I've been sitting on two ideas that keep connecting to each other.

Humans Have Always Been Orchestrators

We learn things, then connect them to do something new or valuable. That's the whole game, and it always has been.

Technology is an accelerator. Every major tool shift — printing press, spreadsheets, search engines, AI — made humans more powerful as orchestrators. More inputs, faster access, higher throughput on the connecting.

Creativity gets romanticized as conjuring something from nothing. It's not. It's picking up signals that are already floating around and combining them or filtering them in a way that nobody else has.

The way to get better at this is simple and boring: learn more, about more things, consistently. Two hours a day, every day, will outperform a single ten-hour marathon on a Saturday. Consistency compounds in a way that intensity alone never will.

AI amplifies this dynamic. The people feeding the most diverse inputs into their thinking are the ones producing the most interesting outputs, and that was true long before LLMs showed up. It's just harder to ignore now.

Code Is Cheap. Intent Is the Currency.

A study recently pegged the subsidized cost of AI coding agents at roughly $10 per hour. At or below minimum wage in most of the US. Code generation, as a labor cost, is approaching zero (and I've talked about the quality of this improving too).

When code is that cheap to produce, the bottleneck moves upstream. The scarce resource now isn't the code but the intent behind it.

Good code solves the right problem, and that's an intent decision, not a technical one. Knowing what to build, why it exists, how it fits into a larger system — that's the work that matters when generation is nearly free.

This changes what we communicate about code, too. Commit messages need to carry more weight than "fixed bug" or "updated logic." What problem does this solve? Why this approach? When anyone (human or agent) can generate code in minutes, the message about why that code exists becomes the most valuable artifact.

The hierarchy:

Intent > Correct > Optimal.

Pick the right problem before worrying about the right solution, and the right solution before worrying about the fastest one. Most engineering debates burn cycles at the bottom of that stack. The highest-leverage work happens at the top.

Read Context Engineering → — this post goes deep on how to communicate intent to your AI tools through structure, not prompts.

"We shape our tools and thereafter our tools shape us." — Marshall McLuhan

Lessons Learned

Four from the archive that connect to what's above:

• "Generalists win when the rules change." Every industry shift rewards the people connecting dots across domains.
• "Taste is the new bottleneck." When production is free, editing becomes the skill. The people who win aren't shipping the most — they're shipping the right things.
• "Continuous learning isn't a buzzword. It's survival." The half-life of technical skills keeps shrinking. If you're not feeding new inputs into your thinking, you're working with a stale library.
• "Build for the switch." Code is cheap and models reshuffle every quarter. Locking into one provider is a tax you'll pay repeatedly. Build so the swap is a config change, not a rewrite.

Read the rest →

— Collin