Argues that Anthropic's accumulated pattern of unannounced rate limit tightening, mid-cycle Claude Code cap recalibrations, short-runway model deprecations, and template support replies signals a company that has stopped treating developers as partners. The individual changes are unremarkable industry behavior, but the aggregate erodes the trust Anthropic explicitly cultivated as its differentiator.
Contends that OpenAI and Google have long treated developers this way, but Anthropic explicitly positioned itself as the grown-up lab that would honor API contracts and communicate clearly. The frustration isn't the behavior itself — it's the collapse of a marketed premise that many developers paid a premium to buy into.
Dozens of first-person reports describe startups whose Claude Code seats were throttled to uselessness the week after signing annual contracts, indie devs discovering claude-3-opus returning 404s in production, and teams whose p95 latency doubled after silent backend routing changes. These aren't hypothetical grievances but material breakages of running systems.
Centers the complaint on Anthropic's terse one-line changelog entries like 'Updated Claude Code rate limits' that tell developers something changed without specifying what, when it takes effect, or how to appeal. The lack of public response to the post itself reinforces the pattern that Anthropic no longer sees clear communication as part of its product.
A blog post titled *Anthropic's Method to Losing Goodwill in a Few Easy Steps* hit the Hacker News front page (196 points) with a plainly-argued grievance: the company that positioned itself as the safety-and-craft alternative to OpenAI has been treating its paying developers with the same casual contempt it once criticized. The post is not a hit piece. It's a checklist. Rate limits tightened without warning. Claude Code usage caps recalibrated mid-billing-cycle. Model deprecations announced with weeks, not quarters, of runway. Support tickets closed with template replies. A pricing page that keeps quietly reshuffling.
None of these on their own would matter — every API vendor tunes limits and retires models — but the accumulation reads as a company that has stopped pretending its developers are partners. The comment thread on Hacker News amplified the pattern with dozens of first-person reports: startups whose Claude Code seats got throttled to uselessness the week after signing an annual contract, indie devs who woke up to find `claude-3-opus` returning 404s in production, teams that discovered their observed p95 latency had doubled because a silent routing change moved them to a different backend pool.
Anthropic hasn't publicly responded to the post as of writing. Its status page and changelog remain, per the author's central complaint, terse to the point of unhelpful — the kind of one-liners ("Updated Claude Code rate limits") that tell you something changed without telling you what, when it takes effect, or how to appeal.
The frustrating part isn't that Anthropic is doing anything unusual for a frontier lab. OpenAI has burned developers this way for two years. Google's model deprecation cadence on Vertex is a running joke. The frustrating part is that Anthropic explicitly sold itself as the grown-up in the room — the lab that would respect the API contract, communicate clearly, and treat developers as a first-class audience rather than a distribution channel for the consumer app. That pitch worked. Claude Code shipped and immediately became the default choice for a big slice of the professional coding market. Enterprise deals followed. The developer surplus was real, and it was earned by being visibly better at the boring stuff — docs, SDK ergonomics, model behavior on realistic code — not by any single benchmark.
That surplus is what's now being spent. The problem is structural, not personal: a lab racing to catch OpenAI on model capability, catch Cursor on IDE integration, and catch Microsoft on enterprise distribution will, at some point, discover that the operational discipline required to run a developer platform is at war with the velocity required to ship frontier models. Something gets sacrificed. The Hacker News post is a diagnosis of what got sacrificed.
Compare to how the mature API providers handle the same tradeoffs. AWS deprecates services on 12-month timelines with named migration paths and, often, migration tooling. Stripe changes pricing with 90-day notice and grandfathering. Even OpenAI, after enough public floggings, now publishes multi-quarter model sunset schedules. The bar isn't "never change things" — it's "tell me what's changing, when, and how to adapt, before I discover it in a paging alert." Anthropic's changelog culture, as multiple commenters noted, still reads like an internal Slack channel accidentally piped to the public. Terse. Assumptive. Written by people who know the context, for people who don't.
The second-order effect is the one that should worry Anthropic's revenue team. Developer trust is a two-way ratchet: you can only turn it in one direction at a time, and it takes an order of magnitude longer to turn back up than it does to turn down. The Claude Code users writing angry Hacker News comments today are the same people who two months from now will be quietly evaluating Cursor's shift to Composer, or building an abstraction layer over Bedrock, or resurrecting their local-model experiments. Not because any of those are strictly better, but because "strictly better" isn't the axis they're now optimizing on. They're optimizing on *predictability*.
If your production system has a hardcoded `model="claude-3-5-sonnet-20241022"` and a hardcoded assumption about tokens-per-minute, you have technical debt you didn't know you had. The concrete moves:
Abstract the provider. Even a thin wrapper — one file, model ID passed by config, provider passed by env var — buys you the ability to failover to Bedrock's Claude, or to a non-Claude model, without a code deploy. If you use one of the AI SDKs (Vercel's `ai`, LangChain, LiteLLM), you already have this. If you're calling `anthropic.messages.create` directly from three different files, you have work to do this sprint.
Track your model IDs like dependencies. Add a smoke test that hits every model ID your code references and fails CI if any of them returns a 404 or a deprecation header. Anthropic's API does return deprecation warnings in response headers before a hard sunset — most SDKs surface them but most codebases never log them. Two lines of middleware fix that.
Instrument for silent regressions. The specific complaint that hurt most in the Hacker News thread was users whose *quality* changed without their limits changing — a suggestion that routing or quantization had moved under them. You cannot detect this without an eval suite you run on every provider change. If you don't have one, the cost of building a 50-example eval that runs on every model version is roughly one afternoon, and it is the single highest-leverage thing you can do to survive the next quiet change.
Read the fine print on rate limits. Anthropic's tier structure ties limits to spend, and moving between tiers is not automatic. If you're planning a launch that will 10x your usage, open the ticket now, not the day of.
None of this is unfixable, and Anthropic has the institutional muscle to fix it — a real changelog, a deprecation policy in writing, a status page that treats "we tightened limits on Sonnet in the EU region" as newsworthy rather than routine. The question is whether the company treats the goodwill leak as an operational problem to solve or as a cost of doing frontier-lab business. The former is a Tuesday afternoon of policy writing and a headcount on developer relations. The latter is how you become the vendor developers use because they have to, not because they want to. Anthropic has, until recently, been the second kind. It should want to stay that way.
> Your Claude subscription—which is a cheaper version of the Anthropic API—is restricted to use with the Claude Code CLI/Desktop, Claude CoWork, or @Claude in Slack.Thats not true at all. You can use the Agent SDK [1], which uses your subscription [2]. I use it via ACP [3] with custom system
It worth noting that – just to add to the confusion – they apparently cancelled the June 15th change just before it was due to go live:https://support.claude.com/en/articles/15036540-use-the-clau...https://the-decoder.com/anthropic-backs-off-unpopular-billin..
The more interesting question is why would you at any point ever extend any goodwill to an unbounded corporate entity?Whenever you are faced with a corpo you should conceptualize it as a system that will happily mow you over for increased profits, unless it is legally bound to not prioritize profits
Not sure why this was flagged, Anthropic has obviously been burning bridges. I thought this line was funny though:> If you want to autocomplete, like I do, you don’t need Fable, or even Opus; Sonnet works fine.It reads like "if you want to go to the grocery store, you don't need a space
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
They're subsidizing their tokens as long as you use their software. That's a fair exchange, I never understood why people took issue with it.If you don't want to get locked in to Claude Code, you can pay more. Just like you can pay more for an unlocked non-carrier subsidized phone. (W