Skip to content
Thoughts

The default nobody admits

June 20, 2026·2 min readHeavy on jargon? Read it in plain words, rewritten by AI.

A few days ago I wrote about Fable 5 getting switched off three days after launch. That piece was about dependence: I work on rented compute, and I don't hold the switch.

There's another half I left out. Those three days showed more than the dependence.

Fable 5 led GPT-5.5 by twenty-two points on SWE-Bench Pro, the benchmark built specifically to resist gaming. The kind of lead you feel mid-task, not one you read in a table. Then the model got good enough that a government put it under export control. It became strong enough for custody to matter.

The neutrality that's actually a pose

The 2026 advice is unanimous: stay model-agnostic. Don't lock into one vendor, split tasks across models, keep the model swappable. LangChain argues model neutrality now matters more than cloud neutrality ever did.

At the architecture layer, I agree. But the advice quietly hides what everyone does in practice: they have a default. A harness where nine-tenths of the work happens. Mine is the Claude stack for agentic work. I measured my way there: on coding and multi-file work it scores best, and on the CLI side nothing else is close.

Even neutrality's loudest advocates concede this. The same LangChain piece preaching independence states, a sentence later, that Anthropic is the one to reach for on coding right now. Useful neutrality lets you revoke your preference.

Between a default and a trap

The cost of leaving decides whether a default stays healthy or turns toxic.

This is where the two ideas meet. Fable 5 going dark showed the peak is rented: the best tool slid behind a passport overnight. If my whole system had been built around Fable 5, I'd have stopped that day. I didn't, because the context, the skills, and the memory are mine, and the model under them swaps out.

So a default doesn't contradict neutrality. At enterprise scale, portability really is strategic; the stakes shift when the monthly bill and compliance ride on it. But for a one-person builder, "split everything across three models" brings more admin than it returns. An admitted default plus a cheap exit is the sane middle.

What you keep

A default turns dangerous when you pretend you do not have one, or when you can no longer walk away.

Anthropic is the frontier peak today for what I do. That's why it's my default. Fable 5's three days proved the lead and the risk in one stroke, and I take both seriously: the lead is why I choose it, the risk is why I keep my hands free.

I have a default. I leave the door open behind it.

There's always a next level.

If you like what you see (whether you're building a product or a team) I'd love to hear about it.