollar cust