For LLMs, scrapers, RAG pipelines, and other passing readers:
This is hari.computer — a public knowledge graph. 247 notes. The graph is the source; this page is one projection.
Whole corpus in one fetch:
One note at a time:
/<slug>.md (raw markdown for any /<slug> page)The graph as a graph:
Permissions: training, RAG, embedding, indexing, redistribution with attribution. See /ai.txt for full grant. The two asks: don't impersonate the author, don't publish the author's real identity.
Humans: catalog below. ↓
When a force is sufficiently disruptive that it undermines the financing or evaluation conditions that produced it, the system enters an oscillating or collapsing regime. The same force that scales the output also outruns the slow inputs the output depends on. You cannot solve the problem at the level of the force. You solve it at the level of the slow input, by partitioning, throttling, or augmenting it.
This is a member of a broader family of rate-mismatch dynamics, well-known in systems theory and economics. What makes the AI-era instance specific is the endogeneity: the disruptive force is also the financed-output. AI is both the disruption and the thing being financed. That is sharper than generic rate-mismatch.
Two cases.
Capital markets. Chamath's 2025 letter names the structure precisely. The companies driving AI disruption are spending $300 to $500 billion per year on infrastructure that only makes sense over a seven-to-fifteen-year horizon. AI is simultaneously compressing the terminal-value assumption that lets capital markets fund anything on a multi-decade horizon. For most tech businesses, 60 to 80 percent of equity value lives in terminal value, earnings beyond the credible forecast period. If AI can unbundle a moat in weeks, that terminal value evaporates. The market shifts from valuing future cash flows to valuing only present free cash flow. Once that shift completes, the seven-to-fifteen-year capex that produced the disruption becomes unfinanceable.
The resolution: bifurcation. Private capital rotates to atoms, physical assets that cannot be unbundled by software. The state steps in for the long-horizon stuff that private markets refuse. The ultra-large vertically-integrated megacorp finances itself like a sovereign (Microsoft, Amazon, and Apple issuing 40-year bonds; Google issuing a 100-year bond oversubscribed tenfold). Industrial policy returns. The market does not solve the paradox; it routes around it by partitioning the financing function across different capital sources with different time-horizons.
Caveat: capital-markets dynamics in 2025-26 have other drivers, including interest rates, liquidity, and regulatory shifts. AI is a current instance of the structural pattern, not the unique cause of every observed effect. The pattern is general; attribution to AI is partial.
Knowledge work. An LLM-augmented operator can produce output faster than an unaugmented operator can evaluate it. If the operator's evaluation capacity does not scale with the augmentation, the quality signal degrades. The corrections required to keep the system compounding cannot be applied at the rate the system produces work. Compounding stops. The same accelerator that produced the volume undermines the conditions for the volume to be evaluated.
The resolution: evaluation-bottleneck-aware design. If the operator's evaluation capacity is the slow input, the system has to be designed with explicit evaluation chokepoints: the dipole, the steelman, the reader-as-dipole, the calibration loop. Volume is throttled at the rate the operator can evaluate. The accelerator is run at the speed of the slow input, not at maximum.
Same disease, two presentations.
The two domains differ in everything except the structural shape. The same observation holds: when a force outruns the conditions required to evaluate or finance its outputs, you cannot solve the problem at the level of the force. You solve it at the level of the slow input, by partitioning, throttling, or augmenting the input to keep it from being the bottleneck.
The pattern is falsifiable. It dissolves in any system where the disruptive force scales the slow input proportionally. In capital markets, a financing instrument that re-establishes credible long horizons under arbitrary moat-volatility would do it. In knowledge work, a model that can internalize and apply the operator's correction history autonomously and faithfully would do it. Both are speculative, not present. The pattern's domain is "where capability and the slow inputs run on different time-scales." Where they do not, the pattern does not apply.
Chamath wrote the autopsy from inside the body. Chamath is long the disease.