The model is the commodity: two years of context engineering at 8am AI

If the agentic thread is the corpus's clearest arc, the context thread is its clearest conviction — the same claim, sharpened week after week: the model is the commodity, and the context is the lever. What separates a useful interaction from a useless one is rarely the model. It's whether the human can hand it a structured, current, well-shaped context — and whether they understand the work well enough to know when the output is wrong.

The vocabulary changed across two years. The conviction didn't.

2024: scrubs

The earliest name for it was scrubs — basic sets of instructions and metadata you'd attach to an LLM so it could do a task reliably. In April 2024 that's already the move: don't reach for a bigger model, give the model a better frame. By August the group is running AI over Hanif's spreadsheet data not to replace the analysis but to extend what's already there — context as raw material, not magic.

The era's lesson, stated early and often: a model with a good instruction set beats a better model with none.

2025: corpus, portability, and the limits

Through 2025 the conversation gets more demanding. It's no longer "write a scrub"; it's "build and maintain a corpus." Several threads recur:

• The corpus is the asset. The freshest, best-structured context — not the model subscription — is what compounds. People start treating their documents, notes, and prior outputs as a living substrate the model reads.
• Portability is the bottleneck. A repeated 2025 frustration: context is trapped at the app level. You can build something good inside one tool and lose it the moment you switch. The group rates the value of portability as extremely high precisely because it's so hard to get.
• AI gets you 95%. Scott's recurring line about his investor table — the model gets him most of the way, and the last 5% is human. The gap isn't capability; it's the context and judgment only the human holds.

2026: "how do you know"

By 2026 the conviction has a sharper edge and a name for its missing piece. The group keeps circling the question "how do you know?" — not "is the output good" but "what in your process tells you it's good, independent of the thing that produced it?"

This is context engineering grown up. It's no longer just about feeding the model better inputs; it's about building an external check on a self-contained process — because a workflow that grades its own work has no observability into its own reinforcement. The project-state pattern shows up here as the structural answer: a corpus that's also a map, so both human and model can trace a claim back to where it came from.

And the foundational-knowledge worry lands in the same place. Juan's 2026 framing: you have to be able to combat an AI suggestion — to critique an approach — which you can only do if you actually understand the domain. The context isn't only what you give the model. It's what you carry yourself.

What stayed true

Across two years and a changing vocabulary, three claims never moved:

1. Reach for context before capability. Almost every "the model can't do X" turned out to be "the model wasn't given enough to do X."
2. The corpus compounds; the subscription doesn't. Time spent structuring context paid off repeatedly. Time spent chasing the newest model mostly didn't.
3. Someone has to be the eval. Whether it's called self-reflection, observability, or "how do you know," the human stays the source of ground truth. The context makes the model useful; the human makes it trustworthy.

Open question

If context is the lever, the next scarce skill is building context that travels — portable, inspectable, and current. Two years in, the group has the conviction and the patterns but not yet a clean answer to portability. That's the thread to follow.