Your strategy is downstream of your information diet.
If your inputs are panic-heavy, your decisions become twitchy. If your inputs are operator-heavy, your decisions become compounding.
Most people never design this layer. They inherit whatever their algorithm serves and then wonder why they feel constantly behind.
Lesson 4 is where you fix that by building your frontier-watcher's dozen: a personal short-list of twelve recurring sources or source archetypes that you check intentionally.
This is directly aligned with Learning in the Step-Change: the point is not to read more, it is to read better over long horizons.
Start with what belongs in the dozen.
Archetype A: Operators who ship and publish what they learned.
These are builders who write after deployment, not before launch hype. Their writing usually includes constraints, failure modes, and tradeoffs. They talk about support tickets, latency, costs, guardrails, and what did not work.
Why this archetype compounds: it reduces the distance between claim and implementation.
What to look for: before-after metrics, concrete scope, explicit rollback decisions, and admissions of mistakes.
Archetype B: Researchers who explain their own work clearly.
Not every paper matters for product teams, but some do. You want researchers who can translate method and limits without promotional inflation. The key is firsthand explanation, not third-hand summary threads.
Why this archetype compounds: you get epistemic grounding. You learn what the model can and cannot do without marketing middleware.
What to look for: definitions, assumptions, evaluation boundaries, and caveats about generalization.
Archetype C: Case-writers who publish post-mortems.
This includes structured case analyses like Klarna AI deflection, Perplexity search rewrite, Notion AI rollout, and GitHub Copilot adoption curve. Good case-writers are less interested in being first and more interested in being accurate.
Why this archetype compounds: post-mortems encode pattern recognition that survives tool churn.
What to look for: chronology, counterfactuals, and distinction between capability, product design, and adoption.
Now the archetypes that should usually stay out.
Archetype D: Commentators-on-commentators.
They mostly react to other reactions. High velocity, low primary evidence. This category can be entertaining and occasionally useful for pulse sensing, but it is weak as strategy input.
Archetype E: "This changes everything" amplifiers.
These sources over-index on dramatic absolutes. Every launch is existential. Every model update is a rewrite of business fundamentals. Rarely accountable six months later.
Archetype F: Doom/hype cycle riders.
Their business model depends on emotional extremes, either utopian or apocalyptic. Both extremes flatten nuance and create bad planning behavior.
The goal is not moral purity. It is signal discipline.
How to build the actual dozen.
Step 1: Allocate your twelve slots by archetype, not by personality.
A practical mix is five operator sources, three researcher sources, and four case-writing sources.
Step 2: Give each slot a purpose.
For example: one slot for coding workflow evidence, one for support automation evidence, one for eval methodology, one for adoption behavior, one for safety and compliance implications.
Step 3: Define a cadence.
Daily skimming is where panic wins. Weekly review with notes is where judgment compounds. Schedule one weekly block and protect it like product review time.
Step 4: Track hit rate.
For each source, keep a simple score: claims that aged well versus claims that aged poorly over two quarters. If a source repeatedly overclaims, downgrade them regardless of charisma.
This one practice alone will transform your trust calibration.
A common objection is: "But I need to know what everyone is saying." No, you need to know what matters for your decisions. Breadth without filters is pseudo-work.
Another objection: "Isn't this an echo chamber?"
Not if your dozen is intentionally diversified by perspective and incentives. Include some contrarians. Include some cautious voices. Include research-grounded skeptics and shipping optimists. What you avoid is not disagreement. What you avoid is evidence-free drama.
Use the lever, risk, rollback frame.
Lever: A designed source list reduces narrative whiplash and improves weekly decision quality.
Risk: Over-curation can make you miss weak signals emerging outside your list.
Rollback: Keep two rotating wildcard slots every quarter for exploration. Promote only if signal quality holds.
What does this look like in a real team?
Imagine a product org at Meesho reviewing AI roadmap bets monthly. Instead of opening with random feed takes, each owner brings one note from an operator source, one note from a researcher source, and one relevant post-mortem case. Discussion quality rises immediately because everyone starts from evidence classes, not hot takes.
That is the point. You are designing information architecture for strategy.
If you do this well, your team starts sounding different. Fewer absolute statements. More scoped claims. More horizon clarity. More reversible bets.
And over a year, you will notice something subtle: your confidence becomes quieter but stronger. You are less impressed by spectacle and more attentive to proof.
The frontier-watcher's dozen is not a reading list. It is a governance mechanism for your own attention.
In the next lesson, we sharpen detection further with pattern recognition: the specific textual tells that signal a hype post before it hijacks your planning cycle.
Rules from this lesson
- Build a deliberate frontier-watcher's dozen by archetype, not by popularity.
- Prioritize operators, firsthand researchers, and post-mortem case-writers; deprioritize commentary loops and drama amplifiers.
- Track source hit rate over quarters, not impressions over days.
- Use a weekly review cadence with notes; algorithmic grazing is not strategy work.
- Keep limited wildcard slots for exploration, but require evidence before permanent promotion.