AI 시대의 공방 — 한 사람이 넓고 깊게 일할 수 있는 시대

For two centuries, you couldn’t be both broad and deep. You had to choose.

Going broad meant going shallow — generalist, dilettante, jack-of-all-trades. Going deep meant going narrow — specialist, expert, lifer in one field. Academia institutionalized this trade-off. Industry reinforced it.

The trade-off wasn’t a preference; it was real. The cost of catching up in any field — its current literature, its tools, its idioms — was 5 to 10 years. By the time you learned one, the next had moved. Solo work across many fields, at depth, became a Renaissance memory.

That trade-off is collapsing.

What changed

Tools changed.

Code that took weeks now takes hours. Literature review that took months now takes days. New frameworks that took semesters now take weekends. Experiment cycles that took weeks now take days.

When the cost of operating in a field drops by an order of magnitude, the cost of spanning multiple fields drops similarly. The trade-off — broad versus deep — that defined modern professional life starts to bend.

I’ve spent the last six months as one data point in that bending:

• Transformer interpretability and surgical correction (paper9)
• Korean Road Name Address as a coordinate-free linear reference system (three-paper series)
• Medical imaging model compression with an integrated diagnostic framework (CheXNet paper)
• Narrative dynamics — a cognitive-energy framework (overview)
• Long-form fiction — a novel serialized on Royal Road and Munpia (Flow)

Five fields. Not parallel — integrated. The same framework — how attention is built, sustained, and released — runs through everything. The integration is the point, not the count.

The bottega returns

Historically, work across many fields required one of two models.

The Renaissance bottega: a single master plus apprentices. Leonardo’s workshop in Florence was this. The master held integration; apprentices executed under direction.

The 20th-century lab or company: principal investigator plus graduate students, or founder plus employees. Wolfram Research, in part, runs this way.

Both required a team. Solo work across multiple fields, at depth, was effectively impossible.

A third model is now viable.

Solo plus AI tools: one master, with tools in the apprentice role.

This is structurally closer to the Renaissance model than to the modern lab. Integration happens in one head. Execution is delegated, but to instruments rather than people. The instruments work continuously, don’t have research interests of their own, and don’t compete for credit.

I call this site River Bottega for this reason. Not a lab — an atelier. One person, several fields, an integrated thesis, with tool-assisted execution.

The new bottleneck

When tools collapse the cost of execution, the bottleneck moves.

Old bottleneck	New bottleneck
Time to code	What to code
Time to learn a field	Which field is worth entering
Tool fluency	Maintaining an integrated thesis
Experiment setup	Quality control across domains

The shift is from technical capacity to taste. Strategic judgment. The ability to see what matters in five fields at once and not lose any of them.

This makes the new model harder than it sounds. Doing five fields adequately is now easy. Doing five fields with depth — with an integrated thesis that holds across all of them — is still difficult. The difficulty has changed shape; it has not disappeared.

A new category, no name yet

Existing categories:

• Specialist — one field, deep
• Generalist — many fields, shallow
• Manager — integrates but does not produce

A new category, undefined for now:

• AI-assisted solo operator integrating one framework across multiple fields

The category does not have a clean name yet. There is likely a first wave of people working in this mode right now, mostly invisible to each other because the existing categories make us look like five different things doing five separate jobs.

What this is not

This is not the end of the specialist. Thirty years deep in one field is still irreplaceable. The new mode runs alongside the older one, not in place of it. Most working researchers are still in the older model and should remain there if it serves their goals.

This is also not a claim that I am executing this mode well. The mode itself is what’s interesting; whether the execution holds up is for the work to demonstrate, not for an essay to assert.

Why this matters

For most of recorded modern history, the form of intellectual work was determined by what one head plus one pair of hands could process per day. That constraint produced the trade-off: depth required narrowness because there was only so much one person could read, code, and verify in a year.

When that constraint loosens — when one head plus one pair of hands plus tools can do the work of a small team — the form of intellectual work changes. New combinations become possible. Old categories stop describing what is actually happening.

This site is one place trying to live in that change. New work goes up regularly. The fields differ; the thesis does not.

Thanks for reading.