Anthropic Didn’t Get the Memo and Is About to Find Out.
The 90’s taught us two lessons. If you weren’t there, you might hear that and think it’s nostalgia from older tech people who miss CRT monitors and dial-up sounds. It’s not nostalgia. It’s a warning label.
Because the 90’s weren’t just a decade of clunky UIs and beige boxes. They were the decade where the modern rules of platform dominance got written in permanent ink: rules that still decide who wins today when a technology becomes ““”default”.
And those rules were simple:
- Microsoft showed the world how to win by feeding developers.
- Apple showed the world how close a company can get to dying by locking the ecosystem down until a Steve Jobs–level reset saves it.
Now Anthropic is making a move that looks like it learned the wrong lesson from history.
Lesson #1: Microsoft didn’t just sell software: it seeded a generation
Let me translate the Microsoft playbook into plain language for anyone who didn’t live through it:
Microsoft didn’t become dominant because it had the most elegant product in every category. It became dominant because it became the language people learned. And people learned it because it was everywhere: at home, at school, in internet cafés, on office desks, inside small businesses.
The messy part, and this is where the younger audience often underestimates the era, was that distribution wasn’t “clean”: piracy existed. Grey-market installs existed. Copies were passed around. Microsoft obviously didn’t love it, but the company understood a deeper truth: the fastest way to win a platform war is to let people grow up inside your ecosystem.
If a teenager learns your OS, your office suite, your dev tools, your workflow, then ten years later, when they’re hired as a sysadmin, dev lead, or CTO, your stack isn’t a vendor choice. It’s muscle memory. It’s what feels normal. That’s how “enterprise standards” happen: not with a sales brochure, but with a generation of builders who already default to you.
Microsoft didn’t just monetize. It cultivated.
Lesson #2: Apple nearly learned “control” the hard way… permanently
Apple’s pre-comeback era is often rewritten as a heroic prelude. In reality, it was frightening.
It wasn’t just that Apple had product problems. It was that Apple had a posture problem: closed, defensive, internally obsessed with control, and structurally bad at letting the outside world build around it. The ecosystem didn’t compound. It resisted.
And yes, Apple’s turnaround is legendary but, here’s the part people forget: Apple didn’t escape because the gated mindset was secretly genius. Apple escaped because it had a Steve Jobs and a Jony Ive.
That’s not a moral judgment. It’s a strategic point.
A closed ecosystem is survivable only if you can create something so uniquely compelling that people tolerate the friction. Apple eventually did. But the “closed years” nearly killed the company before it got there.
Which brings us to Anthropic.
The Claude Code crackdown is not a policy footnote: it’s a platform signal
Anthropic’s latest “ban” around Claude Code usage via third-party bots/tools is framed as legal/compliance. But functionally, it’s a strategic statement:
Your Claude subscription identity (Free/Pro/Max OAuth) is for Anthropic’s official surfaces. Not for third-party products. Not for external agent harnesses. Not for tooling that wraps Claude Code.
In other words: don’t build platforms around our platform unless you do it on our metered API terms.
I’m not pretending I don’t understand the economics. Agent workflows can burn compute in ways a human never will. Tool use adds cost. Running inference is expensive. Subscriptions were never priced to become a universal, programmable fuel source for external automation.
But platform strategy isn’t about what’s “understandable.” It’s about what it trains the market to do.
And this trains the market to do something deadly to a model company:
It trains developers to architect Claude as replaceable.
The part Anthropic seems to be missing: models are getting swappable, fast
Claude is excellent. In many cases it’s top-of-the-line.
That’s not the issue.
The issue is that being excellent is not the same as being defensible. Especially in a world where developers are already building multi-provider systems by default because, they know vendors change rules, prices, rate limits, and access paths overnight.
Anthropic’s crackdown accelerates the exact behavior that makes a model provider weaker over time:
- abstraction layers
- provider fallbacks
- “config switch” replacements
- routing logic
- cost-optimizing model selection
The more you push developers to treat you as a metered utility rather than a partner ecosystem, the faster you become a commodity in their architecture.
And here’s the kicker: unlike Apple, Anthropic does not have a “Jony Ive moat”. They don’t have industrial design and an OS layer and a hardware stack. They have a model and, in this market, models are increasingly substitutable for large classes of work.
That’s why this is such a faux-pas.
You’re choosing the gated-ecosystem posture without owning the kind of unique end-to-end experience that makes gating survivable.
While Anthropic gates, defaults are being captured elsewhere
This part is where the 90’s analogy stops being abstract and starts being scary.
In the browser era, Internet Explorer didn’t win because it was beloved by power users. It won because it was there. Default. Preinstalled. “Good enough”. And most people didn’t change it.
We are watching the assistant era form the same way.
If Google’s Gemini becomes the preinstalled, OS-integrated AI for the “non-IT savvy” majority, people who won’t bother switching providers, then the default layer hardens. Habits form. Integrations standardize. The ecosystem compounds around whoever owns the default distribution channel.
In a default war, the biggest mistake you can make is alienating the builders who could have made you the “unofficial default” in their worlds.
And that’s the meta-problem with Anthropic’s move: it punishes exactly the kind of developer experimentation that creates grassroots dominance.
OpenAI is leaning into the opposite lesson: UX scales human parallelism
This is why Greg Brockman’s point about the Codex app matters so much, and why it belongs in this discussion.
He’s basically saying: the CLI is powerful, but it doesn’t scale the human mind well when you’re coordinating parallel agents, multiple worktrees, multiple repos, and state across tasks. The GUI isn’t decoration, it’s cognitive infrastructure:
“…the Codex app changed my workflow… it becomes much easier to juggle… multiple parallel threads of execution… all of that complexity is something that’s much better managed in a GUI than in a terminal interface… The Codex app lets you go further, lets you do more in parallel, and lets you go deeper on the problems you care about.”
That observation maps directly to what’s happening with developer tooling right now: we’re moving from “one human, one terminal, one thread” to “one human orchestrating many threads”.
Codemus (the application I am creating that allows to orchestrate AI Agents and enables Agile Vibecoding) was being built on the premise that GUI-based orchestration unlocks levels of parallel work that CLI-based workflows make unnecessarily painful. And it was totally built around Claude because Claude was (and is) extremely strong for coding and reasoning.
But a crackdown like this doesn’t just change a login mechanism. It changes the risk profile of the entire platform relationship. It tells every builder of GUIs, orchestration layers, and third-party toolchains:
“Don’t assume we want you in the ecosystem unless you’re on our terms”.
So the rational reaction is exactly what you’re doing: rewrite to support other providers. Not because you want to. Because you’d be irresponsible not to.
And every time a serious builder does that, Claude becomes a little less like “the platform” and a little more like “one of the engines you can swap” or even to avoid altogether.
What Anthropic is about to find out
Here’s the “about to find out” part, and it’s the part the 90’s makes painfully predictable:
If you lock down too early, you don’t stop innovation. You relocate it.
Developers won’t stop building agentic workflows, wrappers, orchestration GUIs, and automation layers. They’ll just build them on top of the providers who behave like Microsoft did: welcoming enough that the ecosystem can compound.
And if Anthropic doesn’t change course, the long-term cost won’t be a few angry GitHub issues or some loud Twitter threads.
The cost will be far worse:
- Claude becomes one backend among many
- developer tools stop defaulting to Claude
- the ecosystem compounds around someone else
- and the “default” layer hardens without Anthropic inside it
Apple escaped the gated era because it had a once-in-a-generation comeback story. Anthropic is not Apple. Anthropic doesn’t ship the device, the OS, the interface layer, and the distribution channel. Anthropic ships a model and, in a market that’s rapidly learning how to swap models, gating the ecosystem is not a power or even clever move.
It’s a surrender move dressed up as compliance.
The 90’s taught us that the winners don’t just build good technology. They nurture the people who decide what becomes normal.
Anthropic didn’t get the memo.
And if they don’t course-correct, they’re about to relearn it the hard way.