| name | stack-year-review |
| description | Structures longform year or phase reviews: stack at t0 vs t1, personally notable bullets, benchmark fatigue honesty, paradox TL;DR, and deep links. Use when synchronizing audience mental models after a fast-moving year in tech. |
Stack Year Review
Distilled pattern: once a year (or after a phase change), publish a before/after picture of the production stack plus a personal highlight reel—notable papers, releases, and intuitions—closed with tensions instead of a single hype arc.
When to use
- The field moved fast and your audience only saw fragments; you want to synchronize mental models.
- You need a longform piece that still rewards skimming.
Steps
- Draw the stack at t₀ and t₁ — What did “default serious stack” look like at year start vs end? Boxes and arrows beat prose.
- List what actually moved capability — RL, reasoning traces, tool use, evals, product layer—whatever you believe mattered; keep each bullet falsifiable.
- Name one ontology shift — How should readers feel the new object? (e.g., from “training animals” to “summoning ghosts”—use your own metaphor; do not copy others’ wording.)
- Admit benchmark fatigue — If leaderboards diverged from felt reality, say so and point to what you trust instead (tasks, demos, production metrics).
- Product / ecosystem paragraph — How did the “app layer” thicken? (IDEs, agents, vertical bundles.) Ask the open economic question: who captures value?
- TL;DR with paradox — One paragraph that holds both “smarter than expected” and “dumber than expected” (or your field’s equivalent)—signals intellectual honesty.
- Deep links — Point to your longer essays or talks for readers who want rabbit holes.
Anti-patterns
- A list of headlines anyone could copy from TechCrunch.
- One-note optimism or one-note doom.
- No personal timestamp (“as practitioner/builder/researcher, I noticed…”).
Output shape
- Diagram or bullet stack: then vs now.
- 5–15 “notable” bullets (mixed scope: infra + product + science).
- TL;DR: tension + forward-looking open field.
