• source https://www.joanwestenberg.com/the-coherence-premium/
  • by JA Westenberg
  • … people using AI are producing more stuff faster without any clear sense of whether the stuff is good or consistent …
  • … single person with the right tools can do what used to require a department.
    • … AI can execute tasks that once demanded teams of specialists.
  • … economics that made large firms necessary are reversing.
  • … solopreneur’s …
    • … advantage is coherence.
  • … coherence … : the degree to which every part of an operation derives from the same understanding, the same model of reality and set of priorities and tradeoffs.
    • … entire loop happens inside one mind.
  • Every handoff loses information, every translation introduces drift, and every layer of abstraction moves further from ground truth.
  • Organizations have spent decades trying to solve this problem.
    • … fighting against something that won’t budge:
      • knowledge is sticky and context is lossy,
      • and understanding doesn’t transfer perfectly between humans.
  • … company starts small, with the founders doing everything themselves.
    • … decisions quickly because they understand everything about the business, and the business works.
  • … hire people … transfer processes. … new employees don’t have the context that generated those processes …
  • … adding coordination mechanisms like project managers, alignment meetings, and cross-functional reviews.
  • … 2022 Microsoft study found … spend over 50% of their time on internal communication and coordination.
  • context fragmentation
    • … information isn’t understanding …
      • Understanding = integration
        • … integration happens in minds
    • Each person reads a fragment, interprets it through their own context, and forms their own understanding.
      • People don’t disagree on facts; they’re operating from different maps of the same territory.
    • … each employee has their own AI assistant, trained on whatever context they happen to feed it …
      • AI amplifies individual perspectives rather than creating shared ones.
  • … operating alone, you have one context, one understanding, one model ..
  • ^^… humans are good at integration and judgment …^^
  • ^^… using AI to do random shit faster, you’re not capturing the advantage.^^
  • coherence stack
    • … mind layer, which is you: your understanding, your judgment, your integrated model … layer can’t be automated or delegated, and it’s the source of coherence.
    • context layer … documents that AI tools can consume. … your constraints and tradeoffs, your voice guidelines, your decision history.
    • execution layer, where AI operates.
    • output layer …nothing reaches this layer without passing through a coherence check:
      • does this output reflect my model? Would I have produced something like this? Does it fit with everything else?
  • Most people using AI skip the context layer entirely.
  • context layer
    • What problem do I solve, and for whom specifically?
    • What’s my actual thesis for why my approach works
    • What are the core tradeoffs I’ve chosen?
      • … maintain a decision log. When you make a significant choice, write down what you decided and why.
    • What do I explicitly not do?
      • … build your constraints file.
        • For example:
          • … When choosing between comprehensive and focused, choose focused. …
    • How do I sound?
      • Include examples. Point to a piece you wrote that exemplifies the voice, and one that doesn’t.
  • … short enough to include in AI prompts. … compression without loss of generative power.
  • Every output that ships should pass through a coherence check.
    • You can automate part of this.
    • Feed your AI tool your recent outputs and ask it to compare a new draft against them, flagging inconsistencies.
    • … final judgment has to be yours.
  • anti-patterns that break coherence
    • Context starvation … ask AI to do something without feeding it your operating model, your constraints, your voice.
    • AI outputs without proper review … deviations accumulate.
    • Model staleness … understanding evolves but your context documents don’t.
    • Fragmented tooling = using different AI tools with different contexts.
    • Decision amnesia = making choices without recording the reasoning.
  • … audit your operation for coherence as often as possible. … frequency matters less than doing it at all.
    • … audit also reveals context layer gaps.
    • Add constraints, add examples, make the implicit explicit.
  • … coherence advantage works best in domains where the value comes from understanding rather than from physical or regulatory scale.
  • … large organizations haven’t figured out how to respond. … give everyone AI tools and hope for productivity gains, which accelerates their fragmentation …
  • … transaction costs are collapsing …
  • Coasean logic that justified large firms is weakening.
    • Coase Theorem
      • Example: A factory pollutes a fisherman’s river. Whether the factory has the right to pollute, or the fisherman has the right to clean water, they will negotiate to the same efficient outcome — the one that maximizes total value. The initial assignment of rights only affects distribution (who pays whom), not efficiency.
  • If there is a new moat (and I’ll admit, that’s a big “if”) it probably looks like coherence; the ability to operate as one mind, one understanding, one model, …